Args: args: parameters of the model

Args: args: parameters of the model

Args: args: parameters of the model

Args: args: parameters of the model

Args: args: parameters of the model

Args: args: parameters of the model

Return the absolute tick when the last note end Note that the length is recomputed each time the function is called

Create a new song from a numpy array A note will be created for each non empty case of the array. The song will contain a single track, and u

Convert a given song to a numpy multi-dimensional array (piano roll) The song is temporally normalized, meaning that all ticks and duration w

Create songs from the decoder outputs. Reshape the list of outputs to list of piano rolls Args: outputs (List[np.array]):

Just a constant learning rate

Just a constant learning rate

Every decay period, the learning rate is divided by 2

Every decay period, the learning rate is divided by 2

Initialize the weights of the model

Initialize the weights of the model

Initialize the weights of the model

Initialize the weights of the model

Initialize the weights of the model

Initialize the weights of the model

In case of a generator (batches non precomputed), compute the batch given the batch id passed Args: batch: the current te

Create a placeholder compatible with the batch input Allow to control the dimensions Return: tf.placeholder: the placehol

Create a placeholder compatible with the target Allow to control the dimensions Return: tf.placeholder: the placeholder f

Recursively extract all files from a given directory

Generator over the mini-batch training samples Warning: the last samples will be ignored if the number of batch does not match the number

Generator over the mini-batch training samples Warning: the last samples will be ignored if the number of batch does not match the number

Args: target(Bool): is true if the bach also need to generate the target

Return the batches which initiate the RNN when generating The initial batches are loaded from a json file containing the first notes of the s

Predict the next keyboard state Args: prev_keyboard (?): the previous keyboard configuration prev_state_enco (?): the

Simple 1 hidden layer perceptron See parent class for arguments details

Predict the next keyboard state Args: prev_keyboard (tf.Tensor): the previous keyboard configuration prev_state (Tupl

Predict the next keyboard state Args: prev_keyboard (tf.Tensor): the previous keyboard configuration prev_state (Tupl

a RNN encoder See parent class for arguments details

a RNN encoder See parent class for arguments details

Return the input dimension Return: int:

Return the learning rate associated at the current training step Args: glob_step (int): Number of iterations since the beginning

Return the learning rate associated at the current training step Args: glob_step (int): Number of iterations since the beginning

Compute the batches for the current epoch Is called twice (for training and testing) Args: dataset (list[Objects]): the t

Return the unique id associated with the builder Ultimately, the id will be used for saving/loading the dataset, and as parameter arg

Return the list of added modules Useful for instance for the command line parser Returns: list[str]: the list of modules

Return the chosen labels from the softmax distribution Allows to reconstruct the song

Return the list of the different modes Useful when parsing the command lines arguments

Return the list of the different modes Useful when parsing the command lines arguments

Return the list of the different modes Useful when parsing the command lines arguments

Return RNN cell, constructed from the parameters Args: args: the rnn parameters scope_name (str): encapsulate variables Retur

Return the initial cell state

Return the initial cell state

Return the initial cell state

Return the initial cell state

Return the initial cell state

Extract data from midi file Args: filename (str): a valid midi file Return: Song: a song object containing th

Loop function used to connect one output of the rnn to the next input. Will re-adapt the output shape to the input one. This

Loop function used to connect one output of the rnn to the next input. The previous input and returned value have to be from the same sha

Create the batch associated with the song Called when generating songs to create the initial input batch Args: raw_song (

Apply some pre-processing to the songs so the song already get the right input representation. Do it once globally for all songs

Project the output of the decoder into the note space Args: X (tf.Tensor): input value

Create the song associated with the network output Args: output (list[np.Array]): The ouput of the network (size batch_size*outpu

Reconstruct the original raw song from the preprocessed data We should have: reconstruct_song(process_song(my_song)) == my_song

Rename all files of the given directory following some rules

Single layer perceptron Project X on the output dimension Args: shape: a tuple (input dim, output dim) scope_name (str): enca

Forward/training step operation. Does not perform run on itself but just return the operators to do so. Those have then to be run by the

Save the song on disk Args: song (Song): a song object containing the tracks and melody filename (str): the path were