NMMH_TH3_21C29001 (1).ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import math\n",
    "import requests\n",
    "import io"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "train=pd.read_csv( \"https://github.com/huynhthanh98/ML/raw/ML-2022/Lab03/bt_train.csv\")\n",
    "valid=pd.read_csv( \"https://github.com/huynhthanh98/ML/raw/ML-2022/Lab03/bt_valid.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>x1</th>\n",
       "      <th>x2</th>\n",
       "      <th>label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>55.808717</td>\n",
       "      <td>54.602061</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>50.882672</td>\n",
       "      <td>52.736342</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>53.196444</td>\n",
       "      <td>44.208365</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>54.727839</td>\n",
       "      <td>60.406845</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>49.771372</td>\n",
       "      <td>48.405605</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          x1         x2  label\n",
       "0  55.808717  54.602061      0\n",
       "1  50.882672  52.736342      0\n",
       "2  53.196444  44.208365      0\n",
       "3  54.727839  60.406845      0\n",
       "4  49.771372  48.405605      0"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0, 1, 2])"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train['label'].unique()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x112048128>"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x1_train = train['x1'].values\n",
    "x2_train = train['x2'].values\n",
    "y_train = train['label'].values\n",
    "\n",
    "plt.scatter(x1_train[y_train == 0], x2_train[y_train==0])\n",
    "plt.scatter(x1_train[y_train == 1], x2_train[y_train==1])\n",
    "plt.scatter(x1_train[y_train == 2], x2_train[y_train==1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x11419aa90>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x1_valid = valid['x1'].values\n",
    "x2_valid = valid['x2'].values\n",
    "y_valid = valid['label'].values\n",
    "\n",
    "plt.scatter(x1_valid[y_valid == 0], x2_valid[y_valid==0])\n",
    "plt.scatter(x1_valid[y_valid == 1], x2_valid[y_valid==1])\n",
    "plt.scatter(x1_train[y_train == 2], x2_train[y_train==1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "#read data\n",
    "X_train_ori=train.values[:,:-1]\n",
    "y_train=train.values[:,-1].reshape(-1, 1)\n",
    "X_valid_ori=valid.values[:,:-1]\n",
    "y_valid=valid.values[:,-1].reshape(-1, 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### NORMALIZE FEATURE AND ONEHOT ENDCODING TARGET"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "def onehot(y_train):\n",
    "    nb_classes = len(np.unique(y_train))\n",
    "    targets = y_train.reshape(-1)\n",
    "    one_hot_targets = np.eye(nb_classes)[targets.astype(int)]\n",
    "    return one_hot_targets "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "def normalization(X_train, y_train):\n",
    "    X_mean = X_train.mean(axis=0)\n",
    "    X_std = X_train.std(axis=0) \n",
    "    X_train_norm=(X_train-X_train.mean(axis=0))/X_train.std(axis=0)\n",
    "    X_train_full=np.concatenate([X_train_norm, \n",
    "                           np.ones([X_train.shape[0], 1])], axis=-1)\n",
    "    \n",
    "    return (X_train_norm, X_train_full)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train_norm, X_train = normalization(X_train_ori, y_train)\n",
    "y_train_onehot=onehot(y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_valid_norm, X_valid = normalization(X_valid_ori, y_valid)\n",
    "y_valid_onehot=onehot(y_valid)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### MULTIPLE LOGISTICS REGRESSION USING SOFTMAX\n",
    "\n",
    "\n",
    "**- Ta cần dự đoán phân phối:**  $P_{\\theta}(c | x)$; Với c in {1,2,3}  \n",
    "\n",
    "**- Giả thiết**: \n",
    "$P(y^{(i)} = c  | x^{(i)}, \\theta) = \\frac{exp(\\theta^{(c)\\top} * x^{(i)})}{\\sum_{j=0}^{2} exp(\\theta^{(j)\\top} * x^{(i)})} $ với c in {1,2,3} \n",
    "\n",
    "**- Hàm mất mát**: (Cross Entropy Loss)\n",
    "\n",
    "$J(\\theta) = - \\frac{1}{N}\\sum_{i=1}^{N}\\sum_{c = 0}^{2} 1\\{y^{y^{(i)}=c}\\} log(\\frac{exp(\\theta^{(c)\\top} * x^{(i)})}{\\sum_{j=0}^{2} exp(\\theta^{(j)\\top} * x^{(i)})})$\n",
    " \n",
    "**- Gradient Descent:** \n",
    "\n",
    "$\\Delta_{\\theta^{(k)}} J(\\theta) = - \\sum_{i = 1}^{N}[ x_{i}(1\\{y_{(i)}=c\\}  - P(y^{(i)} = c  | x^{(i)}, \\theta)]   $   "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "def softmax(Z):\n",
    "    return np.exp(Z)*1/np.sum(np.exp(Z), axis=1).reshape(-1,1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# X = X * theta\n",
    "def calculate_Z(X, theta):\n",
    "    return X*theta # shape (n, p+1) * (p+1, k)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "def initialize_parameters(X, y_train):\n",
    "    # X (n, p+1), num_clas = c\n",
    "    num_class=len(np.unique(y_train))\n",
    "    theta = np.zeros([num_class, X.shape[1]]) #shape (c, p+1) \n",
    "    return theta "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "def one_forward_pass(X, theta):\n",
    "    Z = np.dot(X, theta.T)  # shape (n, p+1) * (p+1)*c = (n, c)\n",
    "    y_hat_matrix = softmax(Z) # shape: (n, c) \n",
    "    return y_hat_matrix # shape (n, c) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**- Gradient Descent:** \n",
    "\n",
    "$\\Delta_{\\theta^{(k)}} J(\\theta) = - \\sum_{i = 1}^{N}[ x_{i}(1\\{y_{(i)}=c\\}  - P(y^{(i)} = c  | x^{(i)}, \\theta)]   $   "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calc_grad(X, y_matrix_onehot, y_hat_matrix):\n",
    "    error=y_hat_matrix-y_matrix_onehot #(n, c) ; X (n, p+1)\n",
    "    classes=np.unique(y_train).astype(int)\n",
    "    grads=np.array([np.sum(X * error[:, i].reshape(-1,1), axis=0) for i in classes])\n",
    "    # 1 c la 1 rows \n",
    "    return grads"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "**- Hàm Loss**: (Cross Entropy Loss)\n",
    "\n",
    "$J(\\theta) = - \\frac{1}{N}\\sum_{i=1}^{N}\\sum_{c = 0}^{2} 1\\{y^{y^{(i)}=c}\\} log(\\frac{exp(\\theta^{(c)\\top} * x^{(i)})}{\\sum_{j=0}^{2} exp(\\theta^{(j)\\top} * x^{(i)})})$ "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "def loss_calc(y_onehot, y_hat_matrix): #(n, c)  \n",
    "    N=y_onehot.shape[0]\n",
    "    return -1/N*np.sum(np.sum(y_onehot*np.log(y_hat_matrix), axis=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_hat = np.array([0.14, 0., .8])\n",
    "np.argmax(y_hat)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_accuracy(y_train, y_predict_score_matrix):\n",
    "    '''\n",
    "    y_predict_score_matrix: result is list if float in (0,1)\n",
    "    '''\n",
    "    y_predict_onehot=np.argmax(y_predict_score_matrix, axis=1)\n",
    "    acc_score=np.mean((y_train.reshape(-1)==y_predict_onehot))\n",
    "    return acc_score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iter: 0, Loss Train: 1.0986 - Loss Valid: 0.9190 | Train Acc 0.3333 - Valid Acc 0.6567 \n",
      "Iter: 10000, Loss Train: 0.2876 - Loss Valid: 1.0009 | Train Acc 0.8856 - Valid Acc 0.7000 \n",
      "Iter: 20000, Loss Train: 0.2876 - Loss Valid: 1.0009 | Train Acc 0.8856 - Valid Acc 0.7000 \n",
      "Iter: 30000, Loss Train: 0.2876 - Loss Valid: 1.0009 | Train Acc 0.8856 - Valid Acc 0.7000 \n",
      "Iter: 40000, Loss Train: 0.2876 - Loss Valid: 1.0009 | Train Acc 0.8856 - Valid Acc 0.7000 \n",
      "Iter: 50000, Loss Train: 0.2876 - Loss Valid: 1.0009 | Train Acc 0.8856 - Valid Acc 0.7000 \n",
      "Iter: 60000, Loss Train: 0.2876 - Loss Valid: 1.0009 | Train Acc 0.8856 - Valid Acc 0.7000 \n",
      "Iter: 70000, Loss Train: 0.2876 - Loss Valid: 1.0009 | Train Acc 0.8856 - Valid Acc 0.7000 \n",
      "Iter: 80000, Loss Train: 0.2876 - Loss Valid: 1.0009 | Train Acc 0.8856 - Valid Acc 0.7000 \n",
      "Iter: 90000, Loss Train: 0.2876 - Loss Valid: 1.0009 | Train Acc 0.8856 - Valid Acc 0.7000 \n",
      "tham số Theta đọc được:  [[-5.25252678  1.04712485 -0.72833223]\n",
      " [-0.51678667 -1.63432787  1.00848156]\n",
      " [ 5.76931346  0.58720302 -0.28014934]]\n"
     ]
    }
   ],
   "source": [
    "theta=initialize_parameters(X_train,y_train)\n",
    "n_iters = 100000\n",
    "learning_rate=0.001 \n",
    "\n",
    "for i in range(n_iters): \n",
    "    y_matrix_onehot=onehot(y_train) \n",
    "    y_train_hat_matrix=one_forward_pass(X_train, theta)\n",
    "    loss_train=loss_calc(y_train_onehot, y_train_hat_matrix) \n",
    "    grads=calc_grad(X_train, y_matrix_onehot, y_train_hat_matrix)\n",
    "    theta -= learning_rate*grads \n",
    "    \n",
    "    if i%10000==0:\n",
    "        y_valid_predict_score_matrix=one_forward_pass(X_valid, theta)\n",
    "        loss_val=loss_calc(y_valid_onehot, y_valid_predict_score_matrix) \n",
    "        \n",
    "        train_acc=get_accuracy(y_train, y_train_hat_matrix) \n",
    "        valid_acc=get_accuracy(y_valid, y_valid_predict_score_matrix) \n",
    "        \n",
    "        print(\"Iter: {}, Loss Train: {:.4f} - Loss Valid: {:.4f} | Train Acc {:.4f} - Valid Acc {:.4f} \".format(\n",
    "            i, \n",
    "            loss_train, \n",
    "            loss_val,\n",
    "            train_acc, \n",
    "            valid_acc, \n",
    "            )\n",
    "         )\n",
    "    # printing the Loss and \n",
    "print(\"tham số Theta đọc được: \", theta)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## SKlearn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.metrics import accuracy_score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = LogisticRegression(multi_class='multinomial', solver='lbfgs')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LogisticRegression(multi_class='multinomial')"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train, y_train.reshape(-1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred= model.predict(X_valid)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.6933333333333334"
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(y_valid, y_pred) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Ta thấy kết quả từ mô hình ta xây dựng và từ SKlearn có Accurracy trên tập 'bt_valid.csv' khá tương đương nhau. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python (reco)",
   "language": "python",
   "name": "reco_base"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}