def apply_finite_conductance_states_to_crossbar(crossbar,
num_conductance_states):
crossbar_min = torch.tensor(1 / (np.vectorize(lambda x: x.r_off)(
crossbar.devices))).view(-1).to(device).float()
crossbar_max = torch.tensor(1 / (np.vectorize(lambda x: x.r_on)(
crossbar.devices))).view(-1).to(device).float()
quantization.quantize(crossbar.conductance_matrix,
num_conductance_states, crossbar_min,
crossbar_max)
return crossbar
def quantization_delegate(image, channels, *args):
bins = get_quantization_bins()
quantized_channels = quantization.quantize(channels, bins)
if not any(bins):
bins = (255, 255, 255)
return quantized_channels, bins
def run(
self
): # Note: running an experiment on a model will leave its parameters unchanged after completion.
for experiment in self.experiments:
torch.save(experiment.model.state_dict(), self.BACKUP_PATH)
quantization.quantize(experiment.model,
experiment.parameter_types,
experiment.quantizer_type,
experiment.n,
base=experiment.base)
accuracy = utilities.testAccuracy(experiment.model,
experiment.test_ds)
print(experiment.id + " accuracy:", accuracy)
self.accuracies.append(accuracy)
self.visualizer.plot_value(accuracy, 1, "accuracy")
for parameter_type in experiment.parameter_types:
self.visualizer.plot_distribution(
experiment.model, self.visualizer.layout.plot_resolution,
experiment.id, parameter_type)
experiment.model.load_state_dict(torch.load(self.BACKUP_PATH))
if self.output_plots is True:
self.visualizer.output_plots()
os.remove(self.BACKUP_PATH)
def test_quantize(filename, dist):
im = Image.open(filename)
# different cases
expections = [128, 32, 8, 4, 2]
print "Quantiztion"
for level in expections:
out = quantize(im, level)
out_name = "quantization-%d.png" % level
out_path = os.path.join(dist, out_name)
out.save(out_path)
print " Picture %s has been saved to the assets folder !" % out_name
def main():
n_clusters = 5
dataset_path = join('datasets', 'adl')
X, labels = quantize(dataset_path, n_clusters)
y, encoder, classes = encode_labels(labels)
X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.7)
random_forest = RandomForestClassifier(tree_funcs=(learn_tree,
predict_tree),
n_trees=50,
max_depth=3,
min_leaf_size=5,
min_split_size=10,
feature_subset_size='sqrt')
random_forest.fit(X_train, y_train)
predictions = random_forest.predict(X_test)
acc = np.mean(predictions == y_test)
print(f'Test dataset predictions accuracy: {acc:2.2%}')
action='store',
dest='layer_size_hl1')
parser.add_argument('--layer_size_hl2',
action='store',
dest='layer_size_hl2')
parser.add_argument('--layer_size_hl3',
action='store',
dest='layer_size_hl3')
parser.add_argument('--learning_rate',
action='store',
dest='learning_rate')
parser.add_argument('--optimizer', action='store', dest='optimizer')
args = parser.parse_args()
wandb.init(project="amcpy-team",
config=HyperParameter(None).get_dict())
config = wandb.config
model_id = train_rna(config)
loaded_model, loaded_model_id = get_model_from_id(model_id)
evaluate_rna(loaded_model)
if not training:
loaded_model, _ = get_model_from_id(' ')
load_dict, info_dict = quantization.quantize(
loaded_model, np.concatenate((X_train, X_test)))
for info in info_dict:
print(info + ' -> ' + info_dict[info])
weights = load_dict['weights']
biases = load_dict['biases']
# serial_communication()
resize_randomness=Param.d['resize_random'],
**Param.d)
images = np.empty([len(image_dirs)] + list(Param.d['image_size']),
dtype=np.float32)
print('Loading representative images...', end=' ')
for i, (idir, ldir) in enumerate(zip(image_dirs, label_dirs)):
img, _ = train_set._load_function(idir, ldir)
images[i] = img
print('Done.')
print('')
(tflite_model_file,
tflite_model_quant_file) = quantization.quantize(model,
images,
ckpt_to_load,
Param.save_dir,
overwrite=overwrite,
saved_model=saved_model,
**Param.d)
if evaluate_models:
quantization.evaluate_quantized_model(tflite_model_file,
tflite_model_quant_file,
test_set,
evaluator,
num_processes=Param.d.get(
'num_parallel_calls', 4),
**Param.d)
if write_tensors:
quantization.write_tensors(tflite_model_file,
images[0],
tensor_list=None,
