Spaces:
Sleeping
Sleeping
| import os, sys | |
| import streamlit as st | |
| import pandas as pd | |
| import numpy as np | |
| from sklearn.metrics.pairwise import paired_cosine_distances | |
| from sklearn.preprocessing import normalize | |
| from rolaser import RoLaserEncoder | |
| laser_checkpoint = f"{os.environ['LASER']}/models/laser2.pt" | |
| laser_vocab = f"{os.environ['LASER']}/models/laser2.cvocab" | |
| laser_tokenizer = 'spm' | |
| laser_model = RoLaserEncoder(model_path=laser_checkpoint, vocab=laser_vocab, tokenizer=laser_tokenizer) | |
| rolaser_checkpoint = f"{os.environ['ROLASER']}/models/RoLASER/rolaser.pt" | |
| rolaser_vocab = f"{os.environ['ROLASER']}/models/RoLASER/rolaser.cvocab" | |
| rolaser_tokenizer = 'roberta' | |
| rolaser_model = RoLaserEncoder(model_path=rolaser_checkpoint, vocab=rolaser_vocab, tokenizer=rolaser_tokenizer) | |
| c_rolaser_checkpoint = f"{os.environ['ROLASER']}/models/c-RoLASER/c-rolaser.pt" | |
| c_rolaser_vocab = f"{os.environ['ROLASER']}/models/c-RoLASER/c-rolaser.cvocab" | |
| c_rolaser_tokenizer = 'char' | |
| c_rolaser_model = RoLaserEncoder(model_path=c_rolaser_checkpoint, vocab=c_rolaser_vocab, tokenizer=c_rolaser_tokenizer) | |
| def add_text_inputs(i): | |
| col1, col2 = st.columns(2) | |
| with col1: | |
| text_input1 = st.text_input('Enter standard text here:', f'std{i}') | |
| with col2: | |
| text_input2 = st.text_input('Enter non-standard text here:', f'ugc{i}') | |
| return text_input1, text_input2 | |
| def main(): | |
| st.title('Pairwise Cosine Distance Calculator') | |
| num_pairs = st.sidebar.number_input('Number of Text Input Pairs', min_value=1, max_value=10, value=1) | |
| std_text_inputs = [] | |
| ugc_text_inputs = [] | |
| for i in range(num_pairs): | |
| pair = add_text_inputs(i) | |
| std_text_inputs.append(pair[0]) | |
| ugc_text_inputs.append(pair[1]) | |
| if st.button('Add Text Input Pair'): | |
| pair = add_text_inputs(len(std_text_inputs)) | |
| std_text_inputs.append(pair[0]) | |
| ugc_text_inputs.append(pair[1]) | |
| if st.button('Submit'): | |
| X_std_laser = normalize(laser_model.encode(std_text_inputs)) | |
| X_ugc_laser = normalize(laser_model.encode(ugc_text_inputs)) | |
| X_cos_laser = paired_cosine_distances(X_std_laser, X_ugc_laser) | |
| X_std_rolaser = normalize(rolaser_model.encode(std_text_inputs)) | |
| X_ugc_rolaser = normalize(rolaser_model.encode(ugc_text_inputs)) | |
| X_cos_rolaser = paired_cosine_distances(X_std_rolaser, X_ugc_rolaser) | |
| X_std_c_rolaser = normalize(c_rolaser_model.encode(std_text_inputs)) | |
| X_ugc_c_rolaser = normalize(c_rolaser_model.encode(ugc_text_inputs)) | |
| X_cos_c_rolaser = paired_cosine_distances(X_std_c_rolaser, X_ugc_c_rolaser) | |
| outputs = pd.DataFrame(columns=[ 'model', 'pair', 'ugc', 'std', 'cos']) | |
| outputs['model'] = np.repeat(['LASER', 'RoLASER', 'C-RoLASER'], 3) | |
| outputs['pair'] = np.tile(np.arange(1,num_pairs+1), 3) | |
| outputs['std'] = np.tile(std_text_inputs, 3) | |
| outputs['ugc'] = np.tile(ugc_text_inputs, 3) | |
| outputs['cos'] = np.concatenate([X_cos_laser, X_cos_rolaser, X_cos_c_rolaser], axis=1) | |
| st.write('## Cosine Distance Scores:') | |
| st.bar_chart(outputs, x='pair', y='cos', color='model', title='Cosine Distance Scores', xlabel='Text Input Pair', ylabel='Cosine Distance', legend='Model') | |
| st.write('## Average Cosine Distance Scores:') | |
| st.write(f'LASER: {outputs[outputs["model"]=="LASER"]["cos"].mean()}') | |
| st.write(f'RoLASER: {outputs[outputs["model"]=="RoLASER"]["cos"].mean()}') | |
| st.write(f'C-RoLASER: {outputs[outputs["model"]=="C-RoLASER"]["cos"].mean()}') | |
| if __name__ == "__main__": | |
| main() | |