Model checkpoints only

README.md

@@ -1,77 +0,0 @@
----
-license: mit
-library_name: pytorch
-tags:
-- time-series
-- forecasting
-- lstm
-- multi-task
-- finance
-- weather
-- multi-domain
-language:
-- en
-metrics:
-- mae
-- mape
-pipeline_tag: time-series-forecasting
----
-
-# Future Prediction Models (Multi-Domain LSTM)
-
-An end-to-end multi-domain forecasting system across **7 topics** (AI, Programming, Finance, Sports, Weather, Economy, Energy). Two model modes ship in this repo:
-
-## 🧠 Unified model — one checkpoint for everything
-
-`unified_model.pt` is a **single multi-task model** trained jointly on 31,735 sliding windows from all 7 domains: shared 2-layer LSTM backbone (128 hidden) + per-domain embedding + per-domain heads.
-
-| Topic | H1 MAPE (unified) | H1 MAPE (separate) |
-|---|---|---|
-| AI (NVDA) | 2.36% | 1.86% |
-| Economy (S&P 500) | 0.75% | 0.64% |
-| Energy (WTI) | 1.90% | 2.83% |
-| Finance (Bitcoin) | 1.74% | 1.52% |
-| Programming (react) | 5.42% | 7.12% |
-| Sports (ATP #1 Elo) | 0.04% | 0.06% |
-| Weather (Chennai) | 0.18% | 0.18% |
-
-## Quick start
-
-```bash
-pip install -r requirements.txt
-python data.py --topic finance --refresh      # fetch real data
-python train_unified.py                       # train the single combined model
-python ask.py "what will bitcoin do next week?"
-python predict.py --topic finance --mode unified
-python ask.py "compare all topics"
-```
-
-## Files
-
-- `unified_model.pt` / `unified_config.json` — the single combined model + per-domain metrics
-- `model_<topic>.pt` / `config_<topic>.json` — 7 dedicated per-topic LSTMs (alternative mode)
-- `data/*.csv` — the real datasets used for training (BTC, NVDA, S&P 500, WTI, npm downloads, ATP tennis, Chennai weather)
-- Full source: data pipeline (`data.py`), training (`train.py`, `train_unified.py`), inference (`predict.py`), and the ChatGPT-style chat layer (`ask.py`, `response.py`, `response_formatter.py`, `prediction_engine.py`)
-
-## Architecture
-
-```
-user question -> intent detection (response.py)
-              -> prediction_engine.py   (structured metrics, prints nothing)
-              -> response_formatter.py  (natural summary + Markdown)
-              -> ask.py                 (chat CLI, stdout)
-```
-
-Metrics are kept strictly separate: `net_change_pct` (path change), `forecast_range_pct` (path min-max range), `current_to_forecast_pct` (vs latest observed value), `direction` (configurable threshold). No confidence is fabricated; markets are near random-walk, and 100% accuracy is impossible.
-
-## Weather for any city
-
-```bash
-export WEATHER_NAME='Chennai, India' WEATHER_LAT=13.0827 WEATHER_LON=80.2707
-python data.py --topic weather --refresh
-python train.py --topic weather
-```
-
-## License
-
-MIT