# Issue with creating a mathematical function through neural network

Dear all,

I need to imitate a mathematical function (u(x,y)) by a neural network.
The loss I need to minimize is (-u’‘(x,y)-f(x,y))^2 where u’’ is the second derivative of u and f is a function corresponding to this second derivative. In fact, I am comparing the Laplacian of both functions.
In other word, I am trying to first determine u’’ for a given f(x,y), and after, using autograd pytorch function, I try to estimate u.

The following code is working fine but the second dimension (y) is not taken into account. See the 2 graphs after the codes : the 1st one stands for the nnfunction and the second one is the real function I wish to imitate.
Any clue why it still does not work properly ?

import numpy as np
import matplotlib.pyplot as plt
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
learning_rate = 0.01
num_epochs = 200
N = 200
A = []
U = lambda x: torch.sin(torch.pi * x) * torch.sin(2 * torch.pi * x)
f = lambda x, u: (5 * torch.pi**2* torch.sin(torch.pi * x) * torch.sin(2 * torch.pi * x))
x = np.linspace(0, 1, N)
y = np.linspace(0, 1, N)
z = []
for i in range(N):
z.append([x[i],y[i]])
class Net(nn.Module):
def init(self):
super(Net, self).init()
self.fc = nn.Sequential(nn.Linear(2, 64), nn.Tanh(), nn.Linear(64, 1, bias = False))
self.l1 = nn.Linear(2,128,)
self.l2 = nn.Linear(128,10)
self.l3 = nn.Linear(10,1)

def forward(self,input):
x = input
y = input
out = F.tanh(self.l1(input))
out = F.tanh(self.l2(out))
out = self.l3(out)
return out * x * (1-x) * y * (1-y)
model = Net()
def loss(input, U, A, f):
nablaU = None
for i in input:
output = model(i)
#print(“output”, output)
#print(“Uneuronal_dx”, Uneuronal_dx)
#print(“Uneuronal_d2x”, Uneuronal_d2x)

``````Uneuronal_dy = torch.autograd.grad(output, i, grad_outputs = torch.ones_like(output), create_graph=True, retain_graph=True)
#print("Uneuronal_dy", Uneuronal_dy)
#print("Uneuronal_d2y", Uneuronal_d2y)

v = torch.unsqueeze(torch.pow(torch.sub(torch.mul(torch.add(Uneuronal_d2x, Uneuronal_d2y), -1),f(i, U)), 2), dim=-1)
#print("v", v)
if nablaU is None:
nablaU = v
#print("nablaU", nablaU)
else:
nablaU = torch.cat( (nablaU ,v), dim=0)
#print("nablaU", nablaU)
``````

return(torch.mean(nablaU))
for i in range(num_epochs):
l = loss(input, U, A, f)
l.backward()
print(“loss”, l)
def u_plot(x, y):
return np.sin(np.pi * x) * np.sin(2 * np.pi * y)
input1 = input[:,0].detach().numpy()
input2 = input[:,1].detach().numpy()
output = []
for i in input:
output.append([model.forward(i).detach().numpy()])
output = np.array(output).reshape(N, 1)
fig = plt.figure()
plt.subplots(figsize=(15, 8))
ax = plt.axes(projection=‘3d’)
X, Y = np.meshgrid(input1, input2)
ax.plot_surface(Y, X, output, rstride=1, cstride=1, edgecolor=‘none’)
ax.set_title(‘Fontion \$u(x,y)\$’);
ax.set_xlabel(‘y’)
ax.set_ylabel(‘x’)
ax.set_zlabel(‘Network’)
fig = plt.figure()
plt.subplots(figsize=(15, 8))
ax = plt.axes(projection=‘3d’)
X, Y = np.meshgrid(input1, input2)
ax.plot_surface(Y, X, u_plot(X, Y), rstride=1, cstride=1, edgecolor=‘none’)
ax.set_title(‘Fontion \$u(x,y)\$’);
ax.set_xlabel(‘y’)
ax.set_ylabel(‘x’)
ax.set_zlabel(‘u(x, y)’);

Idea is after to generalize this into dim 4, 5…
Can someone help?

Looking more into details, I was confirmed by colleagues that there were some bugs in Pytorch since there is no error at all. Can someone from Pytorch just confirm this?

I don’t think anyone would confirm speculative claims about unknown errors, so could you add more details please and explain what exactly is broken?

Also, you have still not followed up on my previous request to check if you are detaching the computation graph as described in this post.

Also, please put formatting for python code correctly. It makes it way more readable for others.

I have followed up your previous request. I checked but the backward function seems to work properly. For a reason which seems to have nothing to do with your comments, the network seems to ignore y. That is our my only worry

OK, I think the code is more easy to read like this :

import numpy as np
import matplotlib.pyplot as plt
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np

learning_rate = 0.01
num_epochs = 200
N = 200
A = []
U = lambda x: torch.sin(torch.pi * x) * torch.sin(2 * torch.pi * x)
f = lambda x, u: (5 * torch.pi**2* torch.sin(torch.pi * x) * torch.sin(2 * torch.pi * x))
x = np.linspace(0, 1, N)
y = np.linspace(0, 1, N)
z = []
for i in range(N):
z.append([x[i],y[i]])

class Net(nn.Module):
def init(self):
super(Net, self).init()
self.fc = nn.Sequential(nn.Linear(2, 64), nn.Tanh(), nn.Linear(64, 1, bias = False))
self.l1 = nn.Linear(2,128,)
self.l2 = nn.Linear(128,10)
self.l3 = nn.Linear(10,1)

def forward(self,input):
x = input
y = input
out = F.tanh(self.l1(input))
out = F.tanh(self.l2(out))
out = self.l3(out)
return out * x * (1-x) * y * (1-y)

model = Net()
def loss(input, U, A, f):
nablaU = None
for i in input:
output = model(i)

v = torch.unsqueeze(torch.pow(torch.sub(torch.mul(torch.add(Uneuronal_d2x, Uneuronal_d2y), -1),f(i, U)), 2), dim=-1)
if nablaU is None:
nablaU = v
else:
nablaU = torch.cat( (nablaU ,v), dim=0)
return(torch.mean(nablaU))

for i in range(num_epochs):
l = loss(input, U, A, f)
l.backward()
print(“loss”, l)
def u_plot(x, y):
return np.sin(np.pi * x) * np.sin(2 * np.pi * y)
input1 = input[:,0].detach().numpy()
input2 = input[:,1].detach().numpy()
output = []
for i in input:
output.append([model.forward(i).detach().numpy()])
output = np.array(output).reshape(N, 1)

fig = plt.figure()
plt.subplots(figsize=(15, 8))
ax = plt.axes(projection=‘3d’)
X, Y = np.meshgrid(input1, input2)
ax.plot_surface(Y, X, output, rstride=1, cstride=1, edgecolor=‘none’)
ax.set_title(‘Fontion \$u(x,y)\$’);
ax.set_xlabel(‘y’)
ax.set_ylabel(‘x’)
ax.set_zlabel(‘Network’)

fig = plt.figure()
plt.subplots(figsize=(15, 8))
ax = plt.axes(projection=‘3d’)
X, Y = np.meshgrid(input1, input2)
ax.plot_surface(Y, X, u_plot(X, Y), rstride=1, cstride=1, edgecolor=‘none’)
ax.set_title(‘Fontion \$u(x,y)\$’);
ax.set_xlabel(‘y’)
ax.set_ylabel(‘x’)
ax.set_zlabel(‘u(x, y)’);

What I meant is to use the symbol ```

You can put the entire code between two of these symbols, like this ```` Your python code ````.
For instance, in the format you just used, nested structure of function definitions and loops is not clear.
If you use the symbol ```, it will respect the original format of python code.
If you change the original post according to this, it will be the best

``````import numpy as np
import matplotlib.pyplot as plt
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np

learning_rate = 0.01
num_epochs = 200

N = 200
A = []

U = lambda x: torch.sin(torch.pi * x) * torch.sin(2 * torch.pi * x)
f = lambda x, u: (5 * torch.pi**2* torch.sin(torch.pi * x) * torch.sin(2 * torch.pi * x))

x = np.linspace(0, 1, N)
y = np.linspace(0, 1, N)

z = []
for i in range(N):
z.append([x[i],y[i]])

class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.fc = nn.Sequential(nn.Linear(2, 64), nn.Tanh(), nn.Linear(64, 1, bias = False))

self.l1 = nn.Linear(2,128,)
self.l2 = nn.Linear(128,10)
self.l3 = nn.Linear(10,1)

def forward(self,input):
x = input
y = input
out = F.tanh(self.l1(input))
out = F.tanh(self.l2(out))
out = self.l3(out)
return out * x * (1-x) * y * (1-y)

model = Net()

def loss(input, U, A, f):
nablaU = None

for i in input:
output = model(i)

v = torch.unsqueeze(torch.pow(torch.sub(torch.mul(torch.add(Uneuronal_d2x, Uneuronal_d2y), -1),f(i, U)), 2), dim=-1)

if nablaU is None:
nablaU = v
else:
nablaU = torch.cat( (nablaU ,v), dim=0)

return(torch.mean(nablaU))

for i in range(num_epochs):
l = loss(input, U, A, f)
l.backward()
print("loss", l)

def u_plot(x, y):
return np.sin(np.pi * x) * np.sin(2 * np.pi * y)

input1 = input[:,0].detach().numpy()
input2 = input[:,1].detach().numpy()

output = []
for i in input:
output.append([model.forward(i).detach().numpy()])

output = np.array(output).reshape(N, 1)

fig = plt.figure()
plt.subplots(figsize=(15, 8))
ax = plt.axes(projection='3d')
X, Y = np.meshgrid(input1, input2)
ax.plot_surface(Y, X, output, rstride=1, cstride=1, edgecolor='none')
ax.set_title('Fontion \$u(x,y)\$');
ax.set_xlabel('y')
ax.set_ylabel('x')
ax.set_zlabel('Network');

fig = plt.figure()
plt.subplots(figsize=(15, 8))
ax = plt.axes(projection='3d')
X, Y = np.meshgrid(input1, input2)
ax.plot_surface(Y, X, u_plot(X, Y), rstride=1, cstride=1, edgecolor='none')
ax.set_title('Fontion \$u(x,y)\$');
ax.set_xlabel('y')
ax.set_ylabel('x')
ax.set_zlabel('u(x, y)');

``````

So I have just sent a readable version.
Again, aim is to train a neural network minimimizing (-u’'-f)^2 which is the loss function. Then, the use of the autograd functions shall help find the real u.
In dim 3, we are comparing laplacian.

According to me, the error is to find in

``````      Uneuronal_dx = torch.autograd.grad(output, i, grad_outputs = torch.ones_like(output), create_graph=True, retain_graph=True)