def main(dataPath, dataPath_val, batch_size):
# must have list of training files
files = glob.glob(dataPath + "/*.csv")[::5]
# validation files
files_val = glob.glob(dataPath_val + "/*.csv")[::5]
# Count number of examples in training data
nexs = get_total_examples(files)
print("Number of training examples: ", nexs)
nexs_val = get_total_examples(files_val)
print("Number of validation examples: ", nexs_val)
# Create data generator
train_gen = DataGenerator(files, nexs, batch_size=batch_size)
val_gen = DataGenerator(files_val, nexs_val, batch_size=batch_size)
linear_feature_columns = train_gen.linear_feature_columns
dnn_feature_columns = train_gen.dnn_feature_columns
# 4.Define Model,train,predict and evaluate
model = DeepFM(linear_feature_columns, dnn_feature_columns, task='binary')
optimizer = keras.optimizers.Adam(lr=0.001,
beta_1=0.9,
beta_2=0.999,
decay=0.0)
model.compile(
optimizer,
"binary_crossentropy",
metrics=['binary_crossentropy', auroc],
)
pbar = ProgbarLogger(count_mode='steps', stateful_metrics=None)
weights_file = "model-5-lr0p001.h5"
model_checkpoint = ModelCheckpoint(weights_file,
monitor="val_binary_crossentropy",
save_best_only=True,
save_weights_only=True,
verbose=1)
history = model.fit_generator(train_gen,
epochs=10,
verbose=1,
steps_per_epoch=nexs / batch_size,
validation_data=val_gen,
validation_steps=nexs / batch_size,
callbacks=[model_checkpoint])
monitor='val_loss',
verbose=1,
save_best_only=True,
mode='min',
save_weights_only=True)
#history = model.fit_generator(generate_arrays_from_file('./data/sample/feature_mapped_valid.data', batch_size=batch_size),
# steps_per_epoch=int(np.ceil(num_train/batch_size)), callbacks=[checkpoint], epochs=50, verbose=1,
# validation_data=generate_arrays_from_file('./data/sample/feature_mapped_valid.data', batch_size=batch_size),
# validation_steps=int(np.ceil(num_valid/batch_size)))
history = model.fit_generator(
generate_arrays_from_file(
'./data/feature_mapped_combined_train.data',
batch_size=batch_size),
steps_per_epoch=int(np.ceil(num_train / batch_size)),
callbacks=[checkpoint],
epochs=50,
verbose=1,
validation_data=generate_arrays_from_file(
'./data/feature_mapped_combined_valid.data',
batch_size=batch_size),
validation_steps=int(np.ceil(num_valid / batch_size)))
elif mode == 'test':
# model.load_weights('model_save/deep_fm_fn-ep002-loss0.148-val_loss0.174.h5') # auc: 0.718467 batch_size=6000
#model.load_weights('model_save/deep_fm_fn-ep001-loss0.149-val_loss0.175.h5') # auc: 0.714243 batch_size = 2048
# model.load_weights('model_save/deep_fm_fn-ep005-loss0.147-val_loss0.173.h5') # auc: 0.722535 batch_size = 10000
# model.load_weights('model_save/deep_fm_fn_bs10000-ep001-loss0.155-val_loss0.153.h5') # auc: 0.738023
#model.load_weights('model_save/deep_fm_fn_bs15000-ep001-loss0.156-val_loss0.152.h5') # auc: 0.739935
#model.load_weights('model_save/deep_fm_fn-ep002-loss0.154-val_loss0.154-bs15000-ee20-hz[128, 128].h5') # auc: 0.741590
model.load_weights(
'model_save/deep_fm_fn-ep020-loss0.153-val_loss0.153-bs15000-ee20-hz[5, 600].h5'
return dataset
linear_feature_columns = varlen_feature_columns + fixed_feature_columns
dnn_feature_columns = varlen_feature_columns + fixed_feature_columns
callbacks = []
GPU = True
if GPU:
strategy = tf.distribute.MirroredStrategy(devices=['/gpu:0', '/gpu:1', '/gpu:2', '/gpu:3'])
# strategy = tf.distribute.MirroredStrategy(devices=['/gpu:3'])
with strategy.scope():
model = DeepFM(linear_feature_columns, dnn_feature_columns, dnn_hidden_units=[1024, 512, 256],
task='binary',
dnn_dropout=0, dnn_activation='relu', dnn_use_bn=False)
model.compile("adam", "binary_crossentropy", metrics=['binary_crossentropy', tf.keras.metrics.AUC()])
# model.run_eagerly = True
model.fit_generator(generator=get_dataset(), steps_per_epoch=None, epochs=10, verbose=2, callbacks=callbacks,
validation_data=get_dataset(eval_data_path), validation_steps=None, validation_freq=1,
class_weight=None,
max_queue_size=100, workers=10, use_multiprocessing=False, shuffle=True, initial_epoch=0)
tf.saved_model.save(model, "./models")
else:
model = DeepFM(linear_feature_columns, dnn_feature_columns, dnn_hidden_units=[1024, 512, 256], task='binary',
dnn_dropout=0, dnn_activation='relu', dnn_use_bn=False)
model.compile("adam", "binary_crossentropy", metrics=['binary_crossentropy', tf.keras.metrics.AUC()])
model.run_eagerly = True
model.fit_generator(generator=get_dataset(), steps_per_epoch=None, epochs=10, verbose=2, callbacks=callbacks,
validation_data=get_dataset(eval_data_path), validation_steps=None, validation_freq=1,
class_weight=None,
max_queue_size=100, workers=10, use_multiprocessing=False, shuffle=True, initial_epoch=0)
