usamaahmedsh/synthetic-elliott-waves

---
license: mit
language:
- en
tags:
- elliott-wave
- synthetic
- finance
- time-series
- wave-pattern
- technical-analysis
size_categories:
- 100K<n<1M
---

# Synthetic Elliott Wave Dataset

High-quality synthetic Elliott Wave patterns generated via rejection sampling
with Fibonacci-ratio constraints and multivariate-normal density scoring.

## Dataset Summary

| Field | Value |
|---|---|
| Total rows | 15,000,779 |
| Real patterns | 779 |
| Synthetic patterns | 15,000,000 |
| Wave types | Corrective, bearish_impulse, impulse |
| All label=1 (good) | Yes — all samples passed all 4 tiers |
| Run | `run_20260318_135833_t5000000` |

## Wave Type Counts

| rule | is_synthetic | count |
|---|---|---|
| Corrective | False | 729 |
| Corrective | True | 5,000,000 |
| bearish_impulse | False | 11 |
| bearish_impulse | True | 5,000,000 |
| impulse | False | 39 |
| impulse | True | 5,000,000 |

## Schema

| Column | Description |
|---|---|
| `rule` | Wave type: `Corrective`, `impulse`, `bearish_impulse` |
| `wave_config` | Wave config string |
| `geo_0..geo_4` | Geometry ratios (wave retracements / extensions) |
| `ensemble_score` | MVN-based quality score (≥0.80 for all synthetic rows) |
| `fib_score` | Fibonacci proximity score |
| `is_synthetic` | `True` for generated, `False` for real market patterns |
| `label` | `1` for all rows in this dataset (good samples only) |

## Generation

- **Tier 1**: Structural EW geometry constraints (wave ordering, direction)
- **Tier 1b**: Monotonicity / non-overlap checks
- **Tier 2**: Fibonacci ratio gates (tolerance per wave type)
- **Tier 3**: Multivariate-normal density score ≥ 0.80 vs real-data distribution
- **Tier 4**: BallTree diversity filter (min distance from existing samples)

## Related Dataset

For scorer training (good + bad samples): [`usamaahmedsh/elliott-wave-scorer-training`](https://huggingface.co/datasets/usamaahmedsh/elliott-wave-scorer-training)