How akshatmittu/machine-learning-model-reliability-analysis-at-production-level works
Project Type and Purpose
• This is a monolithic, full-stack Python application whose primary goal is to provide an interactive dashboard for evaluating and monitoring the reliability of production ML models (regression, classification, NLP, and CV). It leverages Python scripts for the UI (likely Streamlit), a “pages/models” directory as its data store (CSV/HDF5/Joblib), and utility functions for computing key metrics (performance drift, data drift, data quality, prediction drift, and explanations). The “Jupyter Notebooks” folder houses exploratory and offline analyses but is orthogonal to the core runtime.
High-Level Architecture and Components
a. User Interface Layer
• 1_Landing_Page.py
• pages/2_Analysis_Page.py
• pages/Add_Model.py
b. Business Logic & Computation
• utils.py (implements drift detection, metric calculations, explanation generation)
c. Model & Data Storage
• pages/models/… (per-model directories with baseline.csv, model artifacts, ground truths, production-run CSVs)
• pages/output.json, pages/models.json, pages/benchmark.json
• creds.csv (credentials for any remote data sources, if used)
d. Developer/Analyst Notebook Environment
• Jupyter Notebooks (CV.ipynb, Model Monitoring Dashboard.ipynb, NLP Data Quality, etc.)
e. Deployment/Runtime Environment
• Python runtime (per requirements.txt)
• Host (local or cloud VM/container)
Relationships & Data Flows
• User → (interacts with) UI Layer → (invokes) Business Logic
• Business Logic → (reads/writes) Model & Data Storage
• Business Logic → (computes) Metrics & Explanations → (returns) to UI Layer
• Notebook Environment ↔ (imports) Business Logic / raw data for offline experimentation
Key Architectural Patterns & Principles
• Monolithic, multi-page application (Streamlit-style)
• Separation of Concerns: UI (pages/) vs. logic (utils.py) vs. storage (pages/models/)
• File-based “database” for model artifacts and evaluation data—no heavy RDBMS
• Extensibility: New models drop into pages/models/ with the same directory pattern