def __init__(self, config):
tf_config = tf.ConfigProto()
tf_config.gpu_options.per_process_gpu_memory_fraction = 0.25
set_session(tf.Session(config=tf_config))
cuda.select_device(0)
self._verbose = config['verbose']
self._threshold = config['threshold']
if config['model'] not in list(ascii_lowercase)[:11]:
raise ValueError('Model only range from a -- j.')
else:
self._model = get_model(config['model'])
if config['gpu'] >= 0:
os.environ["CUDA_VISIBLE_DEVICES"] = str(config['gpu'])
else:
os.environ["CUDA_VISIBLE_DEVICES"] = ""
cands_to_eval = glob.glob(f'{args.data_dir}/*h5')
if len(cands_to_eval) == 0:
raise FileNotFoundError(f"No candidates to evaluate.")
logging.debug(f'Read {len(cands_to_eval)} candidates')
# Get the data generator, make sure noise and shuffle are off.
cand_datagen = DataGenerator(list_IDs=cands_to_eval,
labels=[0] * len(cands_to_eval),
shuffle=False,
noise=False,
batch_size=args.batch_size)
model = get_model(args.model)
# get's get predicting
probs = model.predict_generator(generator=cand_datagen,
verbose=1,
use_multiprocessing=use_multiprocessing,
workers=args.nproc,
steps=len(cand_datagen))
# Save results
results_dict = {}
results_dict['candidate'] = cands_to_eval
results_dict['probability'] = probs[:, 1]
results_dict['label'] = np.round(probs[:, 1] >= args.probability)
results_file = args.data_dir + f'/results_{args.model}.csv'
pd.DataFrame(results_dict).to_csv(results_file, index=False)
)
data_df = pd.read_csv(args.data_csv)
train_df, val_df = train_test_split(data_df,
test_size=(1 - args.val_split),
random_state=1993)
train_data_generator = DataGenerator(list_IDs=list(train_df['h5']),
labels=list(train_df['label']),
noise=True,
shuffle=True)
validate_data_generator = DataGenerator(list_IDs=list(val_df['h5']),
labels=list(val_df['label']),
noise=False,
shuffle=False)
model_to_train = get_model(args.model)
model_to_train = ready_for_train(model_to_train,
ndt=args.n_dt_layers,
nft=args.n_ft_layers,
nf=args.n_fusion_layers)
trained_model, history = train(model_to_train,
epochs=args.epochs,
patience=args.patience,
output_path=args.output_path,
nproc=args.nproc,
train_obj=train_data_generator,
val_obj=validate_data_generator)
