LSTM Class

LSTM Deep Learning Model

Based on model proposed by Yee-King et al.

“Automatic programming of VST sound synthesizers using deep networks and other techniques.” Yee-King, Matthew John, Leon Fedden, and Mark d’Inverno. IEEE Transactions on Emerging Topics in Computational Intelligence 2.2 (2018): 150-159.

class spiegel.estimator.lstm.LSTM(inputShape, numOutputs, **kwargs)

Bases: spiegel.estimator.tf_estimator_base.TFEstimatorBase

Parameters

  • inputShape (tuple) – Shape of matrix that will be passed to model input

  • numOutputs – Number of outputs the model has

  • kwargs – optional keyword arguments to pass to spiegel.estimator.TFEstimatorBase

addTestingData(input, output, batchSize=64)

Create a tf Dataset from input and output for model testing, batches data if desired

Parameters

  • input (np.array) – matrix of data to use as testing data

  • output (np.array) – matrix of data to use as ground truth for testing data

  • batchSize (int, optional) – If provided, will batch data into batches of this size, set to None or 0 to prevent batching. defaults to 64

addTrainingData(input, output, batchSize=64, shuffleSize=None)

Create a tf Dataset from input and output, and shuffles / batches data for training

Parameters

  • input (np.array) – matrix of training data

  • output (np.array) – matrix of training data ground truth

  • batchSize (int, optional) – If provided, will batch data into batches of this size, set to None or 0 to prevent batching. defaults to 64

  • shuffleSize – If provided, will shuffle data with a buffer size of shuffleSize, defaults to None, so shuffling does not occur

buildModel(hiddenSize=100)

Construct LSTM Model

Parameters

hiddenSize (int, optional) – dimensionality of outer space of hidden layers, defaults to 100

fit(epochs=1, callbacks=[], **kwargs)

Train model on for a fixed number of epochs on training data and validation data if it has been added to this estimator

Parameters

  • epochs (int, optional) – Number of epochs to train model on, defaults to 1

  • callbacks (list, optional) – List of callback functions for training, defaults to []

  • kwargs – Keyword args passed to model fit method. See Tensflow Docs.

loadModelFromCheckpoint()

Load model weights from checkpoint

loadWeights(filepath, **kwargs)

Load model weights from H5 or TensorFlow file

Parameters

  • filepath (string) – filepath to saved model weights

  • kwargs – optional keyword arguments passed to tf load_weights methods, see TensorFlow Docs.

predict(input)

Run prediction on input

Parameters

input (np.array) – matrix of input data to run predictions on. Can be a single instance of data or a batch.

static rootMeanSquaredError(labels, prediction)

Static method for calculating root mean squared error between predictions and targets

Parameters

  • labels (Tensor) – Matrix of ground truth labels

  • prediction (Tensor) – Matrix of predictions

saveWeights(filepath, **kwargs)

Save model weights to a HDF5 or TensorFlow file.

Parameters

  • filepath (string) – filepath to save model weights. Using a file suffix of ‘.h5’ or ‘.keras’ will save in HDF5 format. Otherwise will save as TensorFlow.

  • kwargs – optional keyword arguments passed to tf save_weights method, see Tensflow Docs.