Spaces:
Sleeping
Sleeping
| import os | |
| import zipfile | |
| import streamlit as st | |
| import nltk | |
| from nltk.corpus import stopwords | |
| from sklearn.feature_extraction.text import CountVectorizer | |
| import plotly.express as px | |
| nltk.download('punkt') | |
| nltk.download('stopwords') | |
| def preprocess_text(text): | |
| # Tokenize the text and remove stopwords | |
| tokens = nltk.word_tokenize(text.lower()) | |
| stop_words = set(stopwords.words('english')) | |
| filtered_tokens = [token for token in tokens if token not in stop_words] | |
| return filtered_tokens | |
| def get_context_files(prompt, md_files): | |
| # Preprocess the prompt and context files | |
| processed_prompt = preprocess_text(prompt) | |
| processed_files = {} | |
| for file in md_files: | |
| with open(file, 'r') as f: | |
| content = f.read() | |
| processed_files[file] = preprocess_text(content) | |
| # Calculate word matches and LCS bonus | |
| file_matches = {} | |
| for file, tokens in processed_files.items(): | |
| single_matches = set(tokens) & set(processed_prompt) | |
| double_matches = set(nltk.bigrams(tokens)) & set(nltk.bigrams(processed_prompt)) | |
| triple_matches = set(nltk.trigrams(tokens)) & set(nltk.trigrams(processed_prompt)) | |
| match_count = len(single_matches) + len(double_matches) * 4 + len(triple_matches) * 9 | |
| file_matches[file] = { | |
| 'single_matches': single_matches, | |
| 'double_matches': double_matches, | |
| 'triple_matches': triple_matches, | |
| 'match_count': match_count | |
| } | |
| # Sort the files by the match count | |
| sorted_files = sorted(file_matches.items(), key=lambda x: x[1]['match_count'], reverse=True) | |
| # Create a markdown outline with match counts and word matches | |
| outline = "## Outline\n" | |
| for file, matches in sorted_files: | |
| outline += f"- {file}: {matches['match_count']} matches\n" | |
| if matches['single_matches']: | |
| outline += f" - Single word matches: {', '.join(matches['single_matches'])}\n" | |
| if matches['double_matches']: | |
| outline += f" - Double word matches: {', '.join(' '.join(pair) for pair in matches['double_matches'])}\n" | |
| if matches['triple_matches']: | |
| outline += f" - Triple word matches: {', '.join(' '.join(trio) for trio in matches['triple_matches'])}\n" | |
| # Create a single prompt by concatenating the original prompt and the content of the top ten files | |
| context_prompt = prompt | |
| for file, _ in sorted_files[:10]: | |
| with open(file, 'r') as f: | |
| content = f.read() | |
| # Highlight the matching words in bold | |
| for word in file_matches[file]['single_matches']: | |
| content = content.replace(word, f"**{word}**") | |
| for pair in file_matches[file]['double_matches']: | |
| content = content.replace(' '.join(pair), f"**{' '.join(pair)}**") | |
| for trio in file_matches[file]['triple_matches']: | |
| content = content.replace(' '.join(trio), f"**{' '.join(trio)}**") | |
| context_prompt += '\n\n' + content | |
| # Create a plotly graph showing the match counts for the top ten files | |
| fig = px.bar(x=[file for file, _ in sorted_files[:10]], y=[matches['match_count'] for _, matches in sorted_files[:10]]) | |
| fig.update_layout(xaxis_title='File', yaxis_title='Match Count') | |
| st.plotly_chart(fig) | |
| return outline, context_prompt | |
| # Streamlit app | |
| def main(): | |
| st.title("Context-Aware Prompt Evaluation") | |
| # File upload | |
| uploaded_file = st.file_uploader("Upload a zip file with .md files", type="zip") | |
| if uploaded_file is not None: | |
| # Unzip the uploaded file | |
| with zipfile.ZipFile(uploaded_file, 'r') as zip_ref: | |
| zip_ref.extractall('uploaded_files') | |
| # Get the list of .md files from the uploaded directory | |
| md_files = [os.path.join('uploaded_files', file) for file in os.listdir('uploaded_files') if file.endswith('.md')] | |
| # Show the list of files | |
| st.subheader("Uploaded Files") | |
| for file in md_files: | |
| st.write(file) | |
| # Prompt input | |
| prompt = st.session_state.get('prompt', 'What are the main use cases of generative AI in healthcare that are currently unsolved?') | |
| prompt = st.text_area("Enter your prompt", value=prompt, key='prompt') | |
| # Evaluate the files for the prompt | |
| if st.button("Evaluate"): | |
| outline, context_prompt = get_context_files(prompt, md_files) | |
| st.subheader("Outline") | |
| st.markdown(outline) | |
| st.subheader("Context Prompt") | |
| st.markdown(context_prompt) | |
| if __name__ == '__main__': | |
| main() |