Problem with ECL recognizing a Python class through Python embedding

Question

So, I basically wanted to make use of PyTorch on ECL, through python embedding.When we constructed the architecture of a simple model by creating a class as shown below, we observed a few unusual errors like the class name not being defined:

f() := EMBED(PY)
import torch
import torch.nn as nn
import torch.optim as optim
class model_1(nn.Module):
 def _init_(self):
 super(model_1, self)._init_()
 input_size = torch.randint(10, 100, (1,)).item()
 hidden_size = torch.randint(10, 100, (1,)).item()
 output_size = torch.randint(2, 10, (1,)).item()
 self.layer1 = nn.Linear(input_size, hidden_size)
 self.activation = nn.ReLU()
 self.layer2 = nn.Linear(hidden_size, output_size)
 self.softmax = nn.Softmax(dim=1)
 
 def forward(self, x):
 x = self.layer1(x)
 x = self.activation(x)
 x = self.layer2(x)
 x = self.softmax(x)
 return x
model = model_1()
batch_size = 32
input_size = model.layer1.in_features
input_data = torch.randn(batch_size, input_size)
target = torch.randint(0, model.layer2.out_features, (batch_size,))
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=0.01)
num_epochs = 10
for epoch in range(num_epochs):
 output = model(input_data)
 
 loss = criterion(output, target)
 
 
 print(f'Epoch [{epoch+1}/{num_epochs}], Loss: {loss.item():.4f}')
 
 optimizer.zero_grad()
 loss.backward()
 optimizer.step()
ENDEMBED;
f();

We were getting the following error:

Error: 0: pyembed: name 'model_1' is not defined

We also tried to run the above model without classes just by using functions, and this is what we are getting:

f() := EMBED(PY)
import torch
import torch.nn as nn
import torch.optim as optim
def create_model():
 input_size = torch.randint(10, 100, (1,)).item()
 hidden_size = torch.randint(10, 100, (1,)).item()
 output_size = torch.randint(2, 10, (1,)).item()
 
 layer1 = nn.Linear(input_size, hidden_size)
 activation = nn.ReLU()
 layer2 = nn.Linear(hidden_size, output_size)
 softmax = nn.Softmax(dim=1)
 
 def forward(x):
 x = layer1(x)
 x = activation(x)
 x = layer2(x)
 x = softmax(x)
 return x
 
 
 parameters = list(layer1.parameters()) + list(layer2.parameters())
 
 return forward, parameters
model, parameters = create_model()
batch_size = 32
input_size = parameters[0].size(1) # Input size of the first layer
input_data = torch.randn(batch_size, input_size)
target = torch.randint(0, parameters[-1].size(0), (batch_size,)) 
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(parameters, lr=0.01)
num_epochs = 10
for epoch in range(num_epochs):
 output = model(input_data)
 
 loss = criterion(output, target)
 
 print(f'Epoch [{epoch+1}/{num_epochs}], Loss: {loss.item():.4f}')
 
 optimizer.zero_grad()
 loss.backward()
 optimizer.step()
ENDEMBED;
f();

On running this, we were getting the following error (Although torch is already configured, I ran an import torch in a simple python program, didn't give me an error):

Error: 0: pyembed: name 'torch' is not defined

How can I create an object of the mentioned class above using python embedding?

Answer 1

You need to have a common global area shared between calls. Your class definition is out of scope when you are trying to use it. You also need to define a persist setting. It’s all in the following blog:

Advanced Python Embedding

Here is another blog you may find helpful:

Power Macros in Python

Thanks for your question!