app.py

import streamlit as st
import pandas as pd
from transformers import pipeline

from pprint import pprint
from datasets import load_dataset
from torch.utils.data import DataLoader

st.title("CS634 - milestone2 - Tedi Pano")

@st.cache_resource
def load_data():
    dataset_dict = load_dataset('HUPD/hupd',
        name='sample',
        data_files="https://huggingface.co/datasets/HUPD/hupd/blob/main/hupd_metadata_2022-02-22.feather", 
        icpr_label=None,
        train_filing_start_date='2016-01-01',
        train_filing_end_date='2016-01-21',
        val_filing_start_date='2016-01-22',
        val_filing_end_date='2016-01-31',
    )

    st.write('Loading is done!')
    return dataset_dict

@st.cache_resource
def training_computation(_dataset_dict):
    df = pd.DataFrame(_dataset_dict['train'])
    vf = pd.DataFrame(_dataset_dict['validation'])
    
    accepted_rejected = ['ACCEPTED', 'REJECTED']
    df = df[df['decision'].isin(accepted_rejected)]
    df['patentability_score'] = df['decision'].map({'ACCEPTED': 1, 'REJECTED': 0})
    vf = vf[vf['decision'].isin(accepted_rejected)]
    vf['patentability_score'] = vf['decision'].map({'ACCEPTED': 1, 'REJECTED': 0})
    
    st.write("Processed the data")
    
    
    from sklearn.model_selection import train_test_split
    dftrain, dftest = train_test_split(df, test_size = 0.90, random_state = 0)
    
    from transformers import DistilBertTokenizerFast
    tokenizer = DistilBertTokenizerFast.from_pretrained('distilbert-base-uncased')
    
    X_dtrain = dftrain['abstract'].tolist()
    y_dtrain = dftrain['patentability_score'].tolist()
    
    X_vtrain = vf['abstract'].tolist()
    y_vtrain = vf['patentability_score'].tolist()
    
    X_dtest = dftest['abstract'].tolist()
    y_dtest = dftest['patentability_score'].tolist()
    
    train_encodings = tokenizer(X_dtrain, truncation=True, padding=True)
    val_encodings = tokenizer(X_vtrain, truncation=True, padding=True)
    test_encodings = tokenizer(X_dtest, truncation=True, padding=True)
    
    st.write("tokenizing completed!")
    
    import tensorflow as tf
    
    train_dataset = tf.data.Dataset.from_tensor_slices((
        dict(train_encodings),
        y_dtrain
    ))
    
    val_dataset = tf.data.Dataset.from_tensor_slices((
        dict(val_encodings),
        y_vtrain
    ))
    
    test_dataset = tf.data.Dataset.from_tensor_slices((
        dict(test_encodings),
        y_dtest
    ))
    
    st.write("back to dataset!")
    
    from transformers import TFDistilBertForSequenceClassification, TFTrainer, TFTrainingArguments
    
    training_args = TFTrainingArguments(
        output_dir='./results',          
        num_train_epochs=2,              
        per_device_train_batch_size=16, 
        per_device_eval_batch_size=16,   
        warmup_steps=500,
        eval_steps=500, 
        weight_decay=0.01               
    )
    
    
    with training_args.strategy.scope():
        model = TFDistilBertForSequenceClassification.from_pretrained("distilbert-base-uncased")
    
    trainer = TFTrainer(
        model=model,                         
        args=training_args,                  
        train_dataset=train_dataset,         
        eval_dataset=val_dataset             
    )

    trainer.train()

    st.write("training completed")
    return trainer
    

dataset_dict = load_data()
trainer = training_computation(dataset_dict)


patents = pd.DataFrame(dataset_dict['train'])
patent_selection = st.selectbox("Select Patent",patents['patent_number'])

patent = patents.loc[patents['patent_number'] == patent_selection]
st.write(patent['abstract'])
st.write(patent['claims'])