Different solutions for cvxpy and cvxpylayers

[ManuelP96]

I have a little example of a cvxpy problem

import numpy as np
import cvxpy as cp
import tensorflow as tf
from   cvxpylayers.tensorflow import CvxpyLayer

b=np.ones(6)/6
vol = np.array([0.2, 0.2, 0.15, 0.15, 0.1, 0.1])
rho   = np.array([[   1,  0.7,  0.8,  0.3, -0.2,  0.4],
                  [ 0.7,    1,  0.5,  0.2, -0.1,  0.1], 
                  [ 0.8,  0.5,    1,  0.3, -0.1, 0.1], 
                  [ 0.3,  0.2,  0.3,    1, -0.2,  0.2],
                  [-0.2, -0.1, -0.1, -0.2,    1,    0], 
                  [ 0.4,  0.1,  0.1,  0.2,    0,    1]])
Sigma = np.matmul(np.diag(vol), np.matmul(rho, np.diag(vol)))

btf = tf.Variable(b)
Sigmatf = tf.Variable(Sigma)

w = cp.Variable(6) 

obj = 0.5 * cp.quad_form(w, Sigma) - cp.sum(cp.multiply(b, cp.log(w)))
constr = [w>=0] 
prob = cp.Problem(cp.Minimize(obj), constr)

prob.solve()
w = w/cp.sum(w)

print("Portfolio weights\t", w.value)

but, when I replicate the problem into a cvxpylayer, I get different results. In fact, it seems like the cvxpylayer returns the parameter values:

n,_ = Sigmatf.get_shape()

w = cp.Variable(n)
risk = cp.Parameter(n, nonneg = True)
obj = 0.5 * cp.quad_form(w, Sigma) - risk.T @ cp.log(w)
constr = [w>=0]
prob = cp.Problem(cp.Minimize(obj), constr)
cvxpylayer = CvxpyLayer(prob, parameters=[risk], variables=[w])

b1 = tf.reshape(btf,[6])

with tf.GradientTape() as tape:
    solution,  = cvxpylayer(b1)

solution = solution/tf.reduce_sum(solution)
print(solution.numpy())

[ManuelP96]

I could write the problem using cvxpylayers on torch

b   = np.ones(6)/6                                                          #Risk Budgets
vol = np.array([0.3, 0.25, 0.2, 0.15, 0.1, 0.5])                            #Volatility of Assets
rho = np.array([[   1,  0.7,  0.8,  0.3,  0.2,  0.4],                       #Correlation matrix
                [ 0.7,    1,  0.5,  0.2,  0.1,  0.1],
                [ 0.8,  0.5,    1,  0.3,  0.1,  0.1],
                [ 0.3,  0.2,  0.3,    1,  0.2,  0.2],
                [ 0.2,  0.1,  0.1,  0.2,    1,    0],
                [ 0.4,  0.1,  0.1,  0.2,    0,    1]])
Sigma = np.matmul(np.diag(vol), np.matmul(rho, np.diag(vol)))               #Covariance Matrix

bt = Variable(torch.tensor(b))
Sigmat = Variable(torch.tensor(Sigma))

n = 6

w = cp.Variable(n)
bp = cp.Parameter(n, pos=True)

obj = 0.5 * cp.quad_form(w, Sigma) - cp.sum(cp.multiply(bp, cp.log(w)))
constr = [w>=0]
prob = cp.Problem(cp.Minimize(obj), constr)

cvxpylayer = CvxpyLayer(prob, parameters=[bp], variables=[w])

# solve the problem
solution, = cvxpylayer(bt, solver_args={"solve_method": "ECOS"})

but I need it in tensorflow. Using solver_args={"solve_method": "ECOS"} in tensorflow gives me the error diffcp.cone_program.SolverError: Solver ecos errored.