ds_test = tf.data.Dataset.from_tensor_slices((idx_test, y_label[idx_test]))
ds_test = ds_test.batch(len(idx_test), drop_remainder=False)
ds_test = ds_test.map(lambda x, y: ((five_p_loader(x, ragged_output=True),
three_p_loader(x, ragged_output=True),
ref_loader(x, ragged_output=True),
alt_loader(x, ragged_output=True),
strand_loader(x, ragged_output=True)),
y))
histories = []
evaluations = []
weights = []
for i in range(3):
tile_encoder = InstanceModels.VariantSequence(6, 4, 2, [16, 16, 8, 8])
mil = RaggedModels.MIL(instance_encoders=[tile_encoder.model], output_dim=1, pooling='mean', output_type='regression')
losses = ['mse']
mil.model.compile(loss=losses,
metrics=['mse'],
optimizer=tf.keras.optimizers.Adam(learning_rate=0.001,
)
)
callbacks = [tf.keras.callbacks.EarlyStopping(monitor='val_mse', min_delta=0.001, patience=20, mode='min', restore_best_weights=True)]
history = mil.model.fit(ds_train, steps_per_epoch=10, validation_data=ds_valid, epochs=100000, callbacks=callbacks)
evaluation = mil.model.evaluate(ds_test)
histories.append(history.history)
evaluations.append(evaluation)
weights.append(mil.model.get_weights())
with open(cwd / 'figures' / 'controls' / 'samples' / 'sim_data' / 'regression' / 'experiment_3' / 'instance_model_mean.pkl', 'wb') as f:
ds_test = tf.data.Dataset.from_tensor_slices((idx_test, y_label[idx_test]))
ds_test = ds_test.batch(len(idx_test), drop_remainder=False)
ds_test = ds_test.map(lambda x, y: (
(five_p_loader(x, ragged_output=True),
three_p_loader(x, ragged_output=True),
ref_loader(x, ragged_output=True), alt_loader(x, ragged_output=True),
strand_loader(x, ragged_output=True)), y))
X = False
while X == False:
try:
tile_encoder = InstanceModels.VariantSequence(
6, 4, 2, [16, 16, 8, 8])
mil = RaggedModels.MIL(instance_encoders=[tile_encoder.model],
output_dim=1,
pooling='sum',
output_type='other',
instance_layers=[128, 64])
losses = [Losses.CoxPH()]
mil.model.compile(loss=losses,
metrics=[Losses.CoxPH()],
optimizer=tf.keras.optimizers.Adam(
learning_rate=0.001, ))
callbacks = [
tf.keras.callbacks.EarlyStopping(monitor='val_coxph',
min_delta=0.0001,
patience=20,
mode='min',
restore_best_weights=True)
]
history = mil.model.fit(ds_train,
ds_test = tf.data.Dataset.from_tensor_slices((idx_test, y_label[idx_test]))
ds_test = ds_test.batch(len(idx_test), drop_remainder=False)
ds_test = ds_test.map(lambda x, y: (
(five_p_loader(x, ragged_output=True), three_p_loader(x,
ragged_output=True),
ref_loader(x, ragged_output=True), alt_loader(x, ragged_output=True),
strand_loader(x, ragged_output=True)), y))
histories = []
evaluations = []
weights = []
for i in range(3):
tile_encoder = InstanceModels.VariantSequence(6, 4, 2, [16, 16, 8, 8])
mil = RaggedModels.MIL(instance_encoders=[tile_encoder.model],
output_dim=2,
pooling='sum')
losses = [tf.keras.losses.CategoricalCrossentropy(from_logits=True)]
mil.model.compile(
loss=losses,
metrics=[
'accuracy',
tf.keras.metrics.CategoricalCrossentropy(from_logits=True)
],
optimizer=tf.keras.optimizers.Adam(learning_rate=0.001, ))
callbacks = [
tf.keras.callbacks.EarlyStopping(
monitor='val_categorical_crossentropy',
min_delta=0.00001,
patience=50,
mode='min',