How vinamra3215/prism works
PRISM is a research codebase for adapting the Kronos financial time-series foundation model to Indian equities and evaluating cross-sectional stock rankings. Its documented workflow is NSE OHLCV preparation, parameter-efficient fine-tuning, market-conditioned representations, then walk-forward ranking backtests. The tree also contains multiple experiment branches and plotting utilities, rather than a single evident production service. The README documents the market-gating and ranking stages, but the sampled excerpts do not show their implementation; those links are therefore documented at a high level. Other relationships are limited to visible imports or documented capabilities.
Main entry points: predict_lora_ka_loda.py Stock Forecast · download_data.py Data Download · full_pipeline_tune_predict_full_param.py NIFTY Data Prep · lora_ka_loda.py LoRA Fine-tuning
Components 15
Market Data
- Data Download
Kronos/download_data.py - Feature Selection
__shutupandbendover/het-uchiha/phase_2/features_selection/features.py - NIFTY Data Prep
__shutupandbendover/devansh-swc-wala/Block1/optuna_finetuned/full_pipeline_tune_predict_full_param.py
Model Adaptation
- Kronos Model
Kronos/model/kronos.py - Model Layers
Kronos/model/module.py - LoRA Fine-tuning
Kronos/het/lora_ka_loda.py - Rolling Adaptation
__shutupandbendover/devansh-swc-wala/Block1/optuna_finetuned/nifty50_multistock_individual_lora_pipeline.py - Market Gating
Kronos
Ranking Strategy
- Stock Forecast
Kronos/het/predict_lora_ka_loda.py - Cross-Sectional Rank
Kronos
Evaluation
- Walk-Forward Backtest
__shutupandbendover/devansh-swc-wala/Block1/my_kronos_project/train_and_predict_walk_forward.py - Forecast Metrics
__shutupandbendover/devansh-swc-wala/Block1/optuna_finetuned/metrics_engine.py - Result Visualization
Kronos/visualise_result.py
Other components
- Researcher
- NSE OHLCV
Connections 15
- Researcher → Data Download: initiates research
- NSE OHLCV → Data Download: provides prices
- Data Download → NIFTY Data Prep: prepares data
- NSE OHLCV → NIFTY Data Prep: provides OHLCV
__shutupandbendover/devansh-swc-wala/Block1/optuna_finetuned/full_pipeline_tune_predict_full_param.py - NIFTY Data Prep → LoRA Fine-tuning: trains on data
- Feature Selection → Market Gating: selects signals
- Kronos Model → Model Layers: uses layers
Kronos/model/kronos.py - Kronos Model → LoRA Fine-tuning: adapts backbone
Kronos/het/lora_ka_loda.py - LoRA Fine-tuning → Rolling Adaptation: provides adaptation
- Rolling Adaptation → Stock Forecast: generates forecasts
- Market Gating → Cross-Sectional Rank: conditions ranking
- Stock Forecast → Cross-Sectional Rank: ranks stocks
Kronos/het/predict_lora_ka_loda.py - Cross-Sectional Rank → Walk-Forward Backtest: evaluates strategy
__shutupandbendover/devansh-swc-wala/Block1/my_kronos_project/train_and_predict_walk_forward.py - Walk-Forward Backtest → Forecast Metrics: reports performance
- Stock Forecast → Result Visualization: plots results