augmentednet - v1.9.1-ablation

Special branch for ablation studies, which includes a 5-fold dataset_summary.tsv setup.

- Jupyter Notebook
Published by napulen almost 4 years ago

augmentednet - v1.9.1

• Fixed a bug where the annotations wouldn't generate properly with multiple-voice MusicXML files
• Fixed another bug with Cad64 not generated properly from the predictions
• Removed caching from music21 streams using forceSource=True which lead to unpredictable behavior when parsing input files

What's Changed

• Evaluation bugs by @napulen in https://github.com/napulen/AugmentedNet/pull/117

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.9.0...v1.9.1

- Jupyter Notebook
Published by napulen almost 4 years ago

augmentednet - v1.9.0

This release adds the keymodt (Key Modulation and Tonicization) dataset.

What's Changed

• Separating Bass19 into Bass12+Bass7 and same for Chromagram19. by @napulen in https://github.com/napulen/AugmentedNet/pull/100
• Running mlflow.end_run() at end of experiment. by @napulen in https://github.com/napulen/AugmentedNet/pull/102
• Removing mistakenly added results/summary files. by @napulen in https://github.com/napulen/AugmentedNet/pull/103
• Adding PhD Thesis Notebooks by @napulen in https://github.com/napulen/AugmentedNet/pull/104
• Introduce kmt dataset by @napulen in https://github.com/napulen/AugmentedNet/pull/108
• Notebook ch4 by @napulen in https://github.com/napulen/AugmentedNet/pull/110
• Back to rmsprop as optimizer by @napulen in https://github.com/napulen/AugmentedNet/pull/112
• Revert "Back to rmsprop as optimizer" by @napulen in https://github.com/napulen/AugmentedNet/pull/113
• Include pretrained model and version change for v1.9.0 by @napulen in https://github.com/napulen/AugmentedNet/pull/114

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.8.0...v1.9.0

- Jupyter Notebook
Published by napulen almost 4 years ago

augmentednet - v1.8.0

Several minor changes introduced. - Texturization patterns can now handle durations of a dotted half and dotted quarter notes. - Renaming Duration14 as MeasureNoteOnset14 - It was also separated into MeasureOnset7 and NoteOnset7 - haydnop20 renamed to haydnsun all throughout - TonicizedKey35 and LocalKey35 renamed to TonicizedKey38 and LocalKey38 - Removed padded timesteps from evaluation - Other minor corrections

What's Changed

• Adding Duration14 as a default input representation. by @napulen in https://github.com/napulen/AugmentedNet/pull/91
• Renaming LocalKey38 and TonicizedKey38. by @napulen in https://github.com/napulen/AugmentedNet/pull/92
• Renaming MeasureNoteOnset14 and introducing intermediate representations by @napulen in https://github.com/napulen/AugmentedNet/pull/93
• Adding more options for experiment names by @napulen in https://github.com/napulen/AugmentedNet/pull/94
• Reverse logic of transposition arg. by @napulen in https://github.com/napulen/AugmentedNet/pull/96
• Adding the fix to the existing texturization patterns for 3/4 and 6/8… by @napulen in https://github.com/napulen/AugmentedNet/pull/98
• V180 by @napulen in https://github.com/napulen/AugmentedNet/pull/97

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.7.1...v1.8.0

- Jupyter Notebook
Published by napulen almost 4 years ago

augmentednet - v1.7.1

• Patches an issue where percussion tracks would introduce nonsensical pitch information.
• Patches the RomanText output files, which were sometimes invalid.

What's Changed

• Pointing README to latest dataset automatically. by @napulen in https://github.com/napulen/AugmentedNet/pull/77
• Removing drum parts from music21.stream before processing. by @napulen in https://github.com/napulen/AugmentedNet/pull/80
• Improve the rntxt output writer. by @napulen in https://github.com/napulen/AugmentedNet/pull/81
• Updating version to 1.7.1 by @napulen in https://github.com/napulen/AugmentedNet/pull/83

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.7.0...v1.7.1

- Jupyter Notebook
Published by napulen almost 4 years ago

augmentednet - v1.7.0

The main contribution of this version is that it introduces the mozartpianosonatas (mps) dataset.

What's Changed

• Reorganizing data collections. by @napulen in https://github.com/napulen/AugmentedNet/pull/66
• Fixing trailing newlines. by @napulen in https://github.com/napulen/AugmentedNet/pull/67
• Adding mozartpianosonatas as submodule. by @napulen in https://github.com/napulen/AugmentedNet/pull/69
• Aggregating the MPS dataset by @napulen in https://github.com/napulen/AugmentedNet/pull/70
• Release v1.7.0 by @napulen in https://github.com/napulen/AugmentedNet/pull/76

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.6.0...v1.7.0

- Jupyter Notebook
Published by napulen about 4 years ago

augmentednet - v1.6.0

This version introduces a new input representation, Duration14.

This input encodes the measure-and note-onset information. That is, it indicates when a new measure begins and the duration elapsed since the last measure. Additionally, it indicates when a new onset begins, and the duration elapsed since the last note onset.

It dramatically improves the performance of the HarmonicRhythm7 task.

What's Changed

• Adding Duration14 input representation. by @napulen in https://github.com/napulen/AugmentedNet/pull/63
• Duration14 fixes by @napulen in https://github.com/napulen/AugmentedNet/pull/64
• Updating pretrained model of v1.6.0. by @napulen in https://github.com/napulen/AugmentedNet/pull/65

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.5.1...v1.6.0

- Jupyter Notebook
Published by napulen about 4 years ago

augmentednet - v1.5.1

Same as v1.5.0. However, the pretrained model was changed for a version without Dropout at the end. It turns out that the HarmonicRhythm classifier is more stable this way.

- Jupyter Notebook
Published by napulen about 4 years ago

augmentednet - v1.5.0

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.4.4...v1.5.0

• Improved padding of sequences #56
• Added an additional option for texturizing the synthetic examples at each transposition (as opposed to one texturization per training file) #58
• Improved parsing/resolution of Roman numeral labels at training and inference

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - v1.4.4

The main change in v1.4.X is that the key vocabulary was extended to 38 keys [Bbb/gb, D#/b#].

Some other fixes include:

• Better handling of Roman numeral translation before training
• More transpositions in data augmentation (byproduct of extending the key vocabulary)

Known issues:

• The key-related tasks are still called LocalKey35 and TonicizedKey35, although they should now be named LocalKey38 and TonicizedKey38.
• Left with old name for compatibility, but changing in future releases of v1.4.X.

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.4.3...v1.4.4

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - v1.3.6

Correcting a one-line but crucial fix in inference.py

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - v1.4.0

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - v1.3.5

What's Changed

• Correcting the Roman numeral reconstruction algorithm. by @napulen in https://github.com/napulen/AugmentedNet/pull/48

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.3.4...v1.3.5

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - v1.3.4

What's Changed

• Removing obsolete module evaluate.py. by @napulen in https://github.com/napulen/AugmentedNet/pull/42
• Modularizing duplicated code. by @napulen in https://github.com/napulen/AugmentedNet/pull/44
• Adding pretrained model v1.3.3+ by @napulen in https://github.com/napulen/AugmentedNet/pull/46

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.3.3...v1.3.4

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - v1.3.3

What's Changed

• Making transposition based on tonicizations rather than modulations. by @napulen in https://github.com/napulen/AugmentedNet/pull/37
• String-replacing iio7 to half-diminished and turning I54,i54 into ton… by @napulen in https://github.com/napulen/AugmentedNet/pull/39

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.3.2...v1.3.3

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - v1.3.2

What's Changed

• ChordQuality11 by @napulen in https://github.com/napulen/AugmentedNet/pull/34
• Updating requirements.txt by @napulen in https://github.com/napulen/AugmentedNet/pull/35

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.3.1...v1.3.2

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - v1.3.1

• Fixing unit tests to work with v1.3.0 model
• Fine-tuning the chord vocabulary and forceTonicization algorithm

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.3.0...v1.3.1

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - v1.2.4

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.2.3...v1.2.4

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - v1.2.3

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.2.2...v1.2.3

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - v1.3.0

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.2.4...v1.3.0

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - Redefining default output representations and metrics

The paper originally described the six conventional tasks, and the alternative method based on the 75-most common Roman numerals.

A Roman numeral can also be reconstructed by predicting the notes conforming the chord. This has become the main approach in this version.

Some other additions include:

• An updated README with instructions for inference and training
• A notebook to run an inference demo
• An updated version of the dataset with the new output representations

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - Minor improvements

This is generally the same version of the AugmentedNet network presented in the 2021 paper, with some minor improvements.

• The HarmonicRhythm representation was improved to account for imbalanced classes. This provides better chord segmentation.
• A new representation based on Closed-Position chord outputs is added to the outputs. The number went from 11 tonal tasks to 14. These new representations are used to predict the final Roman numeral label, instead of all previous methods
• Some minor bugfixes

Full Changelog: https://github.com/napulen/AugmentedNet/compare/v1.0.0...v1.1.0

- Jupyter Notebook
Published by napulen over 4 years ago

augmentednet - Camera-ready submission

This is the version of the code used for all the revised experiments in the camera-ready paper.

Accompanying this release will be the experiment logs, pre-processed data, and data splits of the paper.

One (and maybe the only) substantial difference between the architecture submitted for the initial submission and this one, is the use of rmsprop as the optimizer, instead of adam.

I found that rmsprop was generally better and decided to use it as the default.

Every revised experiment in the camera-ready paper uses rmsprop. As observed before, the results went slightly above just because of that.

- Jupyter Notebook
Published by napulen almost 5 years ago

augmentednet - Submitted version (May 2021)

This is the commit version accompanying the paper submission.

The code is not "production ready", but it is the code used to produce the final results in the paper submitted to ISMIR 2021.

The plan is to clean this code (without making changes to the architecture/output of the model) before releasing to the public.

- Jupyter Notebook
Published by napulen about 5 years ago

augmentednet - Initial version

This version is the result of experiments that have been going for a while.

It is not stable yet, but I am planning on removing several of the experimental scripts I wrote, mostly when I was validating the data from different collections such as ABC, TAVERN, Haydn Op. 20, and When-in-Rome.

I have a pretty standard workflow for pairing scores and annotations nowadays, so most of these scripts (e.g., measure_alignment.py) just clutter the repository.

Nothing major should have been omitted from this release.

- Jupyter Notebook
Published by napulen about 5 years ago