ML-LABEL-002: Empirical Sample Uniqueness & Concurrency Study
Executive Summary
Evaluation of Marcos Lopez de Prado's Sample Uniqueness Weighting algorithm on Gold M1 bars.
- Dataset Size: 10,000 bars
- Evaluated Samples: 2,856 samples
- Mean Sample Uniqueness ($u_i$): 0.8988
- Kish Effective Sample Size ($N_{eff}$): 2,765.2
- Sample Redundancy Ratio: 3.18%
- Max Concurrency ($c_t$): 4 concurrent active labels
- Mean Concurrency ($c_t$): 1.14
Quantiles of Sample Uniqueness
| Quantile | Value |
|---|---|
| p05 | 0.5000 |
| p10 | 0.6458 |
| p25 | 0.8333 |
| Median (p50) | 1.0000 |
| p75 | 1.0000 |
| p90 | 1.0000 |
| p95 | 1.0000 |
Empirical Distributions
--- Raw Sample Uniqueness (u_i) (N=2,856) ---
[0.31 - 0.38]: 16 ( 0.6%)
[0.38 - 0.45]: 25 ( 0.9%)
[0.45 - 0.52]: ### 200 ( 7.0%)
[0.52 - 0.59]: 7 ( 0.2%)
[0.59 - 0.66]: 43 ( 1.5%)
[0.66 - 0.73]: # 74 ( 2.6%)
[0.73 - 0.79]: ### 206 ( 7.2%)
[0.79 - 0.86]: ### 221 ( 7.7%)
[0.86 - 0.93]: ### 229 ( 8.0%)
[0.93 - 1.00]: ############################## 1,835 ( 64.3%)--- Concurrency (c_t) (N=7,707) ---
[1.00 - 1.30]: ############################## 6,707 ( 87.0%)
[1.30 - 1.60]: 0 ( 0.0%)
[1.60 - 1.90]: 0 ( 0.0%)
[1.90 - 2.20]: #### 938 ( 12.2%)
[2.20 - 2.50]: 0 ( 0.0%)
[2.50 - 2.80]: 0 ( 0.0%)
[2.80 - 3.10]: 61 ( 0.8%)
[3.10 - 3.40]: 0 ( 0.0%)
[3.40 - 3.70]: 0 ( 0.0%)
[3.70 - 4.00]: 1 ( 0.0%)