def run_all_on_path(in_path, simulator='INI'):
assert os.path.exists(os.path.join(
in_path, 'x_norm.npz')), 'Could not find x_norm.npz on path'
candidates = []
for filename in os.listdir(in_path):
if filename.endswith(
".h5"
) and 'parsed' not in filename and simulator not in filename:
candidates.append(filename)
print('Found a total of {} candidates.'.format(len(candidates)))
durations = []
for i, candidate in enumerate(candidates):
# strip the extension
candidate = candidate.split('.h5')[0]
print('Running candidate {}'.format(candidate))
config = generate_snn_config(in_path, candidate, simulator=simulator)
start_time = time.time()
main(config)
end_time = time.time()
duration = round(end_time - start_time, 2)
durations.append(duration)
print('Duration for candidate {}: {}'.format(candidate, duration))
mean_duration = sum(durations) / len(durations)
n_left = len(candidates[i + 1:])
pred_duration = mean_duration * n_left
print('Predicted duration for remaining candidates: {}'.format(
datetime.timedelta(seconds=pred_duration)))
# rename the log dir
shutil.move(os.path.join(in_path, 'log'),
os.path.join(in_path, candidate))
def setup():
config = configparser.ConfigParser()
config.read("tb/snn_toolbox.ini")
main("tb/snn_toolbox.ini")
h5_file = config["paths"]["filename_ann"] + "_INI.h5"
v_thresh = 0
with h5py.File(h5_file, 'r') as h5:
v_thresh = h5["model_weights"]["0Dense_4"]["v_thresh:0"].shape
# Takes the first threshold and runs with it
param_str = ""
if not v_thresh:
param_str = " -Psnn.THRESHOLD=5"
else:
param_str = " -Psnn.THRESHOLD=" + h5["model_weights"]["0Dense_2"][
"v_thresh:0"].shape[0]
param_str += " -Psnn.NUM_NODES=4"
if "COMPILE_ARGS" in os.environ:
os.environ["COMPILE_ARGS"] = os.environ["COMPILE_ARGS"] + param_str
else:
os.environ["COMPILE_ARGS"] = param_str
print(os.getcwd())
run_make = "make -f " + os.getcwd() + "/tb/tb.mk"
make_proc = subprocess.run(run_make, shell=True, stderr=subprocess.STDOUT)
'simulator': 'brian2', # Chooses execution backend of SNN toolbox.
'duration': 50, # Number of time steps to run each sample.
'num_to_test': 5, # How many test samples to run.
'batch_size': 1, # Batch size for simulation.
'dt': 0.1 # Time resolution for ODE solving.
}
config['input'] = {
'poisson_input': False # Images are encodes as spike trains.
}
config['output'] = {
'plot_vars': { # Various plots (slows down simulation).
'spiketrains', # Leave section empty to turn off plots.
'spikerates',
'activations',
'correlation',
'v_mem',
'error_t'}
}
# Store config file.
config_filepath = os.path.join(path_wd, 'config')
with open(config_filepath, 'w') as configfile:
config.write(configfile)
# RUN SNN TOOLBOX #
###################
main(config_filepath)
""" Script to run a pre-trained neural network for anomaly detection on Loihi, using SNN toolbox """
import argparse
from snntoolbox.bin.run import main
if __name__ == "__main__":
parser = argparse.ArgumentParser()
config_filepath = 'conversion_config_loihi.txt'
parser.add_argument("--model",
help="Path to model with weights in h5 format",
action='store',
dest='model',
default=config_filepath)
args = parser.parse_args()
main(args.model)
