def test():
"""
Execute the tests
Returns: (n_total, n_success)
"""
logger = TestLogger(TESTNAME)
# load net and the data
model = GoldenModel(CONFIG_FILENAME,
NET_FILENAME,
no_scale_between_l1_l2=True)
data = dict(np.load(DATA_FILENAME))
net = dict(np.load(NET_FILENAME))
for i, layer in enumerate(model.layers):
result = test_layer(layer, data)
test_index = i + 2
if i == 0:
test_index = "1+2"
logger.show_subcase_result("Layer {}".format(test_index), result)
result = test_model(model, data)
logger.show_subcase_result("Model", result)
# return summary
return logger.summary()
def gen_stimuli():
"""
This function generates the stimuli (input and output) for the test
"""
model = GoldenModel(CONFIG_FILENAME, NET_FILENAME, clip_balanced=False)
x = np.random.randint(-60, 60, (model.F2, model.T // 8)).astype(int)
x_align = align_array(x)
y_exp = np.transpose(x)
y_exp_align = np.transpose(x_align)
y_exp_align = align_array(y_exp_align)
return x, x_align, y_exp, y_exp_align
def gen_stimuli(random_input):
"""
This function generates the stimuli (input and output) for the test
"""
model = GoldenModel(CONFIG_FILENAME, NET_FILENAME, clip_balanced=False)
if random_input:
x = np.random.randint(-60, 60, (model.C, model.T))
else:
x = np.load(INPUT_FILENAME)["input"][0, :, :]
x = F.quantize_to_int(x, model.input_scale)
y_exp = model.layers[0](x)
return x, y_exp
def gen_stimuli(random_input, no_div=False, pad_data=False):
"""
This function generates the stimuli (input and output) for the test
"""
if no_div:
model = GoldenModel(CONFIG_FILENAME,
NET_FILENAME,
clip_balanced=False,
no_scale_between_l1_l2=True)
layer = model.layers[0]
if random_input:
x = np.random.randint(-60, 60, (model.C, model.T))
else:
x = np.load(INPUT_FILENAME)["input"][0, :, :]
x = F.quantize_to_int(x, layer.input_scale)
y_exp = layer(x)
else:
model = GoldenModel(CONFIG_FILENAME, NET_FILENAME, clip_balanced=False)
layer1 = model.layers[0]
layer2 = model.layers[1]
if random_input:
x = np.random.randint(-60, 60, (model.C, model.T))
else:
x = np.load(INPUT_FILENAME)["input"][0, :, :]
x = F.quantize_to_int(x, layer1.input_scale)
y_exp = layer2(layer1(x))
y_exp_align = align_array(y_exp)
if pad_data:
C, T = x.shape
T_pad = T + 63
assert T_pad % 4 == 0
x_pad = np.zeros((C, T_pad), dtype=np.int)
x_pad[:, 31:31 + T] = x
return x, x_pad, y_exp, y_exp_align
else:
x_align = align_array(x)
return x, x_align, y_exp, y_exp_align
def gen_stimuli():
"""
This function generates the stimuli (input and output) for the test
"""
model = GoldenModel(CONFIG_FILENAME, NET_FILENAME, clip_balanced=False)
x = np.random.randint(-60, 60, (model.F1, model.C, model.T)).astype(int)
x_align = np.zeros(
(model.F1, align_array_size(model.C), align_array_size(model.T)))
x_align = x_align.astype(int)
x_align[:, :model.C, :model.T] = x
y_exp = np.transpose(x, (0, 2, 1))
y_exp_align = np.transpose(x_align, (0, 2, 1))
return x, x_align, y_exp, y_exp_align
def gen_stimuli(random_input):
"""
This function generates the stimuli (input and output) for the test
"""
model = GoldenModel(CONFIG_FILENAME, NET_FILENAME, clip_balanced=False)
layer = model.layers[2]
if random_input:
x = np.random.randint(-60, 60, (model.F2, model.T // 8))
else:
x = np.load(INPUT_FILENAME)["layer2_activ"][0, :, 0, :]
x = F.quantize_to_int(x, layer.input_scale)
y_exp = layer(x)
x_align = align_array(x)
y_exp_align = align_array(y_exp)
return x, x_align, y_exp, y_exp_align
def gen_stimuli(random_input, flip, reorder_bn):
"""
This function generates the stimuli (input and output) for the test
"""
model = GoldenModel(CONFIG_FILENAME, NET_FILENAME, clip_balanced=False, reorder_bn=reorder_bn)
layer = model.layers[1]
if random_input:
x = np.random.randint(-60, 60, (model.F1, model.C, model.T))
else:
x = np.load(INPUT_FILENAME)["layer1_activ"][0, :, :, :]
x = F.quantize_to_int(x, layer.input_scale)
y_exp = layer(x)
if flip:
x_flip = np.transpose(x, (0, 2, 1))
x_align = np.zeros((model.F1, align_array_size(model.T), align_array_size(model.C)), dtype=int)
x_align[:, :model.T, :model.C] = x_flip
else:
x_align = align_array(x)
y_exp_align = align_array(y_exp)
return x, x_align, y_exp, y_exp_align