pycomposer package

Subpackages

• pycomposer.gancomposable package
  • Submodules
  • pycomposer.gancomposable.conditional_gan_composer module
  • Module contents
• pycomposer.noisesampler package
  • Submodules
  • pycomposer.noisesampler.bar_gram_noise_sampler module
  • pycomposer.noisesampler.midi_noise_sampler module
  • Module contents
• pycomposer.truesampler package
  • Submodules
  • pycomposer.truesampler.bar_gram_true_sampler module
  • pycomposer.truesampler.midi_true_sampler module
  • Module contents

Submodules

pycomposer.bar_gram module

class pycomposer.bar_gram.BarGram(midi_df_list, time_fraction=0.1)

Bases: object

The base class for n-gram representation of pitch in each bar.

dim

getter

extract_features(df)

get_dim()

getter

get_pitch_tuple_list()

getter

pitch_tuple_list

getter

set_readonly(value)

setter

pycomposer.gan_composable module

class pycomposer.gan_composable.GANComposable

Bases: object

The interface to build an Algorithmic Composer based on Generative Adversarial Networks(GANs) or its variants such as Conditional Generative Adversarial Networks(Conditional GANs)(Yang, L. C., et al., 2017) and Algorithmic Composer based on Adversarial Auto-Encoders(AAEs) (Makhzani, A., et al., 2015).

In the general GAN framework, the composer learns observed data points drawn from a true distribution of input MIDI files and generates feature points drawn from a fake distribution that means such as Uniform distribution or Normal distribution, imitating the true MIDI files data.

The components included in this class are functionally differentiated into three models.

1. TrueSampler.
2. Generator.
3. Discriminator.

The function of TrueSampler is to draw samples from a true distribution of input MIDI files. Generator has NoiseSampler`s and draw fake samples from a Uniform distribution or Normal distribution by use it. And `Discriminator observes those input samples, trying discriminating true and fake data.

While Discriminator observes Generator’s observation to discrimine the output from true samples, Generator observes Discriminator’s observations to confuse `Discriminator`s judgments. In GANs framework, the mini-max game can be configured by the observations of observations.

After this game, the Generator will grow into a functional equivalent that enables to imitate the TrueSampler and makes it possible to compose similar but slightly different music by the imitation.

References

• Fang, W., Zhang, F., Sheng, V. S., & Ding, Y. (2018). A method for improving CNN-based image recognition using DCGAN. Comput. Mater. Contin, 57, 167-178.
• Gauthier, J. (2014). Conditional generative adversarial nets for convolutional face generation. Class Project for Stanford CS231N: Convolutional Neural Networks for Visual Recognition, Winter semester, 2014(5), 2.
• Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., … & Bengio, Y. (2014). Generative adversarial nets. In Advances in neural information processing systems (pp. 2672-2680).
• Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional networks for semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 3431-3440).
• Makhzani, A., Shlens, J., Jaitly, N., Goodfellow, I., & Frey, B. (2015). Adversarial autoencoders. arXiv preprint arXiv:1511.05644.
• Yang, L. C., Chou, S. Y., & Yang, Y. H. (2017). MidiNet: A convolutional generative adversarial network for symbolic-domain music generation. arXiv preprint arXiv:1703.10847.

compose(file_path, velocity_mean=None, velocity_std=None)

Compose by learned model.

Parameters:

• file_path – Path to generated MIDI file.
• velocity_mean – Mean of velocity. This class samples the velocity from a Gaussian distribution of velocity_mean and velocity_std. If None, the average velocity in MIDI files set to this parameter.
• velocity_std – Standard deviation(SD) of velocity. This class samples the velocity from a Gaussian distribution of velocity_mean and velocity_std. If None, the SD of velocity in MIDI files set to this parameter.

extract_logs()

Extract update logs data.

Returns:Tuple data. The shape is: - list of probabilities inferenced by the discriminator (mean) in the discriminator’s update turn. - list of probabilities inferenced by the discriminator (mean) in the generator’s update turn.

learn(iter_n=500, k_step=10)

Learning.

Parameters:

• iter_n – The number of training iterations.
• k_step – The number of learning of the discriminator.

pycomposer.midi_controller module

class pycomposer.midi_controller.MidiController

Bases: object

MIDI Controller.

extract(file_path, is_drum=False)

Extract MIDI file.

Parameters:

• file_path – File path of MIDI.
• is_drum – Extract drum data or not.

Returns:

pd.DataFrame(columns=[“program”, “start”, “end”, “pitch”, “velocity”, “duration”])

save(file_path, note_df)

Save MIDI file.

Parameters:

• file_path – File path of MIDI.
• note_df – pd.DataFrame of note data.

Module contents