askrbayern / idyompy

This project propose a Python implementation for the IDyOM model made by Marcus Pearce.

IDyOMpy

IDyOMpy is a Python re-implementation of the IDyOM model made by Marcus Pearce. This implementation was made by Guilhem Marion, a Ph.D. student at Laboratoire des Systèmes Perceptifs, ENS, Paris. You can find the documentation on this website. Please cite the related paper if you use this work.

Get Started

Install conda if you don't have it (we highly recommend)

visit: https://conda.io/projects/conda/en/latest/user-guide/install/index.html

Create a new environment

conda create -n idyompyenv

Activate environment (you will have to activate it every time you need it and deactivate when you don't need it):

conda activate idyompyenv

Get the project from GitHub:

git clone https://github.com/GuiMarion/IDyOM.git

Enter the project

cd IDyOM

Install the dependencies:

pip install -r requirements.txt

Then use it!

To run a cross-validation with 5-folds:

python3 App.py -c yourFolder/ -k 5

To train/test:

python3 App.py -t trainingFolder/ -s testingFolder/

To replicate figures from Music of Silence Part II:

python3 App.py -t trainingFolder/ -d testingFolder/

To change the maximal order to 10:

python3 App.py -t trainingFolder/ -s testingFolder/ -m 10

To only use the pitch information:

python3 App.py -t trainingFolder/ -s testingFolder/ -v pitch

To compute genuine entropies and not use the approximation (it takes 5 times longer):

python3 App.py -t trainingFolder/ -s testingFolder/ -g 1

To compute the training monitoring (c.f. method paper):

python3 App.py -i preTrainingFolder/ -e trainingFolder/

To only use the long-term model:

python3 App.py -t trainingFolder/ -s testingFolder/ -l 1

To only use the short-term model:

python3 App.py -t trainingFolder/ -s testingFolder/ -b 1

Of course, all those parameters can be mixed to reach the behavior you are looking for.

Usage: usage App.py [options]

Options:

-h, --help 	show this help message and exit
-a TESTS, --test=TESTS
 	1 if you want to launch unittests
-t TRAIN_FOLDER, --train=TRAIN_FOLDER
 	Train the model with the passed folder
-s TRIAL_FOLDER, --surprise=TRIAL_FOLDER
 	Compute surprise over the passed folder. We use -t argument to train, if none are privided, we use the passed folder to cross-train.
-n TRIAL_FOLDER_SILENT, --silentNotes=TRIAL_FOLDER_SILENT
 	Compute silent notes probabilities over the passed folder. We use -t argument to train, if none are provided, we use the passed folder to cross-train.
-d THRESHOLD_MISSING_NOTES, --threshold_missing_notes=THRESHOLD_MISSING_NOTES
 	Define the threshold for choosing the missing notes (0.2 by default)
-z ZERO_PADDING, --zero_padding=ZERO_PADDING
 	Specify if you want to use zero padding in the surprise output, enable time representation (default 0)
-b SHORT_TERM_ONLY, --short_term=SHORT_TERM_ONLY
 	Only use short term model (default 0)
-c CROSS_EVAL, --cross_eval=CROSS_EVAL
 	Compute likelihoods by pieces over the passed dataset using k-fold cross-eval.
-l LONG_TERM_ONLY, --long_term=LONG_TERM_ONLY
 	Only use long term model (default 0)
-k K_FOLD, --k_fold=K_FOLD
 	Specify the k-fold for all cross-eval, you can use -1 for leave-one-out (default 5).
-q QUANTIZATION, --quantization=QUANTIZATION
 	Rythmic quantization to use (default 24).
-v VIEWPOINTS, --viewPoints=VIEWPOINTS
 	Viewpoints to use: pitch, length, interval, and velocity, separate them with comas, default pitch,length.
-m MAX_ORDER, --max_order=MAX_ORDER
 	Maximal order to use (default 20).
-g GENUINE_ENTROPIES, --genuine_entropies=GENUINE_ENTROPIES
 	Use this parameter to NOT use the entropy approximation. It takes longer (5 times) to compute but generate the genuine entropies, not an approximation (default 0).
-r FOLDER_DUPLICATES, --check_dataset=FOLDER_DUPLICATES
 	Check whether the passed folder contains duplicates.
-e TRAIN_TEST_FOLDER, --evolution=TRAIN_TEST_FOLDER
 	Train and evaluate over training on the passed folder (cross-val).
-i INTIALIZATION, --init_evolution=INTIALIZATION
 	Folder to initialize the evolution on.
-p NB_PIECES, --nb_pieces=NB_PIECES
 	Number of pieces to evaluate on during evolution training.
-o ORIGINAL_PPM, --original_ppm=ORIGINAL_PPM
 	Use original PPM algorithm to calculate likelihoods(default 0).