utils/MLP.py

import torch.nn as nn
import torch

class MLP(nn.Module):
    def __init__(self, input_size, output_size, n_neurons, dropout_rates):
        super(MLP, self).__init__()
        self.input_size = input_size
        self.output_size = output_size
        self.n_neurons = n_neurons
        self.dropout_rates = dropout_rates
        self.fc = nn.ModuleList()
        self.dropout = nn.ModuleList()
        self.fc.append(nn.Linear(input_size, n_neurons[0]))
        self.dropout.append(nn.Dropout(dropout_rates[0]))
        for i in range(1, len(n_neurons)):
            self.fc.append(nn.Linear(n_neurons[i-1], n_neurons[i]))
            self.dropout.append(nn.Dropout(dropout_rates[i]))

        self.fc_out = nn.Linear(n_neurons[-1], output_size)

    def forward(self, x, apply_activation= False):
        x = nn.functional.relu(self.fc[0](x))
        x = self.dropout[0](x)
        for i in range(1, len(self.fc)):
            x = nn.functional.relu(self.fc[i](x))
            x = self.dropout[i](x)
        x = self.fc_out(x)

        # Apply activation only if requested (for inference)
        if apply_activation:
            x = torch.sigmoid(x)
                
        return x
    
    def predict(self, x, apply_activation=True):
        """Convenience method for inference with activations applied"""
        self.eval()
        with torch.no_grad():
            return self.forward(x, apply_activation=apply_activation)