Upload folder using huggingface_hub
Browse files- README.md +56 -0
- hmm/param_FHMM1.txt +0 -0
- hmm/param_FHMM2.txt +0 -0
- hmm/param_FHMM3.txt +0 -0
- ppo/conv_jacobs.pt +3 -0
- ppo/conv_mixed_hmm.pt +3 -0
- ppo/decoder_hmm_nll.pt +3 -0
- ppo/simple_hmm_nll.pt +3 -0
- ppo/transformer_hmm_nll.pt +3 -0
- weights/parncutt.json +14 -0
README.md
ADDED
|
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: agpl-3.0
|
| 3 |
+
tags:
|
| 4 |
+
- music
|
| 5 |
+
- piano-fingering
|
| 6 |
+
- reinforcement-learning
|
| 7 |
+
- hidden-markov-model
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# Chirality — trained piano-fingering models
|
| 11 |
+
|
| 12 |
+
Model artifacts for [vibetuned/counterpoint](https://github.com/vibetuned/counterpoint)
|
| 13 |
+
(`chirality` on PyPI), a piano-fingering pipeline combining
|
| 14 |
+
ergonomic rule sets, statistical models, and reinforcement learning.
|
| 15 |
+
|
| 16 |
+
## Contents
|
| 17 |
+
|
| 18 |
+
| Path | What | Human match (M_gen) |
|
| 19 |
+
| :--- | :--- | :--- |
|
| 20 |
+
| `weights/parncutt.json` | Path-difference learned Parncutt rule weights (Radisavljevic & Driessen 2004, trained on PIG) | **65.4** — best non-human system |
|
| 21 |
+
| `hmm/param_FHMM{1,2,3}.txt` | Nakamura et al. (2020) fingering HMMs, orders 1–3, trained on the PIG miscellaneous subset | 61.6 / 64.3 / 64.4 |
|
| 22 |
+
| `ppo/conv_jacobs.pt` | PPO conv policy, jacobs reward | 44.8 (right hand) |
|
| 23 |
+
| `ppo/decoder_hmm_nll.pt` | PPO decoder policy, HMM-NLL reward | 43.2 (right hand) |
|
| 24 |
+
| `ppo/transformer_hmm_nll.pt` | PPO transformer policy, HMM-NLL reward | 42.2 (right hand) |
|
| 25 |
+
| `ppo/simple_hmm_nll.pt` | PPO MLP policy, HMM-NLL reward | 39.9 (right hand) |
|
| 26 |
+
| `ppo/conv_mixed_hmm.pt` | PPO conv policy, mixed rules+HMM reward | 39.5 (right hand) |
|
| 27 |
+
|
| 28 |
+
Human inter-annotator agreement on the same benchmark is 71.4.
|
| 29 |
+
Evaluation protocol and full tables: `docs/experiments_summary.md` in
|
| 30 |
+
the code repository.
|
| 31 |
+
|
| 32 |
+
## Usage
|
| 33 |
+
|
| 34 |
+
Drop the HMM parameters and learned weights where the CLI looks for
|
| 35 |
+
them, then annotate scores:
|
| 36 |
+
|
| 37 |
+
```bash
|
| 38 |
+
pip install chirality
|
| 39 |
+
mkdir -p data/hmm_params data/learned_weights
|
| 40 |
+
# copy hmm/param_FHMM*.txt -> data/hmm_params/
|
| 41 |
+
# copy weights/parncutt.json -> data/learned_weights/parncutt.json
|
| 42 |
+
chirality annotate --input score.mei --output ./annotated --weights --both
|
| 43 |
+
chirality annotate --input score.mei --output ./annotated --hmm --order 2
|
| 44 |
+
```
|
| 45 |
+
|
| 46 |
+
PPO checkpoints load with `chirality test --checkpoint ppo/<name>.pt
|
| 47 |
+
--policy <arch>` or `chirality annotate --model ...`.
|
| 48 |
+
|
| 49 |
+
## Provenance and licensing
|
| 50 |
+
|
| 51 |
+
Code is AGPL-3.0-or-later. The HMM parameters and learned rule weights
|
| 52 |
+
are statistics estimated from the [PIG dataset](https://beam.kisarazu.ac.jp/research/PianoFingeringDataset/)
|
| 53 |
+
(Nakamura, Saito & Yoshii 2020), which is available for research and
|
| 54 |
+
non-profit use — treat these artifacts under the same terms. The PPO
|
| 55 |
+
checkpoints were trained purely on generated scores (scales, chords,
|
| 56 |
+
arpeggios) and carry no PIG data.
|
hmm/param_FHMM1.txt
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
hmm/param_FHMM2.txt
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
hmm/param_FHMM3.txt
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ppo/conv_jacobs.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5a60ad2bde5a6066eabeab23bfd89f9e7cac414fe6b24dcd00af42ec7e9aed31
|
| 3 |
+
size 12982261
|
ppo/conv_mixed_hmm.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:159c0b949b12d5d91603607e9c1e0c05a3f1d66c2a38e915ede83826ce6d991f
|
| 3 |
+
size 12982261
|
ppo/decoder_hmm_nll.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e5b4cc8070da0484fec327e3cffd7b83801cc3db59b504ed81886ce2d6447715
|
| 3 |
+
size 13700605
|
ppo/simple_hmm_nll.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e43a7ae1425775a12bff5c9185a23fa6e88126fd218921a0ee5626fd847fa0e8
|
| 3 |
+
size 18189493
|
ppo/transformer_hmm_nll.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f4de1f6b183facaa25fba047badc12225de40559253e1b0b6e54743592c2b2ae
|
| 3 |
+
size 10480211
|
weights/parncutt.json
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"str": 0.6003,
|
| 3 |
+
"sma": 1.5466,
|
| 4 |
+
"lar": 1.0416,
|
| 5 |
+
"pcc": 0.6321,
|
| 6 |
+
"pcs": 0.637,
|
| 7 |
+
"wea": 0.1328,
|
| 8 |
+
"345": 0.3534,
|
| 9 |
+
"3t4": 0.1013,
|
| 10 |
+
"bl4": 0.587,
|
| 11 |
+
"bl1": 0.607,
|
| 12 |
+
"bl5": 0.3232,
|
| 13 |
+
"pa1": 4.0985
|
| 14 |
+
}
|