debug_flag.DEBUG = False
model = make_model(n_vert_max, n_feat=n_feat, n_class=n_class)
model.load_weights(args.weights_path)
if args.input_type == 'h5':
from generators.h5 import make_dataset
elif args.input_type == 'root':
from generators.uproot_fixed import make_dataset
elif args.input_type == 'root-sparse':
from generators.uproot_jagged_keep import make_dataset
inputs, truth, _ = make_dataset(args.data_path,
features=features,
n_vert_max=n_vert_max,
n_sample=args.n_sample,
dataset_name=args.input_name)
n_sample = inputs[0].shape[0]
prob = model.predict(inputs, verbose=1)
if n_class == 2:
prob = np.squeeze(prob)
print(
'accuracy',
np.mean(
np.asarray(np.asarray(prob > 0.5, dtype=np.int32) == truth,
dtype=np.float32)))
else:
dataset_name=args.input_name)
fit_kwargs['validation_data'] = valid_gen
fit_kwargs['validation_steps'] = n_valid_steps
callbacks = [sparsity.UpdatePruningStep()]
prune_model.fit_generator(train_gen, **fit_kwargs, callbacks=callbacks)
else:
if args.input_type == 'h5':
from generators.h5 import make_dataset
elif args.input_type == 'root':
from generators.uproot_fixed import make_dataset
elif args.input_type == 'root-sparse':
from generators.uproot_jagged_keep import make_dataset
inputs, truth, shuffle = make_dataset(args.train_path[0],
features=features,
n_vert_max=n_vert_max,
dataset_name=args.input_name)
fit_kwargs = {
'epochs': args.num_epochs,
'batch_size': args.batch_size,
'shuffle': shuffle
}
if args.validation_path:
val_inputs, val_truth, _ = make_dataset(
args.validation_path[0],
format=input_format,
features=features,
n_vert_max=n_vert_max,
y_features=y_features,
dataset_name=args.input_name)