def train_autoencoder_all(methylation_array, latent_dimension):
val_size = int(methylation_array.shape[0] * 0.1)
val_set = AutoencoderGenerator(methylation_array.iloc[:val_size, :])
train_set = AutoencoderGenerator(methylation_array.iloc[val_size:, :])
# Autoencoder training
encoder = Giskard(methylation_array.shape[1],
latent_dimension=latent_dimension,
model_serialization_path="../data/models/")
encoder.fit(train_set,
val_set,
500,
callbacks=[
EarlyStopping(monitor="val_loss",
min_delta=0.05,
patience=10)
])
return encoder
for m in pam50_mirnas:
for gene in mirnas[m]:
if gene[2] > 0.8:
over_rate_genes.append(gene[0])
# dataset = pickle.load(open("../data/mrna_exp_all.pkl", "rb"))[over_rate_mrna]
dataset = dataset[over_rate_genes]
# Generation of training and validation set
val_size = int(dataset.shape[0] * 0.1)
validation_set = AutoencoderGenerator(dataset.iloc[:val_size, :])
training_set = AutoencoderGenerator(dataset.iloc[val_size:, :])
# Autoencoder training
mrna_encoder = Giskard(dataset.shape[1],
latent_dimension=ld,
model_serialization_path="../data/models/")
mrna_encoder.fit(
training_set,
validation_set,
2000,
callbacks=[EarlyStopping(monitor="val_loss", min_delta=0.05, patience=10)])
# Creating an embedded representation of the mRNA methylation array
mrna_to_encode = pickle.load(open("../data/mrna_exp_ma.pkl", "rb"))
mrna_to_encode["beta"] = mrna_to_encode["beta"].rename(
columns=lambda g: g.split('.')[0])[over_rate_genes]
mrna_dataset = mrna_encoder.encode_methylation_array(mrna_to_encode)
pickle.dump(mrna_dataset, open("../data/mrna_embedded_pam.pkl", "wb"))
# Just a check on ground truth