sentiment-lstm

pytorchv1.0.0

Classifies text sequences into positive/negative sentiment using a bidirectional LSTM with learned attention weights. The attention mechanism computes importance scores for each timestep, enabling the model to focus on sentiment-bearing words. Includes embedding layer, dropout, and synthetic data generation for self-contained training.

NLPtext-classificationsentimentlstmintermediate

Install

1openmodelstudio install sentiment-lstm

SDK Usage

1import openmodelstudio as oms
2
3model = oms.use_model("sentiment-lstm")
4handle = oms.register_model("my-sentiment-lstm", model=model)
5job = oms.start_training(handle.model_id, wait=True)

Source Preview(104 lines)

View full source

1"""Sentiment analysis with bidirectional LSTM."""
2
3import torch
4import torch.nn as nn
5
6class SentimentLSTM(nn.Module):
7    def __init__(self, vocab_size=10000, embed_dim=128,
8                 hidden_dim=256, n_classes=2):
9        super().__init__()
10        self.embedding = nn.Embedding(vocab_size, embed_dim)
11        self.lstm = nn.LSTM(embed_dim, hidden_dim,
12                            batch_first=True, bidirectional=True)
13        self.attention = nn.Linear(hidden_dim * 2, 1)
14        self.classifier = nn.Linear(hidden_dim * 2, n_classes)
15
16    def forward(self, x):
17        embedded = self.embedding(x)
18        lstm_out, _ = self.lstm(embedded)
19        attn = torch.softmax(self.attention(lstm_out), dim=1)
20        context = (lstm_out * attn).sum(dim=1)
21        return self.classifier(context)

Details

Author: openmodelstudio
License: MIT
Source: 104 lines
Version: 1.0.0

Dependencies

torch>=2.0

numpy>=1.24

Hyperparameters

epochs10

Number of training epochs

lr0.001

Learning rate

hidden_dim256

LSTM hidden dimension

vocab_size10000

Vocabulary size

View on GitHub