def run_predict(self, args):
"""Predict pathogenic potentials from a fasta/npy file."""
if args.tpu is None:
config_cpus(args.n_cpus)
config_gpus(args.gpus)
if args.output is None:
args.output = os.path.splitext(args.input)[0] + "_predictions.npy"
if args.sensitive:
model = self.bloader.load_sensitive_model(training_mode=False, tpu_resolver=self.tpu_resolver)
elif args.rapid:
model = self.bloader.load_rapid_model(training_mode=False, tpu_resolver=self.tpu_resolver)
else:
if self.tpu_resolver is not None:
tpu_strategy = tf.distribute.experimental.TPUStrategy(self.tpu_resolver)
with tpu_strategy.scope():
model = tf.keras.models.load_model(args.custom)
else:
model = tf.keras.models.load_model(args.custom)
if args.rc_check:
compare_rc(model, args.input, args.output, args.plot_kind, args.alpha, replicates=args.replicates,
batch_size=args.batch_size)
elif args.array:
predict_npy(model, args.input, args.output, replicates=args.replicates, batch_size=args.batch_size)
else:
predict_fasta(model, args.input, args.output, args.n_cpus, replicates=args.replicates,
batch_size=args.batch_size)
def run_train(self, args):
"""Parse the config file and train the NN on Illumina reads."""
if args.tpu is None:
config_cpus(args.n_cpus)
config_gpus(args.gpus)
if args.sensitive:
paprconfig = self.bloader.get_sensitive_training_config()
elif args.rapid:
paprconfig = self.bloader.get_rapid_training_config()
else:
config = configparser.ConfigParser()
config.read(args.custom)
paprconfig = RCConfig(config)
if args.train_data:
paprconfig.x_train_path = args.train_data
if args.train_labels:
paprconfig.y_train_path = args.train_labels
if args.val_data:
paprconfig.x_val_path = args.val_data
if args.val_labels:
paprconfig.y_val_path = args.val_labels
if args.run_name:
paprconfig.runname = args.run_name
paprconfig.log_dir = os.path.join(paprconfig.log_superpath,
"{runname}-logs".format(runname=paprconfig.runname))
paprconfig.set_tpu_resolver(self.tpu_resolver)
paprnet = RCNet(paprconfig)
paprnet.load_data()
paprnet.compile_model()
paprnet.train()
def run_tester(args):
tpu_resolver = global_setup(args)
if args.tpu is None:
n_cpus = config_cpus(args.n_cpus_rec)
config_gpus(args.gpus)
else:
n_cpus = args.n_cpus_rec
if args.custom:
args.command = None
run_tests(args.command, args.model, n_cpus, args.keep, args.scale,
tpu_resolver)
def run_tests(self, args):
"""Run tests."""
if args.tpu is None:
n_cpus = config_cpus(args.n_cpus)
n_gpus = config_gpus(args.gpus)
scale = args.scale * max(1, n_gpus)
else:
n_cpus = multiprocessing.cpu_count()
scale = args.scale
tester = Tester(n_cpus, self.builtin_configs, self.builtin_weights,
args.explain, args.gwpa, args.all, args.quick, args.keep, scale,
tpu_resolver=self.tpu_resolver, input_modes=args.input_modes,
additivity_check=(not args.no_check), large=args.large)
tester.run_tests()
def run_receiver(args):
tpu_resolver = global_setup(args)
if args.tpu is None:
n_cpus = config_cpus(args.n_cpus_rec)
config_gpus(args.gpus)
else:
n_cpus = args.n_cpus_rec
if args.custom:
args.command = None
receiver = Receiver(args.command,
model=args.model,
read_length=args.read_length,
input_dir=args.rec_in_dir,
output_dir=args.rec_out_dir,
n_cpus=n_cpus,
threshold=args.threshold,
tpu_resolver=tpu_resolver)
cycles = [int(c) for c in args.cycle_list.split(',')]
barcodes = args.barcodes.split(',')
receiver.run(cycles=cycles,
barcodes=barcodes,
mode=args.format,
discard_neg=args.discard_neg)