def single_gpu_train(batch_size,
num_gpu,
data_path,
num_epochs,
model_path,
learning_rate,
loss=qa_pair_loss):
global_batch_size = batch_size * num_gpu
print("global_batch_size", global_batch_size)
afaf
learning_rate = learning_rate
# ================================================================================
d = create_dataset_for_ffn(data_path,
batch_size=global_batch_size,
shuffle_buffer=500000)
eval_d = create_dataset_for_ffn(data_path,
batch_size=batch_size,
mode='eval')
medical_qa_model = MedicalQAModel()
optimizer = tf.keras.optimizers.Adam(lr=learning_rate)
medical_qa_model.compile(optimizer=optimizer,
loss=loss,
metrics=[qa_pair_batch_accuracy])
epochs = num_epochs
medical_qa_model.fit(d, epochs=epochs, validation_data=eval_d)
medical_qa_model.save_weights(model_path)
return medical_qa_model
def train_ffn(
model_path='/mnt/1T-5e7/mycodehtml/bio_health/NLP/Santosh_Gupta/ffn_cross_entropy/drive-download-20190429T033651Z-001',
data_path='data/mqa_csv',
num_epochs=300,
num_gpu=1,
# batch_size=64,
batch_size=4,
learning_rate=0.0001,
validation_split=0.2,
loss='categorical_crossentropy'):
if loss == 'categorical_crossentropy':
loss_fn = qa_pair_cross_entropy_loss
else:
loss_fn = qa_pair_loss
# ================================================================================
eval_d = create_dataset_for_ffn(data_path,
batch_size=batch_size,
mode='eval')
if num_gpu > 1:
medical_qa_model = multi_gpu_train(batch_size, num_gpu, data_path,
num_epochs, model_path,
learning_rate, loss_fn)
else:
medical_qa_model = single_gpu_train(batch_size, num_gpu, data_path,
num_epochs, model_path,
learning_rate, loss_fn)
medical_qa_model.summary()
medical_qa_model.save_weights(model_path, overwrite=True)
def multi_gpu_train(batch_size,
num_gpu,
data_path,
num_epochs,
model_path,
learning_rate,
loss=qa_pair_loss):
mirrored_strategy = tf.distribute.MirroredStrategy(
devices=DEVICE[:num_gpu])
global_batch_size = batch_size * num_gpu
learning_rate = learning_rate * 1.5**num_gpu
with mirrored_strategy.scope():
d = create_dataset_for_ffn(data_path,
batch_size=global_batch_size,
shuffle_buffer=100000)
d_iter = mirrored_strategy.make_dataset_iterator(d)
medical_qa_model = tf.keras.Sequential()
medical_qa_model.add(tf.keras.layers.Input((2, 768)))
medical_qa_model.add(MedicalQAModel())
optimizer = tf.keras.optimizers.Adam(lr=learning_rate)
medical_qa_model.compile(optimizer=optimizer, loss=loss)
epochs = num_epochs
loss_metric = tf.keras.metrics.Mean()
medical_qa_model.fit(d_iter,
epochs=epochs,
metrics=[qa_pair_batch_accuracy])
medical_qa_model.save_weights(model_path)
return medical_qa_model
def train_ffn(model_path='models/ffn_crossentropy/ffn',
data_path='data/mqa_csv',
num_epochs=300,
num_gpu=1,
batch_size=64,
learning_rate=0.0001,
validation_split=0.2,
loss='categorical_crossentropy'):
if loss == 'categorical_crossentropy':
loss_fn = qa_pair_cross_entropy_loss
else:
loss_fn = qa_pair_loss
eval_d = create_dataset_for_ffn(data_path,
batch_size=batch_size,
mode='eval')
if num_gpu > 1:
medical_qa_model = multi_gpu_train(batch_size, num_gpu, data_path,
num_epochs, model_path, loss_fn)
else:
medical_qa_model = single_gpu_train(batch_size, num_gpu, data_path,
num_epochs, model_path, loss_fn)
medical_qa_model.summary()
medical_qa_model.save_weights(model_path, overwrite=True)