Text CNN baseline

text_cnn.py

import torch
import torch.nn as nn

device = 'cuda' is torch.cuda.is_available() else 'cpu'

class NgramModule(nn.Module):
    
    def __init__(self, seq_len, kernel, channels):
        super().__init__()
        self.conv = nn.Conv2d(1, channels, kernel)
        self.pool = nn.MaxPool1d(seq_len - kernel[0] + 1)
        self.relu = nn.ReLU(inplace=True)
    
    def forward(self, x):
        x = self.relu(self.conv(x)).squeeze(-1)
        return self.pool(x).squeeze(-1)

class CNN(nn.Module):

    def __init__(self, ksizes, n_channels, embedding_dim, seq_len,
                 dropout, n_classes, pretrained):
        super().__init__()
        self.embedding = nn.Embedding.from_pretrained(pretrained, freeze=False)
        self.blocks = [NgramModule(seq_len, (ksize, embedding_dim), \
                       n_channels).to(device) for ksize in ksizes]
        self.dropout = nn.Dropout(dropout)
        self.fc = nn.Linear(n_channels*len(ksizes), n_classes)

    def forward(self, x):
        x = self.embedding(x)[:,None,:,:] # [B,1,T,E]
        x = torch.cat([block(x) for block in self.blocks], dim=-1).to(device)
        return self.fc(self.dropout(x))