validpy.ANN.src package

Submodules

validpy.ANN.src.BackPropTrainer module

class validpy.ANN.src.BackPropTrainer.Trainer(network, momentum, learning_rate, lr_decay, data_set)

Backpropagation trainer

Attributes:
trainer (BackpropTrainer): Backpropagation trainer training_errors (list): List containing the training error for each epoch validation_errors (list): List containing the validation error for each epoch
plot_epochs(folder)

Plot the training and validation errors over the epochs

Parameters:folder (str) – path to the plot file (.png)
Returns:None
Return type:None
train(network, valid_bp, path)

Train until convergence, stopping the training when the training doesn’t reduce the validation error after 1000 continuous epochs

Parameters:
  • network (NeuralNetwork.NeuralNetwork) – model
  • valid_bp (SupervisedDataSet) – Validation set
  • path (str) – Path where to save the trained model
Returns:

None
Return type:

None

validpy.ANN.src.NeuralNetwork module

class validpy.ANN.src.NeuralNetwork.NeuralNetwork(hidden_layers, units, function, input_length, output_length)

Class containing the ANN model

Attributes:
network (FeedForwardNetwork): ANN model
valid(validation_set)

Compute the average euclidean distance activating the model over a validation set

Parameters:validation_set (SupervisedDataSet) – validation set
Returns:average euclidean distance
Return type:float
validpy.ANN.src.NeuralNetwork.initialize_connection(connection, prev_layer, layer)

Initialize a connection between two layers

Parameters:
  • connection (string) – layer’s connection topology (possible values: full)
  • prev_layer (NeuronLayer) – first layer
  • layer (NeuronLayer) – second Layer
Returns:

connection
Return type:

Connection
validpy.ANN.src.NeuralNetwork.initialize_layer(function, length)

Initialize a layer with the given parameters

Parameters:
  • function (string) – layer’s activation function (possible values: linear, sigmoid, gaussian)
  • length (int) – layer’s length
Returns:

layer
Return type:

NeuronLayer

validpy.ANN.src.experimentKcross module

validpy.ANN.src.experimentKcross.grid(conf, data_sets)

Runs the k-cross validation over all possible parameter combinations

Parameters:
  • conf (JSON) – configuration JSON
  • data_sets (list[list[tuple(list[float], list[float])]]) – examples as list of tuple (input, output)
Returns:

None
validpy.ANN.src.experimentKcross.k_cross_experiment(hidden_layers, units, function, learning_rate, momentum, lr_decay, data_sets, out_folder, input_length, output_length, n_processes, csv=None)

Executes the k-cross validation

Parameters:
  • hidden_layers (int) – Number of hidden layers
  • units (int) – Number of units for each hidden layer
  • function (str) – Network’s activation function (possible values: linear, sigmoid, gaussian)
  • learning_rate (float) – Learning rate value
  • momentum (float) – Momentum value
  • lr_decay (float) – Learning rate value
  • data_sets (list[list[tuple(list[float], list[float])]]) – List of the k data sets
  • out_folder (str) – Path to the output folder
  • input_length (int) – Input length
  • output_length (int) – Output length
  • n_processes (int) – Number of parallel processes
  • csv (FileIO) – Output csv file
Returns:

Average error and average training time over the k experiments
Return type:

tuple(float, float)
validpy.ANN.src.experimentKcross.k_train(hidden_layers, units, function, input_length, output_length, momentum, learning_rate, lr_decay, train, valid_bp, valid, folder, report, plot, net_path, q)

Runs the training and the model validation, this function have to been called via Process() https://docs.python.org/3/library/multiprocessing.html

Parameters:
  • hidden_layers (int) – number of hidden layers
  • units (int) – number of units for each hidden layer
  • function (str) – network’s activation function (possible values: linear, sigmoid, gaussian)
  • input_length (int) – input length
  • output_length (int) – output length
  • momentum (float) – momentum value
  • learning_rate (float) – learning rate value
  • lr_decay (float) – learning rate value
  • train (list[tuple(list[float], list[float])]) – training set
  • valid_bp (list[tuple(list[float], list[float])]) – backpropagation validation set
  • valid (list[tuple(list[float], list[float])]) – validation set
  • folder (str) – path to the output experiment folder
  • report (str) – path to the output experiment report file (.txt)
  • plot (str) – path to the output experiment plot file (.png)
  • net_path – path to the output experiment model file (.xml)
  • q (Queue) – Queue
Returns:

training time and average validation error
Return type:

list[float]
validpy.ANN.src.experimentKcross.main(conf_file)

Reads configuration file and runs the k-cross validation experiments

Parameters:conf_file (str) – Path to the configuration file
Returns:None
validpy.ANN.src.experimentKcross.split_data_set_file(file_name, input_length, output_length, k, separator=', ')

Reads the data set from csv, split the input and the output, shuffle the example list and split it in to k subsets

Parameters:
  • file_name (str) – path to the csv file
  • input_length (int) – input length
  • output_length (int) – output length
  • k (int) – number of subsets
  • separator (Optional[str]) – csv file separators, default = ‘,’
Returns:

list of subsets
Return type:

list[list[tuple]]
validpy.ANN.src.experimentKcross.validates(net_path, validation_set)

Compute the average euclidean distance activating the model over a validation set

Parameters:
  • net_path (str) – Path to the model
  • validation_set (list[tuple(list[float], list[float])]) – Validation set
Returns:

average euclidean distance
Return type:

float

Module contents