def train_dnn_classifier_deep_cc(classifier, train_set, val_set, test_data):
params = {
"input_shape": train_set["beta"].shape[1],
"model_serialization_path": "../data/models/classifier/",
"dropout_rate": 0.3,
"output_shape": len(train_set["pheno"]["subtype"].unique())
}
model = classifier(**params)
model.fit(MethylationArrayGenerator(train_set, "subtype"),
MethylationArrayGenerator(val_set, "subtype"),
40,
verbose=0,
callbacks=[
EarlyStopping(monitor="val_loss",
min_delta=0.05,
patience=20)
])
dummies = pd.get_dummies(train_set["pheno"]["subtype"]).columns.to_list()
tmp_acc = []
for p, r in zip(model.predict(test_data["beta"].to_numpy()),
test_data["pheno"].values.ravel()):
if dummies[np.argmax(p)] == r:
tmp_acc.append(1)
else:
tmp_acc.append(0)
return model, np.mean(tmp_acc)
def train_dnn_classifier(classifier, train_set, val_set, test_data):
params = {"input_shape": train_set["beta"].shape[1], "model_serialization_path": "../data/models/classifier/",
"dropout_rate": 0.3, "output_shape": len(train_set["pheno"]["subtype"].unique())}
model = classifier(**params)
model.fit(MethylationArrayGenerator(train_set, "subtype"),
MethylationArrayGenerator(val_set, "subtype"),
500,
verbose=0,
callbacks=[EarlyStopping(monitor="val_loss", min_delta=0.05, patience=20)])
test_accuracy = model.evaluate(test_data["beta"].to_numpy(),
pd.get_dummies(test_data["pheno"]["subtype"]).to_numpy())
return model, test_accuracy
def methylation_array_kcv(dataset,
model_class,
model_params,
output_target,
k=10,
verbose=0,
callbacks=[]):
"""
KCV evaluation of a model that implements AbstractClassifier
:param dataset: the methylation array filename or the methylation array itself
:param model_class: the model class
:param model_params: a dictionary containing the parameters to init the class
:param output_target: the label
:param k: the folds
:param verbose: verbose mode for fit method
:param callbacks: the callbacks to pass to the fit method
:return: the average accuracy
"""
test_accuracies, val_accuracies = [], []
for i in range(k):
training_set, test_set, validation_set = split_methylation_array_by_pheno(
dataset, output_target)
model = model_class(**model_params)
model.fit(MethylationArrayGenerator(training_set, output_target),
MethylationArrayGenerator(validation_set, output_target),
500,
verbose=verbose,
callbacks=callbacks)
test_accuracies.append(
model.evaluate(
test_set["beta"].to_numpy(),
pd.get_dummies(test_set["pheno"][output_target]).to_numpy()))
val_accuracies.append(
model.evaluate(validation_set["beta"].to_numpy(),
pd.get_dummies(validation_set["pheno"].to_numpy())))
return sum(val_accuracies) / len(val_accuracies), sum(
test_accuracies) / len(test_accuracies)