# Features

Pereval integrates both attribute-scoped and deep feature metrics under one package. Metrics can be used for 1) assessing distributional discrepancies between two sets of MIDI piano performances and 2) asssigning perceptual pseudo-ratings to individual performances.

**Attribute-scoped metrics:**
- Inter-set correlation
- Intra-set correlation
- KL Divergence

**Deep feature metrics:**
- Fréchet Music Distance (FMD)
- Kernel Music Distance (KMD)
- Kernel Performance Distance (KPD)

**Per-sample pseudo ratings:**
- Mahalanobis Distance
- Relative Mahalanobis Distance
- Marginal Mahalanobis Distance

Note that correlation, FMD, KMD, and KPD can also be used for per-sample evaluation.

|                          | Fidelity | Diversity | Alignment-free | Score-aware | Contextual | Per-sample |
|--------------------------|:--------:|:---------:|:--------------:|:-----------:|:----------:|:----------:|
| Reconstruction error     | ✅       | ❌        | ❌             | ✅          | ❌         | ✅         |
| Inter-correlation        | ✅       | ❌        | ❌             | ✅          | ❌         | ✅         |
| Intra-correlation        | ❌       | ✅        | ❌             | ✅          | ❌         | ❌         |
| KL Divergence            | ✅       | ✅        | ✅             | ❌          | ❌         | ❌         |
| Fréchet Music Distance   | ✅       | ✅        | ✅             | ❌          | ✅         | ✅         |
| Kernel Music Distance    | ✅       | ✅        | ✅             | ❌          | ✅         | ✅         |
| Kernel Performance Distance | ✅    | ✅        | ✅             | ✅          | ✅         | ✅         |
| Mahalanobis Distance     | ✅       | ❌        | ✅             | ✅          | ✅         | ✅         |
| Relative Mahalanobis Distance | ✅  | ❌        | ✅             | ✅          | ✅         | ✅         |
| Marginal Mahalanobis Distance | ✅  | ❌        | ✅             | ❌          | ✅         | ✅         |

*Comparison of attribute-scoped and deep feature metrics for evaluating MIDI piano performances.*

Deep feature metrics are calculated using the embeddings from pretrained self-supervised symbolic music models ([Aria](https://github.com/EleutherAI/aria) and [CLaMP3](https://github.com/sanderwood/clamp3)). Pereval supports feature extraction from the Aria model. To integrate CLaMP3 embeddings, please install CLaMP3 separately.