Open
Description
Hello,
I set in config fixed learning rate and dropout and let PopulationBasedTraining do it's mutations.
Something like this:
hyperparameter_space = {
"lr": 1e-3,
"dense_1": tune.quniform(32, 128, 32),
"count_dense_layers": tune.quniform(1, 4, 1),
"dense_x": tune.quniform(32, 256, 32),
"dropout_rate": 0.0,
"tf_optimizer": tune.choice(["SGD", "Adam"]),
}
pbt_hyperparameter_space = {
"lr": tune.loguniform(1e-4, 1e-2),
"dropout_rate": tune.quniform(0.0, 0.3, 0.1), #
}
Algorithm is running fine, lr and dropout are tuned: config.lr=0.001346 and config.dropout_rate=0.1
Can the whole process of tuning with PopulationBasedTraining reflect also in Tensorboard?
As you can see bellow, the tensorboard show only values from hyperparameter_space.