Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,107 +1,21 @@
|
|
| 1 |
-
|
| 2 |
-
pretty_name: semantaai-crypto_assets
|
| 3 |
-
license: mit
|
| 4 |
-
task_categories:
|
| 5 |
-
- time-series-forecasting
|
| 6 |
-
language:
|
| 7 |
-
- en
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
-
|
| 14 |
-
-
|
| 15 |
-
-
|
| 16 |
-
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
## Current Status
|
| 23 |
-
|
| 24 |
-
- Latest partitioned update UTC: `2026-07-03T20:36:45.579155+00:00`
|
| 25 |
-
- Partition window: `2026-07-01T00:00:00+00:00` to latest closed candle at update time
|
| 26 |
-
- Canonical routine update standard: `partitioned_incremental`
|
| 27 |
-
- Raw partition files: `495`
|
| 28 |
-
- Gold partition files: `495`
|
| 29 |
-
- Partitioned rows in latest update: `194800`
|
| 30 |
-
- Partitioned raw/gold parity: `true`
|
| 31 |
-
- Partitioned duplicate timestamps: `0`
|
| 32 |
-
- Partitioned OHLC violations: `0`
|
| 33 |
-
- Partitioned missing candles inside written partitions: `0`
|
| 34 |
-
|
| 35 |
-
## Data Layers
|
| 36 |
-
|
| 37 |
-
Canonical incremental layers:
|
| 38 |
-
|
| 39 |
-
- `eml/raw/ohlcv_partitioned/`: exchange-native cleaned OHLCV partitions by exchange, market type, timeframe, symbol, year, and month.
|
| 40 |
-
- `eml/gold/ohlcv_partitioned/`: normalized EML-ready OHLCV partitions with UTC datetime, exchange, market type, symbol, base/quote, timeframe, closed-candle flag, and quality flags.
|
| 41 |
-
|
| 42 |
-
Legacy full snapshot layers, kept for backward compatibility:
|
| 43 |
-
|
| 44 |
-
- `eml/raw/ohlcv/`
|
| 45 |
-
- `eml/gold/ohlcv/`
|
| 46 |
-
|
| 47 |
-
Execution/research layers:
|
| 48 |
-
|
| 49 |
-
- `eml/raw/book_ticker/`
|
| 50 |
-
- `eml/gold/book_ticker/`
|
| 51 |
-
- `eml/raw/orderbook_l2/`
|
| 52 |
-
- `eml/gold/orderbook_l2_aggregated/`
|
| 53 |
-
- `eml/gold/tradeflow/`
|
| 54 |
-
- `eml/gold/derivatives/`
|
| 55 |
-
- `eml/gold/arbitrage/`
|
| 56 |
-
- `eml/metadata/`
|
| 57 |
-
- `eml/reports/`
|
| 58 |
-
|
| 59 |
-
## Coverage
|
| 60 |
-
|
| 61 |
-
Primary exchanges:
|
| 62 |
-
|
| 63 |
-
- Binance
|
| 64 |
-
- Bybit
|
| 65 |
-
|
| 66 |
-
Markets:
|
| 67 |
-
|
| 68 |
-
- spot
|
| 69 |
-
- USDT perpetual futures where available
|
| 70 |
-
|
| 71 |
-
Required timeframes:
|
| 72 |
-
|
| 73 |
-
- `1m` for top-5 assets
|
| 74 |
-
- `5m`, `15m`, `1h`, `4h`, `1d` for top assets and available triangle/cross pairs
|
| 75 |
-
|
| 76 |
-
## Important Reports
|
| 77 |
-
|
| 78 |
-
- `eml/reports/final_audit_report.json`
|
| 79 |
-
- `eml/reports/dataset_manifest.json`
|
| 80 |
-
- `eml/reports/partitioned_incremental_update_report.json`
|
| 81 |
-
- `eml/reports/data_quality_report.json`
|
| 82 |
-
- `eml/reports/arbitrage_coverage_report.json`
|
| 83 |
-
- `eml/reports/arbitrage_execution_readiness_report.json`
|
| 84 |
-
- `eml/UPDATE_STRATEGY.md`
|
| 85 |
-
|
| 86 |
-
## EML Readiness
|
| 87 |
-
|
| 88 |
-
The latest `final_audit_report.json` marks:
|
| 89 |
-
|
| 90 |
-
- EML walk-forward ready: `true`
|
| 91 |
-
- Research arbitrage ready: `true`
|
| 92 |
-
- Execution-grade arbitrage ready: `false`
|
| 93 |
-
|
| 94 |
-
Execution-grade arbitrage remains marked false because true execution-grade claims require deeper historical bid/ask, L2 depth, fees, constraints, slippage, latency assumptions, and executable size validation. Current arbitrage outputs are research-only unless explicitly marked executable.
|
| 95 |
-
|
| 96 |
-
## Leakage Policy
|
| 97 |
-
|
| 98 |
-
- Feature/gold layers use only past and present information.
|
| 99 |
-
- Future-derived labels must stay in label-specific layers.
|
| 100 |
-
- No incomplete higher-timeframe candles should be marked as closed.
|
| 101 |
-
- Timestamps are UTC.
|
| 102 |
-
|
| 103 |
-
## Update Policy
|
| 104 |
-
|
| 105 |
-
Routine updates must write only current partitions under `eml/raw/ohlcv_partitioned` and `eml/gold/ohlcv_partitioned`. Do not download and rewrite full historical parquet files for routine updates.
|
| 106 |
-
|
| 107 |
-
See `eml/UPDATE_STRATEGY.md` for the canonical update method.
|
|
|
|
| 1 |
+
---
|
| 2 |
+
pretty_name: semantaai-crypto_assets
|
| 3 |
+
license: mit
|
| 4 |
+
task_categories:
|
| 5 |
+
- time-series-forecasting
|
| 6 |
+
language:
|
| 7 |
+
- en
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# semantaai-crypto_assets
|
| 11 |
+
|
| 12 |
+
Semanta AI market dataset in unified `raw` + `gold` + `labels` standard.
|
| 13 |
+
- updated at UTC: `2026-07-22T22:01:35.864528+00:00`
|
| 14 |
+
- target end UTC: `2026-07-22T20:20:00+00:00`
|
| 15 |
+
- raw rows: `4284625`
|
| 16 |
+
- gold rows: `6174008`
|
| 17 |
+
|
| 18 |
+
Layers:
|
| 19 |
+
- `raw/`: canonical 5m exchange/vendor bars with quality flags.
|
| 20 |
+
- `gold/`: cleaned, synchronized, TA/regime/context features. Past/present only.
|
| 21 |
+
- `labels/`: future-derived path-dependent labels. Do not join into features before train/test split design.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|