clustermap app

This commit is contained in:
2025-08-11 01:59:48 +01:00
parent 80c115b57d
commit 248cc5765f
3 changed files with 296 additions and 0 deletions

View File

@@ -0,0 +1,58 @@
# Discord Chat Embeddings Visualizer
A Streamlit application that visualizes Discord chat messages using their vector embeddings in 2D space.
## Features
- **2D Visualization**: View chat messages plotted using PCA or t-SNE dimension reduction
- **Interactive Plotting**: Hover over points to see message content, author, and timestamp
- **Filtering**: Filter by source chat log files and authors
- **Multiple Datasets**: Automatically loads all CSV files from the discord_chat_logs folder
## Installation
1. Install the required dependencies:
```bash
pip install -r requirements.txt
```
## Usage
Run the Streamlit application:
```bash
streamlit run streamlit_app.py
```
The app will automatically load all CSV files from the `../../discord_chat_logs/` directory.
## Data Format
The application expects CSV files with the following columns:
- `message_id`: Unique identifier for the message
- `timestamp_utc`: When the message was sent
- `author_id`: Author's Discord ID
- `author_name`: Author's username
- `author_nickname`: Author's server nickname
- `content`: The message content
- `attachment_urls`: Any attached files
- `embeds`: Embedded content
- `content_embedding`: Vector embedding of the message content (as a string representation of a list)
## Visualization Options
- **PCA**: Principal Component Analysis - faster, good for getting an overview
- **t-SNE**: t-Distributed Stochastic Neighbor Embedding - slower but may reveal better clusters
## Controls
- **Dimension Reduction Method**: Choose between PCA and t-SNE
- **Filter by Source Files**: Select which chat log files to include
- **Filter by Authors**: Select which authors to display
- **Show Data Table**: View the underlying data in table format
## Performance Notes
- For large datasets, consider filtering by authors or source files to improve performance
- t-SNE is computationally intensive and may take longer with large datasets
- The app caches data and computations for better performance

View File

@@ -0,0 +1,5 @@
streamlit>=1.28.0
pandas>=1.5.0
numpy>=1.24.0
plotly>=5.15.0
scikit-learn>=1.3.0

View File

@@ -0,0 +1,233 @@
import streamlit as st
import pandas as pd
import numpy as np
import plotly.express as px
import plotly.graph_objects as go
from sklearn.decomposition import PCA
from sklearn.manifold import TSNE
import json
import os
from pathlib import Path
import ast
# Set page config
st.set_page_config(
page_title="Discord Chat Embeddings Visualizer",
page_icon="🗨️",
layout="wide"
)
# Title and description
st.title("🗨️ Discord Chat Embeddings Visualizer")
st.markdown("Explore Discord chat messages through their vector embeddings in 2D space")
@st.cache_data
def load_all_chat_data():
"""Load all CSV files from the discord_chat_logs folder"""
chat_logs_path = Path("../../discord_chat_logs")
# Display the path for debugging
st.write(f"Looking for CSV files in: {chat_logs_path}")
st.write(f"Path exists: {chat_logs_path.exists()}")
all_data = []
for csv_file in chat_logs_path.glob("*.csv"):
try:
df = pd.read_csv(csv_file)
df['source_file'] = csv_file.stem # Add source file name
all_data.append(df)
st.write(f"✅ Loaded {len(df)} messages from {csv_file.name}")
except Exception as e:
st.error(f"❌ Error loading {csv_file.name}: {e}")
if all_data:
combined_df = pd.concat(all_data, ignore_index=True)
st.success(f"🎉 Successfully loaded {len(combined_df)} total messages from {len(all_data)} files")
return combined_df
else:
st.error("No data loaded!")
return pd.DataFrame()
@st.cache_data
def parse_embeddings(df):
"""Parse the content_embedding column from string to numpy array"""
embeddings = []
valid_indices = []
for idx, embedding_str in enumerate(df['content_embedding']):
try:
# Parse the string representation of the list
embedding = ast.literal_eval(embedding_str)
if isinstance(embedding, list) and len(embedding) > 0:
embeddings.append(embedding)
valid_indices.append(idx)
except Exception as e:
continue
embeddings_array = np.array(embeddings)
valid_df = df.iloc[valid_indices].copy()
st.info(f"📊 Parsed {len(embeddings)} valid embeddings from {len(df)} messages")
st.info(f"🔢 Embedding dimension: {embeddings_array.shape[1] if len(embeddings) > 0 else 0}")
return embeddings_array, valid_df
@st.cache_data
def reduce_dimensions(embeddings, method="PCA", n_components=2):
"""Reduce embeddings to 2D using PCA or t-SNE"""
if method == "PCA":
reducer = PCA(n_components=n_components, random_state=42)
elif method == "t-SNE":
reducer = TSNE(n_components=n_components, random_state=42, perplexity=min(30, len(embeddings)-1))
reduced_embeddings = reducer.fit_transform(embeddings)
return reduced_embeddings
def create_hover_text(df):
"""Create hover text for plotly"""
hover_text = []
for _, row in df.iterrows():
text = f"<b>Author:</b> {row['author_name']}<br>"
text += f"<b>Timestamp:</b> {row['timestamp_utc']}<br>"
text += f"<b>Source:</b> {row['source_file']}<br>"
# Handle potential NaN or non-string content
content = row['content']
if pd.isna(content) or content is None:
content_text = "[No content]"
else:
content_str = str(content)
content_text = content_str[:200] + ('...' if len(content_str) > 200 else '')
text += f"<b>Content:</b> {content_text}"
hover_text.append(text)
return hover_text
def main():
# Load data
with st.spinner("Loading chat data..."):
df = load_all_chat_data()
if df.empty:
st.stop()
# Parse embeddings
with st.spinner("Parsing embeddings..."):
embeddings, valid_df = parse_embeddings(df)
if len(embeddings) == 0:
st.error("No valid embeddings found!")
st.stop()
# Sidebar controls
st.sidebar.header("🎛️ Visualization Controls")
# Dimension reduction method
method = st.sidebar.selectbox(
"Dimension Reduction Method",
["PCA", "t-SNE"],
help="PCA is faster, t-SNE may reveal better clusters"
)
# Source file filter
source_files = valid_df['source_file'].unique()
selected_sources = st.sidebar.multiselect(
"Filter by Source Files",
source_files,
default=source_files,
help="Select which chat log files to include"
)
# Author filter
authors = valid_df['author_name'].unique()
selected_authors = st.sidebar.multiselect(
"Filter by Authors",
authors,
default=authors[:10] if len(authors) > 10 else authors, # Limit to first 10 for performance
help="Select which authors to include"
)
# Filter data
filtered_df = valid_df[
(valid_df['source_file'].isin(selected_sources)) &
(valid_df['author_name'].isin(selected_authors))
]
if filtered_df.empty:
st.warning("No data matches the current filters!")
st.stop()
# Get corresponding embeddings
filtered_indices = filtered_df.index.tolist()
filtered_embeddings = embeddings[[i for i, idx in enumerate(valid_df.index) if idx in filtered_indices]]
st.info(f"📈 Visualizing {len(filtered_df)} messages")
# Reduce dimensions
with st.spinner(f"Reducing dimensions using {method}..."):
reduced_embeddings = reduce_dimensions(filtered_embeddings, method)
# Create hover text
hover_text = create_hover_text(filtered_df)
# Create the plot
fig = go.Figure()
# Color by source file
colors = px.colors.qualitative.Set1
for i, source in enumerate(selected_sources):
source_mask = filtered_df['source_file'] == source
if source_mask.any():
source_data = filtered_df[source_mask]
source_embeddings = reduced_embeddings[source_mask]
source_hover = [hover_text[j] for j, mask in enumerate(source_mask) if mask]
fig.add_trace(go.Scatter(
x=source_embeddings[:, 0],
y=source_embeddings[:, 1],
mode='markers',
name=source,
marker=dict(
size=8,
color=colors[i % len(colors)],
opacity=0.7,
line=dict(width=1, color='white')
),
hovertemplate='%{hovertext}<extra></extra>',
hovertext=source_hover
))
fig.update_layout(
title=f"Discord Chat Messages - {method} Visualization",
xaxis_title=f"{method} Component 1",
yaxis_title=f"{method} Component 2",
hovermode='closest',
width=1000,
height=700
)
# Display the plot
st.plotly_chart(fig, use_container_width=True)
# Statistics
col1, col2, col3 = st.columns(3)
with col1:
st.metric("Total Messages", len(filtered_df))
with col2:
st.metric("Unique Authors", filtered_df['author_name'].nunique())
with col3:
st.metric("Source Files", len(selected_sources))
# Show data table
if st.checkbox("Show Data Table"):
st.subheader("📋 Message Data")
display_df = filtered_df[['timestamp_utc', 'author_name', 'source_file', 'content']].copy()
display_df['content'] = display_df['content'].str[:100] + '...' # Truncate for display
st.dataframe(display_df, use_container_width=True)
if __name__ == "__main__":
main()