NoteBooks/GreenStone_computational_time.ipynb

{"nbformat":4,"nbformat_minor":0,"metadata":{"colab":{"provenance":[],"collapsed_sections":[],"machine_shape":"hm","mount_file_id":"1vlcey_pB98x7AKpY3hPdajviL_iZ4f8H","authorship_tag":"ABX9TyMT4jy7pLcsQ02ej9GOSw9z"},"kernelspec":{"name":"python3","display_name":"Python 3"},"language_info":{"name":"python"},"accelerator":"GPU"},"cells":[{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/","height":1000},"id":"oofdUGLPaenp","executionInfo":{"status":"ok","timestamp":1638311298663,"user_tz":-60,"elapsed":39079,"user":{"displayName":"george liang","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13911471454046444619"}},"outputId":"893e27ca-bf03-40f0-d393-d9ae3d94a41e"},"source":["!pip install nptyping==1.0.1\n","!pip install SciencePlots\n","!pip install scikit-image==0.18.2rc1\n","!pip install matplotlib==3.4.2"],"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["Collecting nptyping==1.0.1\n","  Downloading nptyping-1.0.1-py3-none-any.whl (7.2 kB)\n","Collecting typish>=1.5.2\n","  Downloading typish-1.9.3-py3-none-any.whl (45 kB)\n","\u001b[K     |████████████████████████████████| 45 kB 1.5 MB/s \n","\u001b[?25hRequirement already satisfied: numpy in /usr/local/lib/python3.7/dist-packages (from nptyping==1.0.1) (1.19.5)\n","Installing collected packages: typish, nptyping\n","Successfully installed nptyping-1.0.1 typish-1.9.3\n","Collecting SciencePlots\n","  Downloading SciencePlots-1.0.9.tar.gz (10 kB)\n","  Installing build dependencies ... \u001b[?25l\u001b[?25hdone\n","\u001b[33m  WARNING: Missing build requirements in pyproject.toml for SciencePlots from https://files.pythonhosted.org/packages/c2/44/7b5c0ecd6f2862671a076425546f86ac540bc48c1a618a82d6faa3b26f58/SciencePlots-1.0.9.tar.gz#sha256=2b002f263734a718acdf1e5be57f0de5ee887e70517dfbe1118a27d0ee3dfb0f.\u001b[0m\n","\u001b[33m  WARNING: The project does not specify a build backend, and pip cannot fall back to setuptools without 'wheel'.\u001b[0m\n","  Getting requirements to build wheel ... \u001b[?25l\u001b[?25hdone\n","  Installing backend dependencies ... \u001b[?25l\u001b[?25hdone\n","    Preparing wheel metadata ... \u001b[?25l\u001b[?25hdone\n","Requirement already satisfied: matplotlib in /usr/local/lib/python3.7/dist-packages (from SciencePlots) (3.2.2)\n","Requirement already satisfied: numpy>=1.11 in /usr/local/lib/python3.7/dist-packages (from matplotlib->SciencePlots) (1.19.5)\n","Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib->SciencePlots) (3.0.6)\n","Requirement already satisfied: python-dateutil>=2.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib->SciencePlots) (2.8.2)\n","Requirement already satisfied: cycler>=0.10 in /usr/local/lib/python3.7/dist-packages (from matplotlib->SciencePlots) (0.11.0)\n","Requirement already satisfied: kiwisolver>=1.0.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib->SciencePlots) (1.3.2)\n","Requirement already satisfied: six>=1.5 in /usr/local/lib/python3.7/dist-packages (from python-dateutil>=2.1->matplotlib->SciencePlots) (1.15.0)\n","Building wheels for collected packages: SciencePlots\n","  Building wheel for SciencePlots (PEP 517) ... \u001b[?25l\u001b[?25hdone\n","  Created wheel for SciencePlots: filename=SciencePlots-1.0.9-py3-none-any.whl size=6483 sha256=931de6b45cae6855de9614887526d6595dce336b8119aa87833d746e392e7d72\n","  Stored in directory: /root/.cache/pip/wheels/37/f8/e9/b2f53a40b336388dfc57b108150daff7d6ffbbfc618dba3924\n","Successfully built SciencePlots\n","Installing collected packages: SciencePlots\n","Successfully installed SciencePlots-1.0.9\n","Collecting scikit-image==0.18.2rc1\n","  Downloading scikit_image-0.18.2rc1-cp37-cp37m-manylinux1_x86_64.whl (29.2 MB)\n","\u001b[K     |████████████████████████████████| 29.2 MB 85.5 MB/s \n","\u001b[?25hRequirement already satisfied: tifffile>=2019.7.26 in /usr/local/lib/python3.7/dist-packages (from scikit-image==0.18.2rc1) (2021.11.2)\n","Requirement already satisfied: scipy>=1.0.1 in /usr/local/lib/python3.7/dist-packages (from scikit-image==0.18.2rc1) (1.4.1)\n","Requirement already satisfied: matplotlib!=3.0.0,>=2.0.0 in /usr/local/lib/python3.7/dist-packages (from scikit-image==0.18.2rc1) (3.2.2)\n","Requirement already satisfied: networkx>=2.0 in /usr/local/lib/python3.7/dist-packages (from scikit-image==0.18.2rc1) (2.6.3)\n","Requirement already satisfied: numpy>=1.16.5 in /usr/local/lib/python3.7/dist-packages (from scikit-image==0.18.2rc1) (1.19.5)\n","Requirement already satisfied: imageio>=2.3.0 in /usr/local/lib/python3.7/dist-packages (from scikit-image==0.18.2rc1) (2.4.1)\n","Requirement already satisfied: PyWavelets>=1.1.1 in /usr/local/lib/python3.7/dist-packages (from scikit-image==0.18.2rc1) (1.2.0)\n","Requirement already satisfied: pillow!=7.1.0,!=7.1.1,>=4.3.0 in /usr/local/lib/python3.7/dist-packages (from scikit-image==0.18.2rc1) (7.1.2)\n","Requirement already satisfied: cycler>=0.10 in /usr/local/lib/python3.7/dist-packages (from matplotlib!=3.0.0,>=2.0.0->scikit-image==0.18.2rc1) (0.11.0)\n","Requirement already satisfied: kiwisolver>=1.0.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib!=3.0.0,>=2.0.0->scikit-image==0.18.2rc1) (1.3.2)\n","Requirement already satisfied: python-dateutil>=2.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib!=3.0.0,>=2.0.0->scikit-image==0.18.2rc1) (2.8.2)\n","Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib!=3.0.0,>=2.0.0->scikit-image==0.18.2rc1) (3.0.6)\n","Requirement already satisfied: six>=1.5 in /usr/local/lib/python3.7/dist-packages (from python-dateutil>=2.1->matplotlib!=3.0.0,>=2.0.0->scikit-image==0.18.2rc1) (1.15.0)\n","Installing collected packages: scikit-image\n","  Attempting uninstall: scikit-image\n","    Found existing installation: scikit-image 0.18.3\n","    Uninstalling scikit-image-0.18.3:\n","      Successfully uninstalled scikit-image-0.18.3\n","\u001b[31mERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.\n","albumentations 0.1.12 requires imgaug<0.2.7,>=0.2.5, but you have imgaug 0.2.9 which is incompatible.\u001b[0m\n","Successfully installed scikit-image-0.18.2rc1\n","Collecting matplotlib==3.4.2\n","  Downloading matplotlib-3.4.2-cp37-cp37m-manylinux1_x86_64.whl (10.3 MB)\n","\u001b[K     |████████████████████████████████| 10.3 MB 3.4 MB/s \n","\u001b[?25hRequirement already satisfied: pyparsing>=2.2.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib==3.4.2) (3.0.6)\n","Requirement already satisfied: python-dateutil>=2.7 in /usr/local/lib/python3.7/dist-packages (from matplotlib==3.4.2) (2.8.2)\n","Requirement already satisfied: cycler>=0.10 in /usr/local/lib/python3.7/dist-packages (from matplotlib==3.4.2) (0.11.0)\n","Requirement already satisfied: pillow>=6.2.0 in /usr/local/lib/python3.7/dist-packages (from matplotlib==3.4.2) (7.1.2)\n","Requirement already satisfied: numpy>=1.16 in /usr/local/lib/python3.7/dist-packages (from matplotlib==3.4.2) (1.19.5)\n","Requirement already satisfied: kiwisolver>=1.0.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib==3.4.2) (1.3.2)\n","Requirement already satisfied: six>=1.5 in /usr/local/lib/python3.7/dist-packages (from python-dateutil>=2.7->matplotlib==3.4.2) (1.15.0)\n","Installing collected packages: matplotlib\n","  Attempting uninstall: matplotlib\n","    Found existing installation: matplotlib 3.2.2\n","    Uninstalling matplotlib-3.2.2:\n","      Successfully uninstalled matplotlib-3.2.2\n","\u001b[31mERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.\n","albumentations 0.1.12 requires imgaug<0.2.7,>=0.2.5, but you have imgaug 0.2.9 which is incompatible.\u001b[0m\n","Successfully installed matplotlib-3.4.2\n"]},{"output_type":"display_data","data":{"application/vnd.colab-display-data+json":{"pip_warning":{"packages":["matplotlib","mpl_toolkits"]}}},"metadata":{}}]},{"cell_type":"code","metadata":{"id":"yLnAoDrtaf0a"},"source":["import tensorflow as tf\n","\n","# gpus = tf.config.experimental.list_physical_devices('GPU') \n","# for gpu in gpus: tf.config.experimental.set_memory_growth(gpu, True)\n","from sys import getsizeof\n","\n","from ipywidgets import interact, interactive\n","import os\n","import numpy as np\n","import sys\n","import timeit\n","import csv\n","sys.path.append('/content/drive/MyDrive/YJ/GP_old')\n","sys.path.append('/content/drive/MyDrive/YJ/')\n","sys.path.append('/content/drive/MyDrive/YJ/Green_Stone')\n","sys.path.append('/content/drive/MyDrive/YJ/Gravity_valid')\n","import matplotlib.pyplot as plt\n","import time\n","\n","import gempy as gp\n","# from gempy.core.tensor.tensorflow_graph_uncon0 import TFGraph\n","from gempy.core.grid_modules.grid_types import CenteredGrid,CenteredRegGrid,RegularGrid\n","import tensorflow as tf\n","import tensorflow_probability as tfp\n","import pandas as pd\n","from gempy import create_data, map_series_to_surfaces\n","from gempy.assets.geophysics import GravityPreprocessing\n","from gempy.plot.visualization_2d_pro import *\n","tfd = tfp.distributions\n","import seaborn as sns\n","from GreenStone2 import *\n","from gravity_util import *\n","# from ML_grav_kernel import *\n","tfconstant = constant32\n","\n","# # %%\n","# grav_res = 10\n","# X_r = np.linspace(704000,740000,grav_res)\n","# Y_r = np.linspace(6.87e6,6.92e6,grav_res)\n","\n","# r = []\n","# for x in X_r:\n","#   for y in Y_r:\n","#     r.append(np.array([x,y]))\n","# receivers = np.array(r)\n","# Z_r = 200\n","# n_devices = receivers.shape[0]\n","# xyz = np.meshgrid(X_r, Y_r, Z_r)\n","# xy_ravel = np.vstack(list(map(np.ravel, xyz))).T\n","\n","# model1 = Model(receivers = receivers,dtype = 'float32')"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"ivC9lHSogcts"},"source":["\n","class Optimizing():\n","  def __init__(self,center_grid_resolution,radius):\n","\n","    self.radius = tfconstant(radius)\n","    min_ = -1\n","    max_ = 1\n","    self._a = make_variables(tf.cast(center_grid_resolution[0]/2+1,tf.int32),initializer= tf.random_uniform_initializer(minval=min_, maxval=max_, seed=None),dtype = tf.float32)\n","    self._b = make_variables(tf.cast(center_grid_resolution[1]/2+1,tf.int32),initializer= tf.random_uniform_initializer(minval=min_, maxval=max_, seed=None),dtype = tf.float32)\n","    self._c = make_variables(tf.cast(center_grid_resolution[2],tf.int32),initializer= tf.random_uniform_initializer(minval=min_, maxval=max_, seed=None),dtype = tf.float32)\n","\n","  def loss(self,a,b,c):\n","    tz = compute_tz(a,b,c,self.radius)\n","    l = tf.math.reduce_std(tz)\n","    return l\n","\n","  @tf.function\n","  def train_step(self,opt):\n","    with tf.GradientTape() as tape:\n","      loss_value = self.loss(self._a,self._b,self._c)\n","      grads = tape.gradient(loss_value,[self._a,self._b,self._c])\n","\n","    opt.apply_gradients(zip(grads, [self._a,self._b,self._c]))\n","    return loss_value\n","\n","  def __call__(self,epochs = 1000):\n","\n","    opt = tf.keras.optimizers.Adam(\n","      learning_rate=0.01, beta_1=0.9, beta_2=0.999, epsilon=1e-07, amsgrad=False,\n","      name='Adam')\n","\n","    for epoch in range(epochs):\n","      # start_time = time.time()\n","      loss_value = self.train_step(opt)\n","      if epoch%200==0:\n","        print(\"\\nStart of epoch %d\" % (epoch,))\n","        print(\"loss: %.10f\" % (float(loss_value),))\n","        # print(\"Time taken: %.2fs\" % (time.time() - start_time))\n","    \n","    return self._a,self._b,self._c\n","\n","\n","class OptimalGrid(CenteredGrid):\n","    \"\"\"\n","    ML optimized spaced grid.\n","    \"\"\"\n","\n","    def __init__(self, centers=None, radius=None, resolution=None, abc = None):\n","        super().__init__(centers=None, radius=None, resolution=None)\n","        self.grid_type = 'centered_grid'\n","        self.values = np.empty((0, 3))\n","        self.length = self.values.shape[0]\n","        self.kernel_centers = np.empty((0, 3))\n","        self.kernel_dxyz_left = np.empty((0, 3))\n","        self.kernel_dxyz_right = np.empty((0, 3))\n","        self.tz = np.empty(0)\n","\n","        if centers is not None and radius  is not None and abc is None:\n","            # if resolution is None:\n","            #     resolution = [10, 10, 20]\n","            \n","            f = Optimizing(resolution,radius)\n","            a,b,c = f(epochs=1000)\n","            self.abc = [a,b,c]\n","\n","            self.set_centered_grid( centers = centers,radius=radius, resolution=resolution,a=a,b=b,c=c)\n","        \n","        # with a given optimized grid\n","        elif centers is not None and radius  is not None and abc is not None: \n","            a, b, c = [abc[i] for i in range(3)]\n","            self.set_centered_grid( centers = centers,radius=radius, resolution=resolution,a=a,b=b,c=c)\n","\n","###################\n","\n","\n","\n","    @staticmethod\n","    def create_irregular_grid_kernel(radius,a=None,b=None,c=None):\n","        \"\"\"\n","        Create an isometric grid kernel (centered at 0)\n","\n","        Args:\n","            resolution: [s0]\n","            radius (float): Maximum distance of the kernel\n","\n","        Returns:\n","            tuple: center of the voxel, left edge of each voxel (for xyz), right edge of each voxel (for xyz).\n","        \"\"\"\n","        import copy\n","        if not isinstance(radius, list) or isinstance(radius, np.ndarray):\n","            radius = np.repeat(radius, 3)\n","\n","        # a,b,c = find_optimal_grid(a,b,c,epochs)\n","   \n","        g_x = tf.cumsum(softmax_space(a))\n","        g_y = tf.cumsum(softmax_space(b))\n","        g_z = tf.cumsum(softmax_space(c))\n","\n","        g2_x = tf.concat((-g_x[::-1], g_x),axis=0) * radius[0]\n","        g2_y = tf.concat((-g_y[::-1], g_y),axis=0) * radius[1]\n","        g2_z =  (tf.concat(([0], g_z),axis=0) + 0.005) * - radius[2]\n","\n","        x_center = (g2_x[:-1]+g2_x[1:])/2 \n","        y_center = (g2_y[:-1]+g2_y[1:])/2\n","        z_center = (g2_z[:-1]+g2_z[1:])/2\n","\n","        g = tf.meshgrid(x_center,y_center,z_center)\n","\n","        d_left_x = tf.math.abs(g2_x[:-1] - x_center)\n","        d_left_y = tf.math.abs(g2_y[:-1] - y_center)\n","        d_right_x = tf.math.abs(g2_x[1:] - x_center)\n","        d_right_y = tf.math.abs(g2_y[1:] - y_center)\n","        d_z = z_center - g2_z[:-1]\n","\n","        d_left = tf.meshgrid(d_left_x,d_left_y,d_z)\n","        d_right = tf.meshgrid(d_right_x,d_right_y,d_z)\n","\n","        kernel_g = tf.concat([tf.reshape(g[0],[-1,1]),tf.reshape(g[1],[-1,1]),tf.reshape(g[2],[-1,1])],axis=1)\n","        kernel_d_left = tf.concat([tf.reshape(d_left[0],[-1,1]),tf.reshape(d_left[1],[-1,1]),tf.reshape(d_left[2],[-1,1])],axis=1)\n","        kernel_d_right = tf.concat([tf.reshape(d_right[0],[-1,1]),tf.reshape(d_right[1],[-1,1]),tf.reshape(d_right[2],[-1,1])],axis=1)\n","\n","        return kernel_g.numpy(), kernel_d_left.numpy(), kernel_d_right.numpy()\n","  \n","\n","    def set_centered_kernel(self, resolution, radius,a=None,b=None,c=None):\n","        \"\"\"\n","        Set a centered\n","\n","        Args:\n","            resolution: [s0]\n","            radius (float): Maximum distance of the kernel\n","\n","        Returns:\n","\n","        \"\"\"\n","        self.kernel_centers, self.kernel_dxyz_left, self.kernel_dxyz_right = self.create_irregular_grid_kernel(\n","             radius,a=a,b=b,c=c)\n","\n","        return self.kernel_centers\n","\n","\n","# centered optimized grid\n","def compute_tz(a,b,c,radius):\n","  g_x = tf.cumsum(softmax_space(a))\n","  g_y = tf.cumsum(softmax_space(b))\n","  g_z = tf.cumsum(softmax_space(c))\n","\n","  g2_x = tf.concat((-g_x[::-1], g_x),axis=0) * radius[0]\n","  g2_y = tf.concat((-g_y[::-1], g_y),axis=0) * radius[1]\n","  g2_z =  (tf.concat(([0], g_z),axis=0) + 0.005) * - radius[2]\n","\n","  x_center = (g2_x[:-1]+g2_x[1:])/2 \n","  y_center = (g2_y[:-1]+g2_y[1:])/2\n","  z_center = (g2_z[:-1]+g2_z[1:])/2\n","\n","  g = tf.meshgrid(x_center,y_center,z_center)\n","\n","  d_left_x = tf.math.abs(g2_x[:-1] - x_center)\n","  d_left_y = tf.math.abs(g2_y[:-1] - y_center)\n","  d_right_x = tf.math.abs(g2_x[1:] - x_center)\n","  d_right_y = tf.math.abs(g2_y[1:] - y_center)\n","  d_z = z_center - g2_z[:-1]\n","\n","  d_left = tf.meshgrid(d_left_x,d_left_y,d_z)\n","  d_right = tf.meshgrid(d_right_x,d_right_y,d_z)\n","\n","  kernel_g = tf.concat([tf.reshape(g[0],[-1,1]),tf.reshape(g[1],[-1,1]),tf.reshape(g[2],[-1,1])],axis=1)\n","  kernel_d_left = tf.concat([tf.reshape(d_left[0],[-1,1]),tf.reshape(d_left[1],[-1,1]),tf.reshape(d_left[2],[-1,1])],axis=1)\n","  kernel_d_right = tf.concat([tf.reshape(d_right[0],[-1,1]),tf.reshape(d_right[1],[-1,1]),tf.reshape(d_right[2],[-1,1])],axis=1)\n","\n","  s_gr_x = kernel_g[:, 0]\n","  s_gr_y = kernel_g[:, 1]\n","  s_gr_z = kernel_g[:, 2]\n","\n","  # # getting the coordinates of the corners of the voxel...\n","  x_cor = tf.transpose(tf.stack([s_gr_x - kernel_d_left[:, 0], s_gr_x + kernel_d_right[:, 0]], axis=0))\n","  y_cor = tf.transpose(tf.stack((s_gr_y - kernel_d_left[:, 1], s_gr_y + kernel_d_right[:, 1]), axis=0))\n","  z_cor = tf.transpose(tf.stack((s_gr_z - kernel_d_left[:, 2], s_gr_z + kernel_d_right[:, 2]), axis=0))\n","\n","  x_matrix = tf.repeat(x_cor, 4, axis=1)\n","  y_matrix = tf.tile(tf.repeat(y_cor, 2, axis=1), (1, 2))\n","  z_matrix = tf.tile(z_cor, (1, 4))\n","\n","  s_r = tf.sqrt(x_matrix ** 2 + y_matrix ** 2 + z_matrix ** 2)\n","\n","  mu = tfconstant([1, -1, -1, 1, -1, 1, 1, -1])\n","\n","  G = tfconstant(6.674e-3)\n","\n","  tz = (\n","        G *\n","        tf.reduce_sum(- 1 *\n","                mu * (\n","                        x_matrix * tf.math.log(y_matrix + s_r) +\n","                        y_matrix * tf.math.log(x_matrix + s_r) -\n","                        z_matrix * tf.math.atan(x_matrix * y_matrix / (z_matrix * s_r))),\n","                axis=1))\n","  return tz"],"execution_count":null,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"axNw3IFLdtYF"},"source":["#### convolution spatial method"]},{"cell_type":"code","metadata":{"id":"dy130DoNdmxn"},"source":["class GravityPreprocessingRegAllLoop(RegularGrid):\n","    def __init__(self, model, regular_grid: RegularGrid = None):\n","\n","        if regular_grid is None:\n","            super().__init__()\n","        elif isinstance(regular_grid, RegularGrid):\n","            self.model = model\n","            # self.kernel_centers = np.repeat(regular_grid.values[:,:,np.newaxis],2,axis=2) - model.xy_ravel.T\n","            self.num_receivers = 1\n","            # self.kernel_dxyz_right = regular_grid.kernel_dxyz_right\n","            # self.kernel_dxyz_left = regular_grid.kernel_dxyz_left\n","        self.tz = np.empty(0)\n","\n","    def set_tz_kernel(self, model_radius,regular_grid_resolution,scale=True, **kwargs):\n","        dx, dy, dz = self.model.grid.regular_grid.get_dx_dy_dz()\n","\n","        # we need to find the closest center for each receiver to keep numerical stability\n","        # here we find the smallest center which is greater than the receiver coordinates for x and y\n","        range_xy = [(self.model.extent[1] - self.model.extent[0])/2,(self.model.extent[3] - self.model.extent[2])/2] \n","        #center of model\n","        new_xy_ravel = np.expand_dims(np.array([self.model.extent[0]+range_xy[0],self.model.extent[2]+range_xy[1]]),axis = 1)\n","\n","        ################\n","        self.dx = dx\n","        self.dy = dy\n","        self.dz = dz\n","        self.re_x = self.model.xy_ravel.T[0]\n","        self.re_x = self.re_x + (dx / 2 - (self.re_x-self.model.extent[0]) % dx)\n","\n","        self.re_y = self.model.xy_ravel.T[1]\n","        self.re_y = self.re_y + (dy / 2 - (self.re_y-self.model.extent[2]) % dy)\n","\n","        self.new_xy_ravel = np.stack(\n","            [\n","                self.re_x,\n","                self.re_y,\n","            ],\n","            axis=0,\n","        )\n","        # concat with z value\n","        self.new_xy_ravel = np.concatenate(\n","            [self.new_xy_ravel, self.model.xy_ravel.T[2, None]]\n","        )\n","\n","        ################\n","        # # sensor location move to the next closest cell center\n","        re_x = new_xy_ravel[0]\n","        re_x = re_x + (dx / 2 - (re_x-self.model.extent[0]) % dx)\n","\n","        re_y = new_xy_ravel[1]\n","        re_y = re_y + (dy / 2 - (re_y-self.model.extent[2]) % dy)\n","\n","        # concat with z value\n","        new_xy_ravel_temp = np.concatenate(\n","            [np.array([re_x,re_y]), np.array([[self.model.extent[-1]]])])\n","\n","        self.center_index_x = (new_xy_ravel_temp[0]-self.model.extent[0])//dx\n","        self.center_index_y = (new_xy_ravel_temp[1]-self.model.extent[2])//dy\n","        self.radius_cell_x = int(self.model.model_radius[0]//dx)\n","        self.radius_cell_y = int(self.model.model_radius[1]//dy)\n","\n","        # kernel_centers = np.repeat(self.model.grid.regular_grid.values[:,:,np.newaxis],self.num_receivers,axis=2)-self.model.xy_ravel.T\n","        kernel_centers = np.squeeze(\n","                self.model.grid.regular_grid.values[:, :, np.newaxis] - new_xy_ravel_temp\n","        )\n","\n","        c_x = int(self.center_index_x[0])\n","        c_y = int(self.center_index_y[0])\n","\n","        slice_kernel_centers = (kernel_centers[:,:].reshape(model.regular_grid_resolution+[3,self.num_receivers])[c_x-self.radius_cell_x:c_x+self.radius_cell_x+1,c_y-self.radius_cell_y:c_y+self.radius_cell_y+1,:,:]).reshape([-1,3])\n","\n","\n","        x_cor = np.stack(\n","            (slice_kernel_centers[:, 0] - dx / 2, slice_kernel_centers[:, 0] + dx / 2), axis=1\n","        )\n","        y_cor = np.stack(\n","            (slice_kernel_centers[:, 1] - dy / 2, slice_kernel_centers[:, 1] + dy / 2), axis=1\n","        )\n","        z_cor = np.stack(\n","            (slice_kernel_centers[:, 2] + dz / 2, slice_kernel_centers[:, 2] - dz / 2), axis=1\n","        )\n","\n","        # ...and prepare them for a vectorial op\n","        x_matrix = np.repeat(x_cor, 4, axis=1)\n","        y_matrix = np.tile(np.repeat(y_cor, 2, axis=1), (1, 2))\n","        z_matrix = np.tile(z_cor, (1, 4)) - 0.005 * (self.model.extent[-1]-self.model.extent[-2])\n","\n","        s_r = np.sqrt(x_matrix ** 2 + y_matrix ** 2 + z_matrix ** 2)\n","\n","        # This is the vector that determines the sign of the corner of the voxel\n","        mu = np.array([1, -1, -1, 1, -1, 1, 1, -1])\n","\n","\n","        G = 6.674e-3  # ugal     cm3⋅g−1⋅s−26.67408e-2 -- 1 m/s^2 to microgal =\n","\n","        tz = G * np.sum(\n","            -1\n","            * mu\n","            * (\n","                x_matrix * np.log(y_matrix + s_r)\n","                + y_matrix * np.log(x_matrix + s_r)\n","                - z_matrix * np.arctan(x_matrix * y_matrix / (z_matrix * s_r))\n","            ),\n","            axis=1,\n","        )\n","        return tz"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"BmDHTh2DlSTn"},"source":["# model1 = Model(receivers = receivers,dtype = 'float32',center_grid_resolution=[9,7,50],model_radius=[5000,5000,20200])"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"xuRglsWhI1UH"},"source":["# model1.activate_regular_grid()\n","# model1.compute_model()"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"J1X2g_7iI1O9"},"source":["# p = Plot2D(model1)\n","# fig, ax = plt.subplots(1,1,figsize = (10,20))\n","# ax.set_ylim(-20000,0)\n","# ax.set_xlim(6863000,6930000)\n","# p.plot_lith(ax,cell_number=[30], direction='x')"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"u0F1C0RdlSC9"},"source":["# p = Plot2D(model1)\n","# # pl = gp.plot.plot_data(model1.geo_data, direction=\"z\") # gempy.plot.plot.plot_data\n","# ax = plt.gca()\n","# ax.set_ylim(model1.extent[2],model1.extent[3])\n","# ax.set_xlim(model1.extent[0],model1.extent[1])\n","# p.plot_contacts(ax,cell_number=[-1], direction='z')\n","# ax.scatter(receivers[:,0],receivers[:,1],s=10)\n","# p.plot_lith(ax,cell_number=[-1], direction='z')\n","# ax.set_xlabel('X')\n","# ax.set_ylabel('Y')\n","# for i in p._color_lot:\n","#   ax.scatter([],[],color=p._color_lot[i],label = i)\n","# ax.legend(loc='center left', bbox_to_anchor=(1, 0.5))\n","# ax.yaxis.set_tick_params(rotation=90)\n","# # plt.savefig('/content/drive/MyDrive/YJ/Green_Stone/figs/lith.png',dpi=500,bbox_inches='tight')"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"jSrnXnFBTQBr"},"source":["\n","# p = Plot2D(model1)\n","# fig, axes = plt.subplots(3,1,figsize = (10,10))\n","# for e,ax in enumerate(axes):\n","#   ax.set_ylim(-20000,0)\n","#   ax.set_xlim(6863000,6930000)\n","#   p.plot_lith(ax,cell_number=[30], direction='x')\n","#   p.plot_scalar_field(ax,sn=e,cell_number=model1.regular_grid_resolution[1]//2)"],"execution_count":null,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"HgoImlvEY8sD"},"source":["# Gravity"]},{"cell_type":"code","metadata":{"id":"fG6uzxD2ZnW7"},"source":["# grav_res = 10\n","# X_r = np.linspace(704000,740000,grav_res)\n","# Y_r = np.linspace(6.87e6,6.92e6,grav_res)\n","\n","# r = []\n","# for x in X_r:\n","#   for y in Y_r:\n","#     r.append(np.array([x,y]))\n","# receivers = np.array(r)\n","# Z_r = 200\n","# n_devices = receivers.shape[0]\n","# xyz = np.meshgrid(X_r, Y_r, Z_r)\n","# xy_ravel = np.vstack(list(map(np.ravel, xyz))).T"],"execution_count":null,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"cYZ9AAT4ZIQN"},"source":["## convoluted spatial"]},{"cell_type":"code","metadata":{"id":"ol3BEEkoPYdb"},"source":["\n","def calculate_grav_convoluted(model):\n","  model.activate_regular_grid()\n","  g = GravityPreprocessingRegAllLoop(model,model.grid.regular_grid)\n","  tz = g.set_tz_kernel(model.model_radius,model.regular_grid_resolution)\n","  model.geo_data.grid.values = model.geo_data.grid.values - np.array([0,0,0.005 * (g.model.extent[-1]-g.model.extent[-2])])\n","  model.geo_data.update_from_grid()\n","  # model.geo_data.rescaling.set_rescaled_grid()\n","  model.grid = model.geo_data.grid\n","  model.from_gempy_interpolator()\n","  Input = model.get_graph_input()\n","  model.create_tensorflow_graph(Input)\n","  start = time.time()\n","  formation_block,property_block,block_mask = model.TFG.compute_series(model.surface_points_coord,\n","              model.dips_position,\n","              model.dip_angles,\n","              model.azimuth,\n","              model.polarity,\n","              model.values_properties)\n","\n","  size = tf.reduce_prod(model.regular_grid_resolution)\n","\n","  densities = property_block[0:size]\n","\n","  center_index_x = (g.new_xy_ravel[0]-g.model.extent[0])//g.dx\n","  center_index_y = (g.new_xy_ravel[1]-g.model.extent[2])//g.dy\n","  grav_convolution_full = tf.TensorArray(tf.float32, size=n_devices, dynamic_size=False, clear_after_read=True)\n","  for i in range(n_devices):\n","    c_x = int(center_index_x[i])\n","    c_y = int(center_index_y[i])\n","\n","    windowed_densities = tf.reshape(densities,model.regular_grid_resolution)[c_x-g.radius_cell_x:c_x+g.radius_cell_x+1,c_y-g.radius_cell_y:c_y+g.radius_cell_y+1,:]\n","    windowed_densities = tf.squeeze(tf.reshape(windowed_densities,(-1,1)))\n","    grav_ = model.TFG.compute_forward_gravity(tfconstant(tz), 0, size, windowed_densities)\n","    grav_convolution_full = grav_convolution_full.write(i, grav_)\n","  grav_convolution_full = tf.squeeze(grav_convolution_full.stack())\n","  end = time.time()\n","  print('time covoluted:',end - start)\n","  print('equivalent center resolutoin: ', g.radius_cell_x*2+1, g.radius_cell_y*2+1)\n","  return grav_convolution_full, end - start"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"lO9HBbA4y4XA"},"source":["def plot_grav(grav_diff,diff = True, ax = None, subplots =False):\n","  p = Plot2D(model1)\n","  if ax is None:\n","    ax = plt.gca()\n","  f = plt.gcf()\n","  ax.set_ylim(model1.extent[2],model1.extent[3])\n","  ax.set_xlim(model1.extent[0],model1.extent[1])\n","  ax.set_ylabel('Y')\n","  ax.set_xlabel('X')\n","  if diff is True:\n","    vmin = -3\n","    vmax = 3\n","  else: \n","    vmin=None\n","    vmax=None\n","    ax.scatter(receivers[:,0],receivers[:,1],s=10)\n","\n","  ax.yaxis.set_tick_params(rotation=90)\n","  p.plot_contacts(ax,cell_number=[-1], direction='z')\n","  im = ax.imshow(grav_diff.reshape(grav_res,grav_res).T,\n","            vmin = vmin, vmax = vmax,\n","            extent=(xy_ravel[:, 0].min() + (xy_ravel[0, 0] - xy_ravel[1, 0]) / 2,\n","                   xy_ravel[:, 0].max() - (xy_ravel[0, 0] - xy_ravel[1, 0]) / 2,\n","                   xy_ravel[:, 1].min() + (xy_ravel[0, 1] - xy_ravel[grav_res, 1]) / 2,\n","                   xy_ravel[:, 1].max() - (xy_ravel[0, 1] - xy_ravel[grav_res, 1]) / 2),\n","            origin='lower',cmap='viridis_r')\n","  if subplots is True:\n","    return im\n","  f.subplots_adjust(right=0.8)\n","  cbar = f.colorbar(im)\n","  cbar.set_label(r'$mgal$',rotation=270)\n","  return f, ax"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"gO9KJ9oGiGK_"},"source":["# model = Model(receivers = receivers,dtype = 'float32',center_grid_resolution=[1,1,1],regular_grid_resolution=[400,400,30],model_radius=[5000,5000,20200])"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"xoabSEnux96p"},"source":["# grav_convolution_full,_ =  calculate_grav_convoluted(model)\n","# # relative anomaly\n","# grav_convolution_full = grav_convolution_full.numpy() - grav_convolution_full.numpy().max()"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"C9Eqfa56yWSe"},"source":["# f,ax = plot_grav(grav_convolution_full,diff=False)\n","# ax.set_title(r'Spatial convolution $ 400 \\times 400 \\times 30$')"],"execution_count":null,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"c7PrI0avklIg"},"source":["### low resolution convolution"]},{"cell_type":"code","metadata":{"id":"qVm3Es_s0_pg"},"source":["# regular_grid_resolution_low = [50,50,30]"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"w8qgJ8FDkpi2"},"source":["# model = Model(receivers = receivers,dtype = 'float32',center_grid_resolution=[1,1,1],regular_grid_resolution=regular_grid_resolution_low,model_radius=[5000,5000,20200])\n","# grav_convolution_full_low,_ =  calculate_grav_convoluted(model)\n","# # relative anomaly\n","# grav_convolution_full_low = grav_convolution_full_low.numpy() - grav_convolution_full_low.numpy().max()"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"dPA2IimOzduu"},"source":["# plot_grav(grav_convolution_full_low,diff =False)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"PZKdCZzakyfo"},"source":["# diff_low_full = (grav_convolution_full_low-grav_convolution_full_low.max())-(grav_convolution_full - grav_convolution_full.max())"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"aI69t6UPk7ZJ"},"source":["# plot_grav(diff_low_full)"],"execution_count":null,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"GNrWiPlifpMd"},"source":["## kernel method"]},{"cell_type":"code","metadata":{"id":"S1edbB5yqMdS"},"source":["# centered regular grid\n","# activate the customized grid\n","def activate_customized_grid(model,grid_kernel):\n","  model.geo_data.grid.custom_grid=grid_kernel\n","  model.geo_data.grid.deactivate_all_grids()\n","  # activate also rescaled the grid\n","  model.geo_data.grid.set_active('custom')\n","\n","  model.geo_data.update_from_grid()\n","  model.geo_data.rescaling.set_rescaled_grid()\n","  model.grid = model.geo_data.grid\n","  model.from_gempy_interpolator()\n","  return model\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"ci6ZReFKtiS_"},"source":["\n","def calculate_grav_center_regular(model,grid_kernel ):\n","  model = activate_customized_grid(model,grid_kernel)\n","  g_center_regulargrid = GravityPreprocessing(grid_kernel)\n","  tz_center_regulargrid = tfconstant(g_center_regulargrid.set_tz_kernel())\n","  input = model.get_graph_input()\n","  model.create_tensorflow_graph(input)\n","  start = time.time()\n","  formation_block,property_block,block_mask = model.TFG.compute_series(model.surface_points_coord,\n","              model.dips_position,\n","              model.dip_angles,\n","              model.azimuth,\n","              model.polarity,\n","              model.values_properties)\n","  size = grid_kernel.values.shape[0]\n","  densities = property_block[:size]\n","\n","  gravity_center_reg = model.TFG.compute_forward_gravity(tz_center_regulargrid, 0, size, densities)\n","  end = time.time()\n","  print('time:',end - start)\n","  return gravity_center_reg,end - start"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"wegJzXiGuVSm"},"source":["# centerReg = CenteredRegGrid(model.xy_ravel,radius=[5000,5000,20200],resolution=[9,7,30])\n","# gravity_center_reg_0,_ = calculate_grav_center_regular(model,centerReg)\n","# gravity_center_reg_0 = gravity_center_reg_0.numpy() - gravity_center_reg_0.numpy().max()"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"lLNdImAa0s1t"},"source":["# plot_grav(gravity_center_reg_0,diff =False)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"2UoaTKtqie3u"},"source":["# diff_cr_full = (gravity_center_reg_0)-(grav_convolution_full )"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"hnRTJzfXiUGN"},"source":["# plot_grav(diff_cr_full)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"u74xSlVmUmKK"},"source":["# centerML = OptimalGrid(model.xy_ravel,radius=[5000,5000,20200],resolution=[9,7,30])\n","\n","# gravity_center_ml,_ = calculate_grav_center_regular(model,centerML)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"StWls5k6mLaw"},"source":["# # centered_optimal_grid\n","# diff_co_full = (gravity_center_ml.numpy()-gravity_center_ml.numpy().max())-(grav_convolution_full - grav_convolution_full.max())"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"F5SPLnCEgGnH"},"source":["# plot_grav(diff_co_full)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"oosb4MKHm17p"},"source":["# rmse_low_full = tf.sqrt(tf.reduce_mean(diff_low_full**2)).numpy()"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"nlFblUCTUmE9"},"source":["# rmse_cr = tf.sqrt(tf.reduce_mean(diff_cr_full**2)).numpy()"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"CoR8QAy_UmCs"},"source":["# rmse_co = tf.sqrt(tf.reduce_mean(diff_co_full**2)).numpy()"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"BFO8ZUX25RTC"},"source":["# fig, (ax0, ax1, ax2) = plt.subplots(nrows=1, ncols=3, sharex=True,\n","#                                     figsize=(18, 6))\n","# plot_grav(diff_low_full,ax = ax0, diff=True,subplots=True)\n","# plot_grav(diff_cr_full,ax = ax1, diff=True,subplots=True)\n","# im = plot_grav(diff_co_full,ax = ax2, diff=True,subplots=True)\n","# ax0.set_title(r'Spatial convolution $ 50 \\times 50 $' +'\\n RMSE: %1.3f'% rmse_low_full)\n","# ax1.set_title(r'Kernel regular grid $ 9 \\times 7 $'+'\\n RMSE: %1.3f'% rmse_cr)\n","# ax2.set_title(r'Kernel optimized grid $ 9 \\times 7 $'+'\\n RMSE: %1.3f'% rmse_co)\n","# fig.subplots_adjust(right=0.8)\n","# cbar_ax = fig.add_axes([0.82, 0.15, 0.01, 0.7])\n","# fig.colorbar(im, cax=cbar_ax)\n","# cbar_ax.set_ylabel('$mgal$', rotation=270,labelpad=15)\n"],"execution_count":null,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"w5L7qFXVYBmj"},"source":["# complexity"]},{"cell_type":"code","metadata":{"id":"gsFOMiePYys1","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1638311407746,"user_tz":-60,"elapsed":83183,"user":{"displayName":"george liang","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13911471454046444619"}},"outputId":"3013b822-0186-44ac-ed25-2c271673a60d"},"source":["# %%\n","time_cov_list = []\n","time_kernel_list = []\n","for i in range(1,41):\n","  X_r = np.linspace(704000,740000,10)\n","  Y_r = np.linspace(6.87e6,6.92e6,i)\n","\n","  r = []\n","  for x in X_r:\n","    for y in Y_r:\n","      r.append(np.array([x,y]))\n","  receivers = np.array(r)\n","  Z_r = 200\n","  n_devices = receivers.shape[0]\n","  xyz = np.meshgrid(X_r, Y_r, Z_r)\n","  xy_ravel = np.vstack(list(map(np.ravel, xyz))).T\n","  model = Model(receivers = receivers,dtype = 'float32',center_grid_resolution=[1,1,1],regular_grid_resolution=[50,50,30],model_radius=[5000,5000,20200])\n","  centerReg = CenteredRegGrid(model.xy_ravel,radius=[5000,5000,20200],resolution=[9,7,30])\n","  _,time_cov = calculate_grav_convoluted(model)\n","  _,time_kernel = calculate_grav_center_regular(model,centerReg)\n","  time_cov_list.append(time_cov)\n","  time_kernel_list.append(time_kernel)"],"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 3.6382720470428467\n","equivalent center resolutoin:  9 7\n","time: 0.2074129581451416\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.2560739517211914\n","equivalent center resolutoin:  9 7\n","time: 0.22203302383422852\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.27686429023742676\n","equivalent center resolutoin:  9 7\n","time: 0.21207475662231445\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.2893505096435547\n","equivalent center resolutoin:  9 7\n","time: 0.2127525806427002\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.31884074211120605\n","equivalent center resolutoin:  9 7\n","time: 0.21477770805358887\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.32686710357666016\n","equivalent center resolutoin:  9 7\n","time: 0.2212381362915039\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.33779287338256836\n","equivalent center resolutoin:  9 7\n","time: 0.2162792682647705\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.3640594482421875\n","equivalent center resolutoin:  9 7\n","time: 0.24548697471618652\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.37909364700317383\n","equivalent center resolutoin:  9 7\n","time: 0.22020959854125977\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.40978312492370605\n","equivalent center resolutoin:  9 7\n","time: 0.2182152271270752\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.41097092628479004\n","equivalent center resolutoin:  9 7\n","time: 0.22706055641174316\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.4185173511505127\n","equivalent center resolutoin:  9 7\n","time: 0.22119569778442383\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.4508061408996582\n","equivalent center resolutoin:  9 7\n","time: 0.22482919692993164\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.48320913314819336\n","equivalent center resolutoin:  9 7\n","time: 0.23346662521362305\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.48491501808166504\n","equivalent center resolutoin:  9 7\n","time: 0.23098063468933105\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.5046298503875732\n","equivalent center resolutoin:  9 7\n","time: 0.2405400276184082\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.5094881057739258\n","equivalent center resolutoin:  9 7\n","time: 0.24924921989440918\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.5444278717041016\n","equivalent center resolutoin:  9 7\n","time: 0.24559879302978516\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.552466869354248\n","equivalent center resolutoin:  9 7\n","time: 0.2403552532196045\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.5688681602478027\n","equivalent center resolutoin:  9 7\n","time: 0.24864530563354492\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.593146800994873\n","equivalent center resolutoin:  9 7\n","time: 0.25374627113342285\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.6036834716796875\n","equivalent center resolutoin:  9 7\n","time: 0.2597999572753906\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.6169583797454834\n","equivalent center resolutoin:  9 7\n","time: 0.2620062828063965\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.647244930267334\n","equivalent center resolutoin:  9 7\n","time: 0.2635478973388672\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.6635572910308838\n","equivalent center resolutoin:  9 7\n","time: 0.26819515228271484\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.6836318969726562\n","equivalent center resolutoin:  9 7\n","time: 0.26787471771240234\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.6957252025604248\n","equivalent center resolutoin:  9 7\n","time: 0.26773858070373535\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.7535476684570312\n","equivalent center resolutoin:  9 7\n","time: 0.27586865425109863\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.7226715087890625\n","equivalent center resolutoin:  9 7\n","time: 0.27730417251586914\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.7460241317749023\n","equivalent center resolutoin:  9 7\n","time: 0.2842981815338135\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.762427568435669\n","equivalent center resolutoin:  9 7\n","time: 0.27834129333496094\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.7909431457519531\n","equivalent center resolutoin:  9 7\n","time: 0.28302907943725586\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.8017804622650146\n","equivalent center resolutoin:  9 7\n","time: 0.2872307300567627\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.8034944534301758\n","equivalent center resolutoin:  9 7\n","time: 0.2882201671600342\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.8420922756195068\n","equivalent center resolutoin:  9 7\n","time: 0.294811487197876\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.8437271118164062\n","equivalent center resolutoin:  9 7\n","time: 0.2969319820404053\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.8775730133056641\n","equivalent center resolutoin:  9 7\n","time: 0.31775355339050293\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.8981227874755859\n","equivalent center resolutoin:  9 7\n","time: 0.31044912338256836\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.905550479888916\n","equivalent center resolutoin:  9 7\n","time: 0.2997004985809326\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","time covoluted: 0.9275012016296387\n","equivalent center resolutoin:  9 7\n","time: 0.3099086284637451\n"]}]},{"cell_type":"code","metadata":{"id":"KdcqP4l2bN4y"},"source":["plt.plot(time_cov_list,label='cov')\n","plt.plot(time_kernel_list,label = 'kernel')\n","plt.legend()"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"CJRbiBrWEj53","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1638311409173,"user_tz":-60,"elapsed":976,"user":{"displayName":"george liang","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13911471454046444619"}},"outputId":"211b179c-2d6f-4db6-b4a9-da645c2cc8cb"},"source":["model = Model(receivers = receivers,dtype = 'float32',center_grid_resolution=[1,1,1],regular_grid_resolution=[50,50,30],model_radius=[5000,5000,20200])\n","centerReg = CenteredRegGrid(model.xy_ravel,radius=[5000,5000,20200],resolution=[9,7,30])\n","model = activate_customized_grid(model,centerReg)\n","g_center_regulargrid = GravityPreprocessing(centerReg)\n","tz_center_regulargrid = tfconstant(g_center_regulargrid.set_tz_kernel())\n","input = model.get_graph_input()\n","model.create_tensorflow_graph(input,gradient = True)\n","\n","@tf.function\n","def forward_kernel(mu,model):\n","  TFG = TFGraph(input, model.fault_drift,\n","                    model.grid_tensor, model.values_properties, model.nugget_effect_grad,model.nugget_effect_scalar, model.Range,\n","                    model.C_o, model.rescale_factor,slope = 10, dtype = model.tfdtype, gradient = True)\n","  formation_block,property_block,block_mask = TFG.compute_series(mu,\n","              model.dips_position,\n","              model.dip_angles,\n","              model.azimuth,\n","              model.polarity,\n","              model.values_properties)\n","  size = centerReg.values.shape[0]\n","  densities = property_block[:size]\n","\n","  gravity_center_reg = model.TFG.compute_forward_gravity(tz_center_regulargrid, 0, size, densities)\n","\n","  print('tracing kernel forwad')\n","  tf.print('executing kernel forwad')\n","\n","  return formation_block"],"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]}]},{"cell_type":"code","metadata":{"id":"BPayzBe6FmH5"},"source":["# start_time = time.time()\n","# forward_kernel(model.surface_points_coord,model)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))\n","# start_time = time.time()\n","# forward_kernel(model.surface_points_coord,model)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))\n","# start_time = time.time()\n","# forward_kernel(model.surface_points_coord,model)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"jnZTxVWntDE0"},"source":["# forward_kernel(model.surface_points_coord)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"ZQySIOfxtHb2"},"source":["# forward_cov(model.surface_points_coord)"],"execution_count":null,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"xi1K68IRFsdd"},"source":["convolution forward"]},{"cell_type":"code","metadata":{"id":"vwG_TFMVFsOO"},"source":["model.activate_regular_grid()\n","g = GravityPreprocessingRegAllLoop(model,model.grid.regular_grid)\n","tz = g.set_tz_kernel(model.model_radius,model.regular_grid_resolution)\n","input = model.get_graph_input()\n","model.create_tensorflow_graph(input,gradient = True)\n","@tf.function\n","def forward_cov(mu,model):\n","    TFG = TFGraph(input, model.fault_drift,\n","                    model.grid_tensor, model.values_properties, model.nugget_effect_grad,model.nugget_effect_scalar, model.Range,\n","                    model.C_o, model.rescale_factor,slope = 1, dtype = model.tfdtype, gradient = True)\n","    \n","    formation_block,property_block,block_mask =TFG.compute_series(mu,\n","                model.dips_position,\n","                model.dip_angles,\n","                model.azimuth,\n","                model.polarity,\n","                model.values_properties)\n","\n","    size = tf.reduce_prod(model.regular_grid_resolution)\n","\n","    densities = property_block[0:size]\n","\n","    center_index_x = tf.constant((g.new_xy_ravel[0]-g.model.extent[0])//g.dx,tf.float32)\n","    center_index_y = tf.constant((g.new_xy_ravel[1]-g.model.extent[2])//g.dy,tf.float32)\n","    grav_convolution_full = tf.TensorArray(tf.float32, size=n_devices, dynamic_size=False, clear_after_read=True)\n","    for i in tf.range(n_devices):\n","      c_x = tf.cast(center_index_x[i],tf.int32)\n","      c_y = tf.cast(center_index_y[i],tf.int32)\n","\n","      windowed_densities = tf.reshape(densities,model.regular_grid_resolution)[c_x-g.radius_cell_x:c_x+g.radius_cell_x+1,c_y-g.radius_cell_y:c_y+g.radius_cell_y+1,:]\n","      windowed_densities = tf.squeeze(tf.reshape(windowed_densities,(-1,1)))\n","      grav_ = TFG.compute_forward_gravity(tfconstant(tz), 0, size, windowed_densities)\n","      grav_convolution_full = grav_convolution_full.write(i, grav_)\n","    grav_convolution_full = tf.squeeze(grav_convolution_full.stack())\n","    print('tracing convolution forward')\n","    tf.print('executing convolution forward')\n","    return formation_block"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"eSNfKbAsGLyC"},"source":["# start_time = time.time()\n","# forward_cov(model.surface_points_coord,model)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))\n","# start_time = time.time()\n","# forward_cov(model.surface_points_coord,model)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))\n","# start_time = time.time()\n","# forward_cov(model.surface_points_coord,model)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"Zs0CCmiAGQte","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1638311609605,"user_tz":-60,"elapsed":200437,"user":{"displayName":"george liang","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13911471454046444619"}},"outputId":"c5ea0f77-3c5c-45b6-c87d-4ea603da4c57"},"source":["# %%\n","time_cov_list = []\n","time_kernel_list = []\n","for i in range(1,100,5):\n","# for i in range(1,2):\n","\n","  X_r = np.linspace(704000,740000,2)\n","  Y_r = np.linspace(6.87e6,6.92e6,i)\n","\n","  r = []\n","  for x in X_r:\n","    for y in Y_r:\n","      r.append(np.array([x,y]))\n","  receivers = np.array(r)\n","  Z_r = 200\n","  n_devices = receivers.shape[0]\n","  xyz = np.meshgrid(X_r, Y_r, Z_r)\n","  xy_ravel = np.vstack(list(map(np.ravel, xyz))).T\n","  model = Model(receivers = receivers,dtype = 'float32',center_grid_resolution=[1,1,1],regular_grid_resolution=[150,150,30],model_radius=[5000,5000,20200])\n","  ### kernel\n","  centerReg = CenteredRegGrid(model.xy_ravel,radius=[5000,5000,20200],resolution=[9,7,30])\n","  model = activate_customized_grid(model,centerReg)\n","  g_center_regulargrid = GravityPreprocessing(centerReg)\n","  tz_center_regulargrid = tfconstant(g_center_regulargrid.set_tz_kernel())\n","  input = model.get_graph_input()\n","  model.create_tensorflow_graph(input,gradient = True)\n","\n","  # compile kernel\n","  forward_kernel(model.surface_points_coord,model)\n","  start_time = time.time()\n","  forward_kernel(model.surface_points_coord,model)\n","  time_kernel_list.append(time.time() - start_time)\n","\n","\n","  ### convolution\n","  model.activate_regular_grid()\n","  g = GravityPreprocessingRegAllLoop(model,model.grid.regular_grid)\n","  tz = g.set_tz_kernel(model.model_radius,model.regular_grid_resolution)\n","  input = model.get_graph_input()\n","  model.create_tensorflow_graph(input,gradient = True)\n","\n","  # compile convolution\n","  forward_cov(model.surface_points_coord,model)\n","  start_time = time.time()\n","  forward_cov(model.surface_points_coord,model)\n","  time_cov_list.append(time.time() - start_time)\n"],"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","WARNING:tensorflow:5 out of the last 9 calls to <function forward_kernel at 0x7ff9422f8e60> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for  more details.\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","WARNING:tensorflow:5 out of the last 9 calls to <function forward_cov at 0x7ff9421ed8c0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for  more details.\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","WARNING:tensorflow:6 out of the last 11 calls to <function forward_kernel at 0x7ff9422f8e60> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for  more details.\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","WARNING:tensorflow:6 out of the last 11 calls to <function forward_cov at 0x7ff9421ed8c0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for  more details.\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","tracing kernel forwad\n","executing kernel forwad\n","executing kernel forwad\n","tracing convolution forward\n","executing convolution forward\n","executing convolution forward\n"]}]},{"cell_type":"code","metadata":{"id":"UnTvEF5GHYau"},"source":["plt.plot(time_cov_list,'.-',label = 'cov')\n","plt.plot(time_kernel_list,'.-',label = 'kernel')\n","plt.legend()"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"KPluwqI_Ie9s"},"source":["# avg_time_forward_cov = sum(time_cov_list)/len(time_cov_list)\n","# avg_time_forward_cov"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"LykpvedbInqa"},"source":["# avg_time_forward_kernel = sum(time_kernel_list)/len(time_kernel_list)\n","# avg_time_forward_kernel"],"execution_count":null,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"la1HtniDYE62"},"source":["# AD"]},{"cell_type":"code","metadata":{"id":"Bepw0MbY5siB"},"source":["# X_r = np.linspace(704000,740000,10)\n","# Y_r = np.linspace(6.87e6,6.92e6,1)\n","\n","# r = []\n","# for x in X_r:\n","#   for y in Y_r:\n","#     r.append(np.array([x,y]))\n","# receivers = np.array(r)\n","# Z_r = 200\n","# n_devices = receivers.shape[0]"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"3m1oNUv4YGP6"},"source":["# model = Model(receivers = receivers,dtype = 'float32',center_grid_resolution=[1,1,1],regular_grid_resolution=[50,50,30],model_radius=[5000,5000,20200])\n","# centerReg = CenteredRegGrid(model.xy_ravel,radius=[5000,5000,20200],resolution=[9,7,30])\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"yBaIZ9X9dqxz"},"source":["# model = activate_customized_grid(model,centerReg)\n","# g_center_regulargrid = GravityPreprocessing(centerReg)\n","# tz_center_regulargrid = tfconstant(g_center_regulargrid.set_tz_kernel())\n","# input = model.get_graph_input()\n","# model.create_tensorflow_graph(input,gradient = True)\n","@tf.function\n","def grad_func_kernel(mu,model):\n","  with tf.GradientTape() as t:\n","    t.watch(mu)\n","    TFG = TFGraph(input, model.fault_drift,\n","                    model.grid_tensor, model.values_properties, model.nugget_effect_grad,model.nugget_effect_scalar, model.Range,\n","                    model.C_o, model.rescale_factor,slope = 10, dtype = model.tfdtype, gradient = True)\n","    formation_block,property_block,block_mask = TFG.compute_series(mu,\n","                model.dips_position,\n","                model.dip_angles,\n","                model.azimuth,\n","                model.polarity,\n","                model.values_properties)\n","    size = centerReg.values.shape[0]\n","    densities = property_block[:size]\n","\n","    gravity_center_reg = model.TFG.compute_forward_gravity(tz_center_regulargrid, 0, size, densities)\n","    # print('tracing')\n","    # tf.print('executing')\n","  grad = t.gradient(gravity_center_reg,mu)\n","  return grad"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"ey27qR90vb1J"},"source":["# start_time = time.time()\n","# grad_func_kernel(model.surface_points_coord,model)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))\n","# start_time = time.time()\n","# grad_func_kernel(model.surface_points_coord,model)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))\n","# start_time = time.time()\n","# grad_func_kernel(model.surface_points_coord,model)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"3LqAM1yr2sIJ"},"source":["# model = activate_customized_grid(model,centerReg)\n","# g_center_regulargrid = GravityPreprocessing(centerReg)\n","# tz_center_regulargrid = tfconstant(g_center_regulargrid.set_tz_kernel())\n","# input = model.get_graph_input()\n","# model.create_tensorflow_graph(input,gradient = True)\n","# # @tf.function\n","# def hvp(mu,tangents):\n","#     with tf.autodiff.ForwardAccumulator(mu, tangents) as acc:\n","#         with tf.GradientTape(watch_accessed_variables=False) as t:\n","#             t.watch(mu)\n","#             TFG = TFGraph(input, model.fault_drift,\n","#                             model.grid_tensor, model.values_properties, model.nugget_effect_grad,model.nugget_effect_scalar, model.Range,\n","#                             model.C_o, model.rescale_factor,slope = 10, dtype = model.tfdtype, gradient = True)\n","#             formation_block,property_block,block_mask = TFG.compute_series(mu,\n","#                         model.dips_position,\n","#                         model.dip_angles,\n","#                         model.azimuth,\n","#                         model.polarity,\n","#                         model.values_properties)\n","#             size = centerReg.values.shape[0]\n","#             densities = property_block[:size]\n","#             print('tracing')\n","#             tf.print('executing')\n","\n","#             gravity_center_reg = model.TFG.compute_forward_gravity(tz_center_regulargrid, 0, size, densities)\n","#         loss = t.gradient(gravity_center_reg,mu)\n","#     hess = acc.jvp(loss)\n","#     return(hess)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"tF6r6PmW7Z2q"},"source":["# mu = model.surface_points_coord\n","# tangents = np.zeros(mu.shape)\n","# tangents[0]=1\n","# tangents = tf.convert_to_tensor(tangents,dtype=tf.float32)\n","# start_time = time.time()\n","# hvp(mu,tangents)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"jJSJ76gc72VF"},"source":["# start_time = time.time()\n","# hvp(mu,tangents)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))\n","# start_time = time.time()\n","# hvp(mu,tangents)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))\n","# start_time = time.time()\n","# hvp(mu,tangents)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))"],"execution_count":null,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"RaH4nButpEwI"},"source":["The above first order works perfectly, but second order derivative is problematic"]},{"cell_type":"markdown","metadata":{"id":"GDkr_wSc3NPm"},"source":["Convolution scheme"]},{"cell_type":"code","metadata":{"id":"nhRhCD_d0yj0"},"source":["# model.activate_regular_grid()\n","# g = GravityPreprocessingRegAllLoop(model,model.grid.regular_grid)\n","# tz = g.set_tz_kernel(model.model_radius,model.regular_grid_resolution)\n","# input = model.get_graph_input()\n","# model.create_tensorflow_graph(input,gradient = True)\n","@tf.function\n","def grad_func(mu,model):\n","  with tf.GradientTape() as t:\n","    t.watch(mu)\n","    TFG = TFGraph(input, model.fault_drift,\n","                    model.grid_tensor, model.values_properties, model.nugget_effect_grad,model.nugget_effect_scalar, model.Range,\n","                    model.C_o, model.rescale_factor,slope = 1, dtype = model.tfdtype, gradient = True)\n","    formation_block,property_block,block_mask =TFG.compute_series(mu,\n","                model.dips_position,\n","                model.dip_angles,\n","                model.azimuth,\n","                model.polarity,\n","                model.values_properties)\n","\n","    size = tf.reduce_prod(model.regular_grid_resolution)\n","\n","    densities = property_block[0:size]\n","\n","    center_index_x = tf.constant((g.new_xy_ravel[0]-g.model.extent[0])//g.dx,tf.float32)\n","    center_index_y = tf.constant((g.new_xy_ravel[1]-g.model.extent[2])//g.dy,tf.float32)\n","    grav_convolution_full = tf.TensorArray(tf.float32, size=n_devices, dynamic_size=False, clear_after_read=True)\n","    for i in tf.range(n_devices):\n","      c_x = tf.cast(center_index_x[i],tf.int32)\n","      c_y = tf.cast(center_index_y[i],tf.int32)\n","\n","      windowed_densities = tf.reshape(densities,model.regular_grid_resolution)[c_x-g.radius_cell_x:c_x+g.radius_cell_x+1,c_y-g.radius_cell_y:c_y+g.radius_cell_y+1,:]\n","      windowed_densities = tf.squeeze(tf.reshape(windowed_densities,(-1,1)))\n","      grav_ = TFG.compute_forward_gravity(tfconstant(tz), 0, size, windowed_densities)\n","      grav_convolution_full = grav_convolution_full.write(i, grav_)\n","    grav_convolution_full = tf.squeeze(grav_convolution_full.stack())\n","    # print('tracing')\n","    # tf.print('executing')\n","  grad = t.gradient(grav_convolution_full,mu)\n","  return grad\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"JdIEcmCD7RUN"},"source":["# start_time = time.time()\n","# grad_func(model.surface_points_coord,model)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))\n","# start_time = time.time()\n","# grad_func(model.surface_points_coord,model)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))\n","# start_time = time.time()\n","# grad_func(model.surface_points_coord,model)\n","# print(\"--- %s seconds ---\" % (time.time() - start_time))"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"4Op26uTdJ1Yl","executionInfo":{"status":"ok","timestamp":1637943967855,"user_tz":-60,"elapsed":1251836,"user":{"displayName":"george liang","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13911471454046444619"}},"outputId":"431a03c1-4bf2-4311-a64f-c8497584c190"},"source":["# %%\n","from tqdm import tqdm\n","time_cov_list = []\n","time_kernel_list = []\n","time_df = pd.DataFrame(columns=['num_receivers','time','method'])\n","for i in tqdm(range(1,21,1), position=0, leave=True):\n","\n","  X_r = np.linspace(704000,740000,10)\n","  Y_r = np.linspace(6.87e6,6.92e6,i)\n","\n","  r = []\n","  for x in X_r:\n","    for y in Y_r:\n","      r.append(np.array([x,y]))\n","  receivers = np.array(r)\n","  Z_r = 200\n","  n_devices = receivers.shape[0]\n","  xyz = np.meshgrid(X_r, Y_r, Z_r)\n","  xy_ravel = np.vstack(list(map(np.ravel, xyz))).T\n","  model = Model(receivers = receivers,dtype = 'float32',center_grid_resolution=[1,1,1],regular_grid_resolution=[50,50,30],model_radius=[5000,5000,20200])\n","  ### kernel\n","  centerReg = CenteredRegGrid(model.xy_ravel,radius=[5000,5000,20200],resolution=[9,7,30])\n","  model = activate_customized_grid(model,centerReg)\n","  g_center_regulargrid = GravityPreprocessing(centerReg)\n","  tz_center_regulargrid = tfconstant(g_center_regulargrid.set_tz_kernel())\n","  input = model.get_graph_input()\n","  model.create_tensorflow_graph(input,gradient = True)\n","\n","  # compile kernel\n","  grad_func_kernel(model.surface_points_coord,model)\n","  for j in range(10):\n","    start_time = time.time()\n","    grad_func_kernel(model.surface_points_coord,model)\n","    time_kernel = (time.time() - start_time)\n","    time_df = time_df.append({'num_receivers': i*10,'time':time_kernel,'method':'kernel'}, ignore_index=True)\n","  time_kernel_list.append(time.time() - start_time)\n","\n","\n","  ### convolution\n","  model.activate_regular_grid()\n","  g = GravityPreprocessingRegAllLoop(model,model.grid.regular_grid)\n","  tz = g.set_tz_kernel(model.model_radius,model.regular_grid_resolution)\n","  input = model.get_graph_input()\n","  model.create_tensorflow_graph(input,gradient = True)\n","\n","  # compile convolution\n","  grad_func(model.surface_points_coord,model)\n","  for j in range(10):\n","    start_time = time.time()\n","    grad_func(model.surface_points_coord,model)\n","    time_cov = (time.time() - start_time)\n","    time_df = time_df.append({'num_receivers': i*10,'time':time_cov,'method':'cov_50'}, ignore_index=True)\n","  time_cov_list.append(time.time() - start_time)\n","\n","\n","  ### high res cov\n","  model = Model(receivers = receivers,dtype = 'float32',center_grid_resolution=[1,1,1],regular_grid_resolution=[150,150,30],model_radius=[5000,5000,20200])\n","  ### convolution\n","  model.activate_regular_grid()\n","  g = GravityPreprocessingRegAllLoop(model,model.grid.regular_grid)\n","  tz = g.set_tz_kernel(model.model_radius,model.regular_grid_resolution)\n","  input = model.get_graph_input()\n","  model.create_tensorflow_graph(input,gradient = True)\n","\n","  # compile convolution\n","  grad_func(model.surface_points_coord,model)\n","  for j in range(10):\n","    start_time = time.time()\n","    grad_func(model.surface_points_coord,model)\n","    time_cov = (time.time() - start_time)\n","    time_df = time_df.append({'num_receivers': i*10,'time':time_cov,'method':'cov_150'}, ignore_index=True)\n","  time_cov_list.append(time.time() - start_time)\n"],"execution_count":null,"outputs":[{"output_type":"stream","name":"stderr","text":["\r  0%|          | 0/20 [00:00<?, ?it/s]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r  5%|▌         | 1/20 [01:06<20:54, 66.01s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 10%|█         | 2/20 [02:06<18:46, 62.59s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 15%|█▌        | 3/20 [03:06<17:26, 61.55s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 20%|██        | 4/20 [04:06<16:17, 61.09s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 25%|██▌       | 5/20 [05:07<15:14, 60.96s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 30%|███       | 6/20 [06:09<14:15, 61.11s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 35%|███▌      | 7/20 [07:09<13:11, 60.86s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 40%|████      | 8/20 [08:11<12:15, 61.31s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 45%|████▌     | 9/20 [09:12<11:12, 61.09s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 50%|█████     | 10/20 [10:15<10:17, 61.75s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 55%|█████▌    | 11/20 [11:16<09:14, 61.58s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 60%|██████    | 12/20 [12:18<08:12, 61.60s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 65%|██████▌   | 13/20 [13:23<07:18, 62.66s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 70%|███████   | 14/20 [14:25<06:15, 62.55s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 75%|███████▌  | 15/20 [15:28<05:12, 62.57s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 80%|████████  | 16/20 [16:34<04:14, 63.54s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 85%|████████▌ | 17/20 [17:37<03:10, 63.34s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 90%|█████████ | 18/20 [18:39<02:06, 63.19s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["\r 95%|█████████▌| 19/20 [19:43<01:03, 63.31s/it]"]},{"output_type":"stream","name":"stdout","text":["Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n","Active grids: ['regular']\n","Active grids: ['regular' 'sections']\n"]},{"output_type":"stream","name":"stderr","text":["100%|██████████| 20/20 [20:51<00:00, 62.59s/it]\n"]}]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"oAF_FU38KUxk","executionInfo":{"status":"ok","timestamp":1637943968251,"user_tz":-60,"elapsed":413,"user":{"displayName":"george liang","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13911471454046444619"}},"outputId":"32152507-7037-41a1-f803-66a7cd7ebdd2"},"source":["!nvidia-smi"],"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["Fri Nov 26 16:26:07 2021       \n","+-----------------------------------------------------------------------------+\n","| NVIDIA-SMI 495.44       Driver Version: 460.32.03    CUDA Version: 11.2     |\n","|-------------------------------+----------------------+----------------------+\n","| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |\n","| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |\n","|                               |                      |               MIG M. |\n","|===============================+======================+======================|\n","|   0  Tesla P100-PCIE...  Off  | 00000000:00:04.0 Off |                    0 |\n","| N/A   60C    P0    46W / 250W |   8787MiB / 16280MiB |      0%      Default |\n","|                               |                      |                  N/A |\n","+-------------------------------+----------------------+----------------------+\n","                                                                               \n","+-----------------------------------------------------------------------------+\n","| Processes:                                                                  |\n","|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |\n","|        ID   ID                                                   Usage      |\n","|=============================================================================|\n","|  No running processes found                                                 |\n","+-----------------------------------------------------------------------------+\n"]}]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/","height":385},"id":"WULcOO2C9b0f","executionInfo":{"status":"ok","timestamp":1637944600398,"user_tz":-60,"elapsed":3290,"user":{"displayName":"george liang","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13911471454046444619"}},"outputId":"516b6051-c1ba-4f2f-8fdc-364c5bf62690"},"source":["import seaborn as sns\n","sns.set_theme(style=\"whitegrid\",font_scale=1.5)\n","f, ax = plt.subplots(figsize=(8, 5.5))\n","sns.lineplot(x=\"num_receivers\", y=\"time\",\n","             hue=\"method\",style=\"method\",\n","             data=time_df,\n","             markers=True, dashes=False)\n","# plt.savefig('/content/drive/MyDrive/YJ/Green_Stone/figs/ADtime505030.png',dpi=500,bbox_inches='tight')"],"execution_count":null,"outputs":[{"output_type":"display_data","data":{"image/png":"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\n","text/plain":["<Figure size 576x396 with 1 Axes>"]},"metadata":{}}]},{"cell_type":"code","metadata":{"id":"-m7Ouv9E8hs7"},"source":["import seaborn as sns\n","sns.set_theme(style=\"darkgrid\")\n","\n","# Load an example dataset with long-form data\n","fmri = sns.load_dataset(\"fmri\")"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/","height":458},"id":"B-diHyHOKpfv","executionInfo":{"status":"ok","timestamp":1637880319195,"user_tz":-60,"elapsed":448,"user":{"displayName":"george liang","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13911471454046444619"}},"outputId":"ce97b8f5-1eb4-44e6-c80f-bf00fee377d4"},"source":["num_receivers = np.arange(1,100,5)\n","plt.plot(num_receivers,time_cov_list,'.-',label = 'covolution')\n","plt.plot(num_receivers,time_kernel_list,'.-',label = 'kernel')\n","plt.legend()"],"execution_count":null,"outputs":[{"output_type":"execute_result","data":{"text/plain":["<matplotlib.legend.Legend at 0x7f67ce030510>"]},"metadata":{},"execution_count":43},{"output_type":"display_data","data":{"image/png":"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\n","text/plain":["<Figure size 748.8x514.8 with 1 Axes>"]},"metadata":{}}]},{"cell_type":"code","metadata":{"id":"saunSm316iS0"},"source":["model.activate_regular_grid()\n","g = GravityPreprocessingRegAllLoop(model,model.grid.regular_grid)\n","tz = g.set_tz_kernel(model.model_radius,model.regular_grid_resolution)\n","input = model.get_graph_input()\n","model.create_tensorflow_graph(input,gradient = True)\n","def hvp(mu,tangents):\n","    with tf.autodiff.ForwardAccumulator(mu, tangents) as acc:\n","        with tf.GradientTape(watch_accessed_variables=False) as t:\n","            t.watch(mu)\n","            TFG = TFGraph(input, model.fault_drift,\n","                            model.grid_tensor, model.values_properties, model.nugget_effect_grad,model.nugget_effect_scalar, model.Range,\n","                            model.C_o, model.rescale_factor,slope = 1, dtype = model.tfdtype, gradient = True)\n","            formation_block,property_block,block_mask =TFG.compute_series(mu,\n","                        model.dips_position,\n","                        model.dip_angles,\n","                        model.azimuth,\n","                        model.polarity,\n","                        model.values_properties)\n","\n","            size = tf.reduce_prod(model.regular_grid_resolution)\n","\n","            densities = property_block[0:size]\n","\n","            center_index_x = tf.constant((g.new_xy_ravel[0]-g.model.extent[0])//g.dx,tf.float32)\n","            center_index_y = tf.constant((g.new_xy_ravel[1]-g.model.extent[2])//g.dy,tf.float32)\n","            grav_convolution_full = tf.TensorArray(tf.float32, size=n_devices, dynamic_size=False, clear_after_read=True)\n","            for i in tf.range(n_devices):\n","              c_x = tf.cast(center_index_x[i],tf.int32)\n","              c_y = tf.cast(center_index_y[i],tf.int32)\n","\n","              windowed_densities = tf.reshape(densities,model.regular_grid_resolution)[c_x-g.radius_cell_x:c_x+g.radius_cell_x+1,c_y-g.radius_cell_y:c_y+g.radius_cell_y+1,:]\n","              windowed_densities = tf.squeeze(tf.reshape(windowed_densities,(-1,1)))\n","              grav_ = TFG.compute_forward_gravity(tfconstant(tz), 0, size, windowed_densities)\n","              grav_convolution_full = grav_convolution_full.write(i, grav_)\n","            grav_convolution_full = tf.squeeze(grav_convolution_full.stack())\n","\n","            print('tracing')\n","            tf.print('executing')\n","        loss = t.gradient(grav_convolution_full,mu)\n","    hess = acc.jvp(loss)\n","    return(hess)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"qqp0Qze0jPMA","executionInfo":{"status":"ok","timestamp":1637858420094,"user_tz":-60,"elapsed":9324,"user":{"displayName":"george liang","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13911471454046444619"}},"outputId":"9b28f16b-8843-4f58-92fe-188fec78a8fe"},"source":["mu = model.surface_points_coord\n","tangents = np.zeros(mu.shape)\n","tangents[0]=1\n","tangents = tf.convert_to_tensor(tangents,dtype=tf.float32)\n","start_time = time.time()\n","hvp(mu,tangents)\n","print(\"--- %s seconds ---\" % (time.time() - start_time))"],"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["tracing\n","executing\n","--- 9.035943508148193 seconds ---\n"]}]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"uW36a_BQ9B20","executionInfo":{"status":"ok","timestamp":1637858446686,"user_tz":-60,"elapsed":26607,"user":{"displayName":"george liang","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13911471454046444619"}},"outputId":"c6f96727-0eda-4842-f090-f09a1708ad5a"},"source":["start_time = time.time()\n","hvp(mu,tangents)\n","print(\"--- %s seconds ---\" % (time.time() - start_time))\n","start_time = time.time()\n","hvp(mu,tangents)\n","print(\"--- %s seconds ---\" % (time.time() - start_time))\n","start_time = time.time()\n","hvp(mu,tangents)\n","print(\"--- %s seconds ---\" % (time.time() - start_time))"],"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["tracing\n","executing\n","--- 8.787880182266235 seconds ---\n","tracing\n","executing\n","--- 8.629806518554688 seconds ---\n","tracing\n","executing\n","--- 9.168853282928467 seconds ---\n"]}]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"YOCBw7nMpcgA","executionInfo":{"status":"ok","timestamp":1637853235068,"user_tz":-60,"elapsed":10497,"user":{"displayName":"george liang","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13911471454046444619"}},"outputId":"7a53233a-41a9-4033-97bf-3877aedac570"},"source":["start_time = time.time()\n","hvp(mu,tangents)\n","print(\"--- %s seconds ---\" % (time.time() - start_time))\n","start_time = time.time()\n","hvp(mu,tangents)\n","print(\"--- %s seconds ---\" % (time.time() - start_time))\n","start_time = time.time()\n","hvp(mu,tangents)\n","print(\"--- %s seconds ---\" % (time.time() - start_time))"],"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["tracing\n","executing\n","--- 3.417574405670166 seconds ---\n","tracing\n","executing\n","--- 3.4176459312438965 seconds ---\n","tracing\n","executing\n","--- 3.341977834701538 seconds ---\n"]}]},{"cell_type":"markdown","metadata":{"id":"jf9pFyWSpRhN"},"source":["## cov"]},{"cell_type":"code","metadata":{"id":"wYu7KjFilyz8"},"source":["for i in tf.range(20,dtype=tf.int32):\n","  print(i)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"fxQVLxo1yZrg"},"source":["@tf.function\n","def foo(x):\n","  for i in tf.range(tf.constant([2,3,2],tf.int32)[0]):\n","    a = 4\n","  return a\n","\n","foo(2)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"8AAOlc271arj"},"source":["m = tf.constant(3,dtype = tf.float32)\n","with tf.GradientTape(watch_accessed_variables=False) as t:\n","  t.watch(m)\n","  loss = foo(m)\n","t.gradient(loss,m)\n"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"M2hacHwb2Ss0"},"source":["x"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"SZLoDqkK2AQA"},"source":[],"execution_count":null,"outputs":[]}]}