10_nlp_python

import stanza
import pandas as pd
from transformers import pipeline
from sklearn.metrics import accuracy_score
import os

# Download the English model
stanza.download('en')
nlp = stanza.Pipeline('en')

# Initialize the pipeline for English
file_path = '~/Desktop/COIi work/State_Laws/cleaned_text_set.csv'
cleaned_bills = pd.read_csv(file_path)
print(cleaned_bills.head())

### DistilBERT ---- first trying
sentiment_analyzer = pipeline("sentiment-analysis", model="distilbert-base-uncased")
# Example text
text = "This immigration bill is a disaster for our community."

# Get sentiment
result = sentiment_analyzer(text)
print(result)

## now on bills ----
cleaned_bills['sentiment'] = cleaned_bills['Summary'].apply(lambda x: sentiment_analyzer(x)[0])  # Extract the first result

# Separate out the label and score into separate columns
cleaned_bills['sentiment_label'] = cleaned_bills['sentiment'].apply(lambda x: x['label'])
cleaned_bills['sentiment_score'] = cleaned_bills['sentiment'].apply(lambda x: x['score'])

# Save the results to a new CSV
output_file = os.path.expanduser('~/Desktop/COIi work/State_Laws/bills_with_sentiment.csv')
cleaned_bills.to_csv(output_file, index=False)


### Seeing accuracy score 
# Step 2: Filter out resolutions (if necessary)
resos = cleaned_bills[cleaned_bills['Summary'].str.contains(r'\bresolution\b', na=False)]
resos_IDs = resos['Bill_Number']

# Step 3: Create the cleaned_bills_only dataframe
cleaned_bills_only = cleaned_bills[~cleaned_bills['Bill_Number'].isin(resos_IDs)]

### Assigning classifications function ----
def categorize_sentiment_pol(row):
    score = row['sentiment_score']
    if score > 0.75:
        return 'P'
    elif score < 0.25:
        return 'A'
    else:
        return 'N'
def categorize_sentiment_symb(row):
    score = row['sentiment_score']
    if score > 0.75:
        return 'SP'
    elif score < 0.25:
        return 'SA'
    else:
        return 'N'

cleaned_bills_only['categorized_sentiment'] = cleaned_bills_only.apply(categorize_sentiment_symb, axis=1)
resos['categorized_sentiment'] = resos.apply(categorize_sentiment_pol, axis=1)


print(cleaned_bills_only[['Bill_Number', 'sentiment_label', 'sentiment_score', 'categorized_sentiment']].head())
# Exclude the first two columns using iloc
cleaned_bills_excluded = cleaned_bills_only.iloc[:, 2:]


full_set = pd.concat([cleaned_bills_excluded, resos], ignore_index=True)
print(full_set.head())


## Importing hand coded file
code_file = '~/Desktop/COIi work/State_Laws/complete_training.csv'
training = pd.read_csv(code_file)

train_IDs = training['Bill_Number']
train_pred = cleaned_bills_excluded[~cleaned_bills_excluded['Bill_Number'].isin(train_IDs)]

comparison_df = pd.merge(train_pred[['Bill_Number', 'categorized_sentiment']], 
                         training[['Bill_Number', 'Class']], 
                         on='Bill_Number', how='inner')

# Check if the predicted sentiment matches the actual sentiment
comparison_df['match'] = comparison_df['categorized_sentiment'] == comparison_df['Class']

# Calculate accuracy
accuracy = comparison_df['match'].mean() * 100
print(f"Accuracy: {accuracy:.2f}%")