Feature: Any GPU optimization features?
jbrepogmailcom opened this issue · 0 comments
jbrepogmailcom commented
Feature Description
It seems to me that there is no advantage in using GPU in autokeras. For example, with low-end PC during fit the step duration is 4ms, while with GPU for thousands of dollars (NVIDIA T4) the step duration is 3ms and GPU is used for 20-25%. It could be good if there could be some profiler that can optimize parameters to use GPU.
I also tried the code below, but it did not improve performance.
Code Example
###### My special code here ##############
config = tf.compat.v1.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.compat.v1.Session(config=config)
##########################################
reg = ak.StructuredDataRegressor(project_name=modelName)
##########################################
with tf.device('/gpu:0'):
##########################################
reg.fit(X_train, y_train, validation_split=vSplit)