How suryatejareddy195/crypto-screener works
Crypto Predictor is a Streamlit dashboard backed by a calibrated next-candle direction model and a research pipeline. The principal workflow refreshes market and funding data, builds causal features, trains and evaluates models, then serves gated UP/DOWN probabilities and diagnostics in the dashboard. The README also describes live derivatives diagnostics and a rule-based baseline. Edges reflect explicit README documentation or the shown SOURCE INDEX references; unsampled implementation details are not inferred.
Main entry points: train_model.py Walk-forward trainer · app.py Streamlit dashboard · retrain.py Refresh and retrain · build_dataset.py OHLCV ingestion
Components 19
User experience
- Streamlit dashboard
app.py - Rule-based baseline
predict.py - Live market signal
predict.py
Market data
- OHLCV ingestion
build_dataset.py - Raw candle cache
build_dataset.py - Funding and live alt-data
altdata.py - Funding cache
altdata.py
Model pipeline
- Causal features and labels
features.py - Feature cache
features.py - Walk-forward trainer
train_model.py - Probability calibration
mlutils.py - Skill and trading evaluation
train_model.py - Calibrated gated inference
train_model.py
Model operations
- Trained model artifacts
models - Latest-model registry
models/manifest.json - Refresh and retrain
retrain.py
Other components
- Analyst
- Binance APIs
- CoinGecko API
Connections 25
- Analyst → Streamlit dashboard: views
README.md - Streamlit dashboard → Calibrated gated inference: requests probabilities
app.py - Streamlit dashboard → Funding and live alt-data: shows diagnostics
app.py - Streamlit dashboard → Rule-based baseline: uses baseline
app.py - Streamlit dashboard → Walk-forward trainer: shows diagnostics
app.py - Analyst → Rule-based baseline: runs baseline
README.md - OHLCV ingestion → Binance APIs: downloads candles
build_dataset.py - OHLCV ingestion → CoinGecko API: falls back to
build_dataset.py - OHLCV ingestion → Raw candle cache: caches candles
build_dataset.py - Funding and live alt-data → Binance APIs: fetches derivatives
altdata.py - Funding and live alt-data → Funding cache: caches funding
altdata.py - Causal features and labels → Raw candle cache: reads candles
README.md - Causal features and labels → Funding cache: joins funding
features.py - Causal features and labels → Feature cache: builds feature cache
README.md - Walk-forward trainer → Feature cache: trains from features
README.md - Walk-forward trainer → Probability calibration: uses calibration
train_model.py - Walk-forward trainer → Skill and trading evaluation: evaluates model
README.md - Walk-forward trainer → Trained model artifacts: saves models
README.md - Walk-forward trainer → Latest-model registry: updates registry
README.md - Calibrated gated inference → Trained model artifacts: loads model
train_model.py - Calibrated gated inference → Latest-model registry: resolves latest
train_model.py - Refresh and retrain → OHLCV ingestion: refreshes candles
retrain.py - Refresh and retrain → Funding and live alt-data: refreshes funding
retrain.py - Refresh and retrain → Causal features and labels: rebuilds features
retrain.py - Refresh and retrain → Walk-forward trainer: retrains models
retrain.py