def _config(_path_wd, _datapath):
path_wd = str(_path_wd)
datapath = str(_datapath)
filename_ann = 'mnist_cnn'
configparser = import_configparser()
config = configparser.ConfigParser()
config.read_dict({
'paths': {
'path_wd': path_wd,
'dataset_path': datapath,
'filename_ann': filename_ann
}
})
with open(os.path.join(path_wd, filename_ann + '.h5'), 'w'):
pass
config_filepath = os.path.join(path_wd, 'config')
with open(config_filepath, 'w') as configfile:
config.write(configfile)
config = update_setup(config_filepath)
return config
def mk_config(path, fold):
"create config file given path to a trained model"
configparser = import_configparser()
config = configparser.ConfigParser()
config['paths'] = {
'path_wd': path + "/" + fold, # Path to model.
'dataset_path': path + "/" + fold, # Path to dataset.
'filename_ann': "ts" + fold # Name of input model.
}
config['tools'] = {
'evaluate_ann': True, # Test ANN on dataset before conversion.
'normalize': True, # Normalize weights for full dynamic range.
'simulate':
True # Simulate model, seems to be necessary for normalization
}
config['simulation'] = {
'simulator': 'nest', # We convert to a pynn model with nest as backend
'duration': 50, # Number of time steps to run each sample.
'num_to_test': 1, # How many test samples to run.
'batch_size': 1, # Batch size for simulation.
'dt': 0.1 # timestep
}
config['input'] = {
'poisson_input': True, # Images are encodes as spike trains.
'input_rate': 1000 # Poisson Neurons firing rate
}
config['cell'] = {
'v_thresh': 0.01, # voltage threshold, different for actual SCNN
'tau_refrac': 0.1, # refractory period
'delay': 0.1 # synaptic delay
}
return config
def generate_snn_config(path_wd, model_name, simulator='INI'):
# Store model so SNN Toolbox can find it.
# SNN TOOLBOX CONFIGURATION #
#############################
# Create a config file with experimental setup for SNN Toolbox.
configparser = import_configparser()
config = configparser.ConfigParser()
config['paths'] = {
'path_wd': path_wd, # Path to model.
'dataset_path': path_wd, # Path to dataset.
'filename_ann': model_name # Name of input model.
}
config['tools'] = {
'evaluate_ann': True, # Test ANN on dataset before conversion.
'normalize': True, # Normalize weights for full dynamic range.
}
config['simulation'] = {
'simulator': simulator, # Chooses execution backend of SNN toolbox.
'duration': 2000, # Number of time steps to run each sample.
'num_to_test': 10000, # How many test samples to run.
'batch_size': 256, # Batch size for simulation.
'keras_backend': 'tensorflow'
}
config['output'] = {
# log desired variables to disk
'log_vars': {
'synaptic_operations_b_t', 'neuron_operations_b_t'
}, #{'spiketrains_n_b_l_t', 'synaptic_operations_b_t', 'neuron_operations_b_t'},
# 'plot_vars': { # Various plots (slows down simulation).
# 'spiketrains', # Leave section empty to turn off plots.
# 'spikerates',
# 'activations',
# 'correlation',
# 'v_mem',
# 'error_t'}
'verbose': 0
}
if simulator is 'loihi':
config['loihi'] = loihi_config_dict
# loihi does not support batching
config.set('simulation', 'batch_size', '1')
# config['simulation']['batch_size'] = 1
# Store config file.
config_filepath = os.path.join(path_wd, model_name) + '.config'
with open(config_filepath, 'w') as configfile:
config.write(configfile)
return config_filepath
def test_updating_settings(params, expect_pass, _path_wd):
configparser = import_configparser()
config = configparser.ConfigParser()
config.read_dict(params)
configpath = os.path.join(str(_path_wd), 'config')
with open(configpath, 'w') as file:
config.write(file)
if expect_pass:
assert update_setup(configpath)
else:
pytest.raises(AssertionError, update_setup, configpath)
def load_config(filepath):
"""
Load a config file from ``filepath``.
"""
from snntoolbox.utils.utils import import_configparser
configparser = import_configparser()
assert os.path.isfile(filepath), \
"Configuration file not found at {}.".format(filepath)
config = configparser.ConfigParser()
config.optionxform = str
config.read(filepath)
return config
model.compile('adam', 'categorical_crossentropy', ['accuracy'])
# Train model with backprop.
model.fit(x_train, y_train, batch_size=64, epochs=1, verbose=2,
validation_data=(x_test, y_test))
# Store model so SNN Toolbox can find it.
model_name = 'mnist_cnn'
keras.models.save_model(model, os.path.join(path_wd, model_name + '.h5'))
# SNN TOOLBOX CONFIGURATION #
#############################
# Create a config file with experimental setup for SNN Toolbox.
configparser = import_configparser()
config = configparser.ConfigParser()
config['paths'] = {
'path_wd': path_wd, # Path to model.
'dataset_path': path_wd, # Path to dataset.
'filename_ann': model_name # Name of input model.
}
config['tools'] = {
'evaluate_ann': True, # Test ANN on dataset before conversion.
'normalize': True, # Normalize weights for full dynamic range.
}
config['simulation'] = {
'simulator': 'brian2', # Chooses execution backend of SNN toolbox.
