raj-jaiswal-98/Sentiment-Analysis-LSTM

1import numpy as np # linear algebra2import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)3import torch4import torch.nn as nn5import torch.nn.functional as F6# from nltk.corpus import stopwords 7import nltk8# from collections import Counter9import string10import re11# import seaborn as sns12# from tqdm import tqdm13# import matplotlib.pyplot as plt14# from torch.utils.data import TensorDataset, DataLoader15# from sklearn.model_selection import train_test_split16import pickle17import streamlit as st18import time19 20 21is_cuda = torch.cuda.is_available()22 23# If we have a GPU available, we'll set our device to GPU. We'll use this device variable later in our code.24if is_cuda:25    device = torch.device("cuda")26    print("GPU is available")27else:28    device = torch.device("cpu")29    print("GPU not available, CPU used")30 31# device = torch.device("cpu")32output_dim = 133#model class34 35class SentimentLSTM(nn.Module):36    def __init__(self, no_layers, vocab_size, hidden_dim, embedding_dim, drop_prob = 0.5):37        super(SentimentLSTM, self).__init__()38        39        self.no_layers = no_layers40        self.output_dim = output_dim41        self.hidden_dim = hidden_dim42        self.vocab_size = vocab_size43        44        #embedding layer45        self.embedding = nn.Embedding(vocab_size, embedding_dim)  46        47        #LSTM48        self.lstm = nn.LSTM(input_size = embedding_dim, hidden_size = self.hidden_dim, num_layers = no_layers, batch_first=True)49        50        #dropout layers51        self.dropout = nn.Dropout(0.3)52        53        #linear and Sigmoid layer54        55        self.fc = nn.Linear(self.hidden_dim, self.output_dim)56        self.sig = nn.Sigmoid()57        58    def forward(self, x, hidden):59        # we just passed a batch60        batch_size = x.size(0) # batch size -> B61        #embed shape -> [B, max_len, embed_dim]62        embeds = self.embedding(x)63        64        65        lstm_out, hidden = self.lstm(embeds, hidden)66        lstm_out = lstm_out.contiguous().view(-1, self.hidden_dim)67        68        69        # drop out and fully connected70        out = self.dropout(lstm_out)71        out = self.fc(out)72        73        # sigmoid 74        75        sig_out = self.sig(out)76        77        #reshape to batch size first78        79        sig_out = sig_out.view(batch_size, -1)80        81        sig_out = sig_out[:, -1]82        83        84        return sig_out, hidden85    86    87    def init_hidden(self, batch_size):88        89        # create hidden state and cell state tensors with size [no_layers x batch_size x hidden_dim]90        91        hidden_state = torch.zeros((self.no_layers, batch_size, self.hidden_dim)).to(device)92        cell_state = torch.zeros((self.no_layers, batch_size, self.hidden_dim)).to(device)93        hidden = (hidden_state, cell_state)94        return hidden95 96# import saved model with weights from pickle file97 98 99# model = pickle.load(open('model.pkl', 'rb'))100vocab = pickle.load(open('vocab.pkl', 'rb'))101PATH = 'model_state.pkl'102model = SentimentLSTM(2, len(vocab)+1, 256, 64)103model.load_state_dict(torch.load(PATH, map_location=device))104model.eval()105 106# pre-processing input data107 108def preprocess_string(s):109    # remove all characters except letters and digits110    s = re.sub(r"[^\w\s]", '', s)111    #remove all extra whites spaces112    s = re.sub(r"\s+", '', s)113    #remove digits114    s = re.sub(r"\d", '', s)115    return s116 117def padding(sents, seq_len):118    features = np.zeros((len(sents), seq_len), dtype = int)119    for i, rev in enumerate(sents):120        if len(rev) != 0:121            features[i, -len(rev):] = np.array(rev)[:seq_len]122    return features123 124 125 126# predict sentiment of given text127 128 129def predict_sentiment(text):130    word_seq = np.array([vocab[preprocess_string(word)] for word in text.split() if preprocess_string(word) in vocab.keys()])131    word_seq = np.expand_dims(word_seq, axis = 0)132    # print(word_seq)133    pad = torch.from_numpy(padding(word_seq, 500))134    135    inputs = pad.to(device)136    batch_size = 1137    h = model.init_hidden(batch_size)138    output, h = model(inputs, h)139    prob = output.item()140    pred = ''141    if prob > 0.5:142        pred = f"This Statement seems Positive 🤗 to us, with probability of {prob}"143    else:144        pred = f"This Statement seems Negative 😤 to us, with probability of {1-prob}"145    return pred146 147 148# Streamlit UI149 150st.title('Analyse the Sentiment of any Statement 😤/🤗')151 152text = st.text_input("Enter your statement/review here!!")153 154if text != '':155    latest_iteration = st.empty()156    bar = st.progress(1)157 158    for i in range(3):159    # Update the progress bar with each iteration.160        # latest_iteration.text(f'Iteration {i+1}')161        bar.progress((i+1) * 33)162        time.sleep(0.1)163    st.write(predict_sentiment(text))164 165