def test_safe_idx_sequence_from_idx(data,
idx=None,
data_seq=1,
target_seq=0,
target_shift=0):
safe_idx = get_safe_idx(data,
idx=idx,
data_seq=data_seq,
target_seq=target_seq,
target_shift=target_shift)
print_ndarray(
f'safe_idx = get_safe_idx({np.shape(data)}, idx={idx}, data_seq={data_seq}, target_seq={target_seq}, target_shift={target_shift})',
safe_idx)
x, y = sequence_from_idx(data,
data,
safe_idx,
data_seq=data_seq,
target_seq=target_seq,
target_shift=target_shift)
print_ndarray(
f'x, _ = sequence_from_idx({np.shape(data)}, {np.shape(data)}, safe_idx, data_seq={data_seq}, target_seq={target_seq}, target_shift={target_shift})',
x)
print_ndarray(
f'_, y = sequence_from_idx({np.shape(data)}, {np.shape(data)}, safe_idx, data_seq={data_seq}, target_seq={target_seq}, target_shift={target_shift})',
y)
pass # test_safe_idx_batch_from_seq()
def test_prep(data,
config,
idx=[28799, 35999],
data_seq=0,
target_seq=0,
target_shift=0):
data.set_params(data_seq=data_seq,
target_seq=target_seq,
target_shift=target_shift)
print(data)
x, y = data.get_dataset(idx)
print_ndarray('\nx = data.get_dataset({})'.format(idx),
x,
10,
frm='8.3f')
print_ndarray('y = data.get_dataset({})\n'.format(idx),
y,
10,
frm='8.3f')
nbins = 15
p_range = [5, 95]
treshold = 0.050
if not tf.executing_eagerly():
data_shape = np.shape(x)[1:]
model = Preproces.build(data_shape,
name=model_name,
config=config,
nbins=nbins,
p_range=p_range,
treshold=treshold)
logdir = 'C:\logs'
tf.compat.v1.summary.FileWriter(
logdir, graph=tf.compat.v1.get_default_graph()).close()
assert False, 'write graph'
output = prep(x,
config,
nbins=nbins,
p_range=p_range,
treshold=treshold)
print_ndarray('\noutput\n', output, 0) # (2,16,16,36)
return output
if __name__ == "__main__":
import numpy as np
from mylib import print_ndarray
print('\ntensorflow helper functions')
print('tensorflow version: {0}'.format(tf.__version__))
print('tf.executing_eagerly(): {0}'.format(tf.executing_eagerly()))
# define test random array
np.random.seed(seed=111)
mu = np.array([0, 2, 5, -5], dtype=np.float32).reshape(-1, 1)
mu = np.array([0, 0, 0, 0], dtype=np.float32).reshape(-1, 1)
sigma = np.array([1, 2, 3, 3], dtype=np.float32).reshape(-1, 1)
logvar = np.log(np.square(sigma))
print_ndarray('mu', mu)
print_ndarray('sigma', sigma)
a_shape = (4, 11000)
n = a_shape[0]
a = np.random.normal(size=a_shape) * sigma + mu
print_ndarray('a = np.random.normal(size={}) * sigma + mu'.format(a.shape),
a)
a = np.cumsum(a, axis=1)
print_ndarray('a = np.cumsum(a, np.float32)'.format(a.shape),
a,
frm='8.2f')
q = [25, 50, 75]
percentile = np.percentile(a, q, axis=-1)
print(details['device_name'])
print('---\n')
input_shape = (16, 16, 36)
config = get_mixer_s4_config()
if not tf.executing_eagerly():
logdir = 'C:\logs'
m = MlpMix2.build(input_shape, config, name='test_model')
tf.compat.v1.summary.FileWriter(
logdir, graph=tf.compat.v1.get_default_graph()).close()
assert False, 'write graph'
m = MlpMix2.build(input_shape, config)
size = (1024, 16, 16, 36)
x = np.random.normal(size=size)
print_ndarray(f'x = np.random.normal(size={size})', x)
log_prob, value, feature = m(x)
print_ndarray('log_prob', log_prob)
print_ndarray('value', value)
print_ndarray(
'feature, mean, var',
np.concatenate([
feature,
np.mean(feature, axis=-1, keepdims=True),
np.var(feature, axis=-1, keepdims=True)
],
axis=-1))
tf.math.exp(hard_positives - hard_negatives))
else:
triplet_loss = tf.maximum(hard_positives - hard_negatives + margin,
0.0)
return triplet_loss # tf.reduce_mean(triplet_loss)
if __name__ == "__main__":
import numpy as np
from mylib import print_ndarray
size = (128, 128)
labels = np.random.randint(0, high=2, size=size[0])
feature = np.random.normal(loc=1, scale=1, size=size)
print_ndarray(f'\nlabels (y_true)', labels)
print_ndarray(f'feature (y_pred)', feature)
angular_dist = angular_distance(feature)
print_ndarray(f'angular_dist', angular_dist)
pairwise_dist = pairwise_distance(feature, squared=False)
print_ndarray(f'pairwise_dist', pairwise_dist)
pairwise_dist = pairwise_distance(feature,
squared=False,
L2_normalize=True)
print_ndarray(f'pairwise_dist', pairwise_dist)
t = triplet_semihard_loss(
labels,
output = test_prep(dr_train,
config256,
idx=idx,
data_seq=256,
target_seq=0,
target_shift=0)
for image in output:
draw_img(image, n_cols=9, n_rows=4)
# test prepare_data()
x, y = dr_train.get_dataset(idx,
data_seq=7200,
target_seq=0,
target_shift=0)
print_ndarray('\nx = .get_dataset({})'.format(idx), x, 20, frm='8.3f')
y = x[0, :, :4]
a = prepare_data(y, config256)
print_ndarray(' = .prepare_data({})'.format(np.shape(y)),
a,
20,
frm='8.3f')
#data_list = ['Open', 'High', 'Low', 'Close', 'AD1', 'AH4', 'AH1', 'AM15', 'AM5', 'LVW1', 'LVD1', 'LVH4', 'LVH1', 'LVM15', 'LVM5', 'LVM2', 'Idx']
config1024 = get_prep1024_config(data_file='hilo66_logvar_2020.csv',
target_file='reward_100_2020~.csv')
dr_train = get_data(config1024, data_path=data_path)
output = test_prep(dr_train,
config1024,
target_seq=target_seq,
target_shift=target_shift)
print_ndarray(
f'x, _ = sequence_from_idx({np.shape(data)}, {np.shape(data)}, safe_idx, data_seq={data_seq}, target_seq={target_seq}, target_shift={target_shift})',
x)
print_ndarray(
f'_, y = sequence_from_idx({np.shape(data)}, {np.shape(data)}, safe_idx, data_seq={data_seq}, target_seq={target_seq}, target_shift={target_shift})',
y)
pass # test_safe_idx_batch_from_seq()
size = 12
data = np.arange(size, dtype=np.float32)
data = np.reshape(data, (-1, 1))
data = np.concatenate([data + 0.1, data + 0.2], axis=-1)
print_ndarray('data', data)
test_safe_idx_sequence_from_idx(data,
idx=None,
data_seq=0,
target_seq=0,
target_shift=0)
test_safe_idx_sequence_from_idx(data,
idx=None,
data_seq=1,
target_seq=0,
target_shift=0)
test_safe_idx_sequence_from_idx(data,
idx=None,
data_seq=2,
target_seq=0,
if __name__ == "__main__":
import numpy as np
from mylib import print_ndarray
print('\nnumpy helper functions')
print('numpy version: {0}'.format(np.__version__))
# define test random array
np.random.seed(seed=111)
mu = np.array([0, 2, 5, -5], dtype=np.float32).reshape(-1, 1)
mu = np.array([0, 0, 0, 0], dtype=np.float32).reshape(-1, 1)
sigma = np.array([1, 2, 3, 3], dtype=np.float32).reshape(-1, 1)
logvar = np.log(np.square(sigma))
print_ndarray('mu', mu)
print_ndarray('sigma', sigma)
a_shape = (4, 11000)
n = a_shape[0]
a = np.random.normal(size=a_shape) * sigma + mu
a = np.asarray(a, dtype=np.float32)
print_ndarray('a = np.random.normal(size={}) * sigma + mu'.format(a.shape),
a)
a = np.cumsum(a, axis=1)
print_ndarray('a = np.cumsum(a, np.float32)'.format(a.shape),
a,
frm='8.2f')
q = [25, 50, 75]