Streamlit Application
Architecture
Interactive web application developed with Streamlit to present analyses from EDA notebooks.
Structure
Main page with sidebar navigation
4 independent analysis modules
Interactive widgets (sliders, selectbox, filters)
Data caching (TTL 1h) for performance
The 4 Analyses
1. Long-Term Trends Analysis
Source: 01_long_term/recipe_analysis_trendline.ipynb
Presents evolution from 1999-2018 of:
Interaction volume
Preparation time
Recipe complexity
Nutritional profiles
Visualizations: 6 synchronized temporal charts with linear regression.
2. Seasonality Analysis
Source: 02_seasonality/recipe_analysis_seasonality.ipynb
Identifies seasonal patterns:
Recipe distribution by season
Monthly nutritional variations
Seasonal activity peaks
Visualizations: Histograms, monthly heatmaps, thematic color palette.
3. Weekend Effect Analysis
Source: 03_week_end_effect/recipe_analysis_weekend.ipynb
Compares weekday vs weekend publications:
Volume by day of week
Impact on complexity/duration
Statistical tests (Chi-2)
Visualizations: 3 comparative panels with displayed p-values.
4. Ratings Analysis
Source: 01_long_term/rating_analysis.ipynb
Studies user rating distribution:
0-5 star distribution
Aggregated statistics
Positive bias detection
Visualizations: Interactive histograms, boxplots, satisfaction metrics.
Data Loading
Data is loaded from S3 at startup via:
DataLoader (error handling)
cached_loaders (Streamlit cache TTL 1h)
178K recipes + 1.1M ratings (~450 MB Parquet)
Performance
First load: 5-10 seconds (S3)
Subsequent loads: <0.1 second (cache)
DNAT optimization: 500+ MB/s