akashkumar5/Multi_Timeframe_Market_Regimes_HMM6_BTCUSD

Multi‑Timeframe Market Regimes (HMM‑6)

Dataset Summary

This dataset provides labeled crypto market regimes derived from multi-timeframe (5m, 15m) OHLCV data and technical indicators. Market regimes are inferred using a 6-state Hidden Markov Model (HMM).

The dataset is suitable for market regime detection, regime-aware modeling, and sequence modeling baselines (LSTM / Transformers).

No future leakage: all indicators are computed using information available up to the timestamp, and labels correspond to the inferred regime at that timestamp only.

Source & Licensing

• Asset: BTC (BTCUSD / BTCUSDT – specify exact symbol)
• Exchange: specify exchange (e.g., Binance)
• Market: spot or perpetual futures
• Timezone: UTC

Files

.
├── README.md
└── data/
    ├── train.csv         # oldest part
    ├── validation.csv    # middle part
    └── test.csv          # most recent data

Optionally provide Parquet equivalents:

└── data/
    ├── train.parquet
    ├── validation.parquet
    └── test.parquet

All splits are chronological. No shuffling is performed.

Columns

Core

• timestamp — UTC timestamp (ISO8601 or UNIX ms)

OHLCV (contextual)

• open_5m, high_5m, low_5m, close_5m, volume_5m
• open_15m, high_15m, low_15m, close_15m, volume_15m

OHLCV columns are included for context and visualization and are not required for training the HMM.

---

Technical Indicators (5m)

• log_ret_1_5m
• ema_ratio_9_21_5m
• macd_hist_5m
• adx_5m
• atr_norm_5m
• bb_width_5m
• rsi_14_5m
• volume_zscore_50_5m

Technical Indicators (15m)

• log_ret_1_15m
• ema_ratio_9_21_15m
• macd_hist_15m
• adx_15m
• atr_norm_15m
• bb_width_15m
• rsi_14_15m
• volume_zscore_50_15m

---

Targets

• state — integer HMM state label (0–5)
• regime — human-readable semantic label mapped from state

Regime Semantics

State	Regime
0	Choppy High-Vol
1	Range
2	Squeeze
3	Strong Trend
4	Volatility Spike
5	Weak Trend

Notes

• state IDs are model-dependent and arbitrary
• regime labels are post-hoc human interpretations
• For modeling, use state as the target

---

Columns

Core

• timestamp — UTC timestamp (ISO8601 or UNIX ms)

OHLCV (contextual)

• open_5m, high_5m, low_5m, close_5m, volume_5m
• open_15m, high_15m, low_15m, close_15m, volume_15m

OHLCV columns are included for context and visualization and are not required for training the HMM.

---

Technical Indicators (5m)

• log_ret_1_5m
• ema_ratio_9_21_5m
• macd_hist_5m
• adx_5m
• atr_norm_5m
• bb_width_5m
• rsi_14_5m
• volume_zscore_50_5m

Technical Indicators (15m)

• log_ret_1_15m
• ema_ratio_9_21_15m
• macd_hist_15m
• adx_15m
• atr_norm_15m
• bb_width_15m
• rsi_14_15m
• volume_zscore_50_15m

---

Targets

• state — integer HMM state label (0–5)
• regime — human-readable semantic label mapped from state

Regime Semantics

State	Regime
0	Choppy High-Vol
1	Range
2	Squeeze
3	Strong Trend
4	Volatility Spike
5	Weak Trend

Notes

• state IDs are model-dependent and arbitrary
• regime labels are post-hoc human interpretations
• For modeling, use state as the target

---

Splitting Protocol (Leak-Free)

1. Sort data by timestamp (ascending)
2. Split into contiguous blocks:
   • train → validation → test
3. Do not shuffle
4. Sliding windows for sequence models must be constructed within each split only

---

How to Load

from datasets import load_dataset

ds = load_dataset("akashkumar5/Multi_Timeframe_Market_Regimes_HMM6_BTCUSD")
train = ds["train"]
valid = ds["validation"]
test = ds["test"]

Known Limitations

• Technical‑indicator engineered features; not raw trades.
• Single asset (if BTCUSDT only) may not generalize to alts/FX.
• Regimes are model‑dependent; HMM re‑fits may shift boundaries.

Changelog

• v1.0.0 — Initial release.