def superlayeredNN_hyperparameters_tuning():
#epigeneticsData = EmbryoDevelopmentDataWithClusters(2, 3, 256, 6)
epigeneticsData = CancerDataWithModalities(5)
training_dataset, validation_dataset, test_dataset = epigeneticsData.get_training_validation_test_datasets(
)
clusters_size = epigeneticsData.clusters_size
output_size = epigeneticsData.output_size
"""superlayered_nn = SuperlayeredNeuralNetwork(
[clusters_size[0], clusters_size[1]],
[[512, 256, 128, 64], [512, 256, 128, 64]],
[128, 64],
output_size)"""
superlayered_nn = SuperlayeredNeuralNetwork(
[clusters_size[0], clusters_size[1]],
[[64, 256, 128, 16], [64, 256, 128, 16]], [64, 16], output_size)
learning_rate = 0.02
weight_decay = 0.01
keep_probability = 0.50
training_accuracy, validation_accuracy, steps_list, test_accuracy = superlayered_nn.train_validate_test(
training_dataset, validation_dataset, test_dataset, learning_rate,
weight_decay, keep_probability)
plot_training_accuracy_and_validation_accuracy(training_accuracy,
validation_accuracy,
steps_list)
def test_SNN(self):
superlayered_nn = SuperlayeredNeuralNetwork(
[cluster_1_num_genes, cluster_2_num_genes],
[[256, 128, 64, 32], [256, 128, 64, 32]], [128, 32], num_classes)
learning_rate = 0.05
weight_decay = 0.05
keep_probability = 0.5
validation_accuraty, confussion_matrix, ROC_points = superlayered_nn.train_and_evaluate(
self.training_dataset_with_clusters,
self.validation_dataset_with_clusters, learning_rate, weight_decay,
keep_probability)
print confussion_matrix
def evaluate_superlayered_neural_network(epigenetic_data_with_clusters):
print "-------------------------------------------------------------------------------"
print "-------------------Evaluating Superlayered Neural Network----------------------"
print "-------------------------------------------------------------------------------"
clusters_size = epigenetic_data_with_clusters.clusters_size
output_size = epigenetic_data_with_clusters.output_size
"""superlayered_neural_network = SuperlayeredNeuralNetwork(
[clusters_size[0], clusters_size[1]],
[[256, 128, 64, 32], [256, 128, 64, 32]],
[128, 32], output_size)"""
superlayered_neural_network = SuperlayeredNeuralNetwork(
[clusters_size[0], clusters_size[1]],
[[512, 256, 128, 64], [512, 256, 128, 64]], [256, 64], output_size)
"""superlayered_neural_network = SuperlayeredNeuralNetwork(
[clusters_size[0], clusters_size[1]],
[[512, 256, 128, 64], [512, 256, 128, 64]],
[256, 64], output_size)"""
confussion_matrix, ROC_points, performance_metrics = nested_cross_validation_on_SNN(
superlayered_neural_network, epigenetic_data_with_clusters)
label_to_one_hot_encoding = epigenetic_data_with_clusters.label_to_one_hot_encoding
class_id_to_symbol_id = compute_class_id_to_class_symbol(
label_to_one_hot_encoding)
class_symbol_to_evaluation_matrix = compute_evaluation_metrics_for_each_class(
confussion_matrix, class_id_to_symbol_id)
#plot_confussion_matrix_as_heatmap(confussion_matrix, label_to_one_hot_encoding, 'Confusion Matrix for SNN')
#plot_ROC_curves(ROC_points)
return confussion_matrix, ROC_points, performance_metrics
def plot_keep_probability(epigenetic_data, epigenetic_data_with_clusters):
_, k_fold_datasets_hyperparameters_tuning = epigenetic_data.get_k_fold_datasets(
)
input_data_size = epigenetic_data.input_size
output_size = epigenetic_data.output_size
MLP = MultilayerPerceptron(input_data_size, [128, 64, 32, 16], output_size)
RNN = RecurrentNeuralNetwork(input_sequence_length=input_data_size / 4,
input_step_size=4,
LSTM_units_state_size=[16, 64],
hidden_units=[32, 16],
output_size=output_size)
print("Got here")
keep_probability_plots(MLP, k_fold_datasets_hyperparameters_tuning, 0.05,
'mlp_kp')
keep_probability_plots(RNN, k_fold_datasets_hyperparameters_tuning, 0.0001,
'rnn_kp')
k_fold_datasets, k_fold_datasets_hyperparameters_tuning = epigenetic_data_with_clusters.get_k_fold_datasets(
)
clusters_size = epigenetic_data_with_clusters.clusters_size
output_size = epigenetic_data_with_clusters.output_size
SNN = SuperlayeredNeuralNetwork([clusters_size[0], clusters_size[1]],
[[128, 64, 32, 16], [128, 64, 32, 16]],
[32, 16], output_size)
keep_probability_plots(SNN, k_fold_datasets_hyperparameters_tuning, 0.05,
'snn_kp')
def plot_learning_rate(epigenetic_data, epigenetic_data_with_clusters):
_, k_fold_datasets_hyperparameters_tuning = epigenetic_data.get_k_fold_datasets(
)
input_data_size = epigenetic_data.input_size
output_size = epigenetic_data.output_size
MLP = MultilayerPerceptron(input_data_size, [256, 128, 64, 32],
output_size)
RNN = RecurrentNeuralNetwork(input_sequence_length=input_data_size / 4,
input_step_size=4,
LSTM_units_state_size=[32, 128],
hidden_units=[128, 32],
output_size=output_size)
#learning_rate_plots(MLP, k_fold_datasets_hyperparameters_tuning, 'mlp_ln')
#learning_rate_plots(RNN, k_fold_datasets_hyperparameters_tuning, 'rnn_ln')
k_fold_datasets, k_fold_datasets_hyperparameters_tuning = epigenetic_data_with_clusters.get_k_fold_datasets(
)
clusters_size = epigenetic_data_with_clusters.clusters_size
output_size = epigenetic_data_with_clusters.output_size
SNN = SuperlayeredNeuralNetwork([clusters_size[0], clusters_size[1]],
[[128, 64, 32, 16], [128, 64, 32, 16]],
[32, 16], output_size)
learning_rate_plots(SNN, k_fold_datasets_hyperparameters_tuning, 'snn_ln')