attributeerror: 'sequential' object has no attribute "get_build_config"
Closed this issue · 3 comments
areflesh commented
This issue appeared after switching from version 1.3.5 to 1.4.2. No code changes were done.
if args.auto_tune:
print('AUTO TUNING ENABLED')
def build_model(hp):
model = Sequential()
model.add(Dense(input_dims+1, input_dim=input_dims, activation='relu'))
if args.normalize:
model.add(tf.keras.layers.BatchNormalization())
for i in range(hp.Int('num_layers', 1,5)):
model.add(Dense(
hp.Choice('units', [32, 64, 128, 256, 512, 1024]),
activation='relu'))
model.add(Dropout(args.dropout_rate))
model.add(Dense(2, activation='softmax'))
if args.gradient_clipping:
optimizer = tf.keras.optimizers.Adam(clipnorm=args.gradient_clipping_value)
else:
optimizer = 'adam'
model.compile(loss="kullback_leibler_divergence", optimizer=optimizer,
metrics=[
tf.keras.metrics.BinaryAccuracy(name='accuracy'),
tf.keras.metrics.Precision(name='precision'),
tf.keras.metrics.Recall(name='recall'),
tf.keras.metrics.AUC(name='auc'),
tf.keras.metrics.AUC(name='prc', curve='PR'),])
return model
if args.objective == 'prc':
objective = keras_tuner.Objective("prc", direction="max")
elif args.objective == 'loss':
objective = keras_tuner.Objective("loss", direction="min")
else:
raise ValueError('objective must be either prc or loss')
tuner = keras_tuner.BayesianOptimization(
build_model,
objective=objective,
max_trials=args.max_trials,
executions_per_trial=1,
directory="s3://{}/{}/model/auto_tune".format(args.s3_bucket, args.s3_dir),
overwrite=True)
tuner.search_space_summary()
tuner.search(dataset,
epochs = args.max_epochs_tune,
verbose = 2,
class_weight = class_weights)
print('TUNNING DONE. BEST HYPERPARAMETERS:')
tuner.results_summary()
model = tuner.get_best_models(num_models=1)[0]
model.build()
Issue appear after each trial and after third trial it says RuntimeError: Number of consecutive failures exceeded the limit of 3 and stops
haifeng-jin commented
Hi @areflesh , it is probably because the TF version you use?
We do model.get_build_config()
under the hood.
As I tested in colab, Sequential
model class has this method.
pacs27 commented
I have the same error when I use tensorflow version 2.10 which does not have this method
haifeng-jin commented
It has been fixed in master. Will be available in the next release. (1.4.3)
It is much appreciated if anyone had the bug could test it on our master branch. Thanks!