def write_ppc_file(fname):
"""Save PYPOWER file.
"""
filename = fname
base = os.path.basename(fname)
casename = os.path.splitext(base)[0]
outfile = open(fname, 'w', newline='')
outfile.write('from numpy import array\n\n')
outfile.write('def ' + casename + '():\n')
outfile.write('\tppc = {"version": ''2''}\n')
outfile.write('\tppc["baseMVA"] = 100.0\n')
outfile.write('\tppc["bus"] = ')
outfile.write(np.array_repr(ppc["bus"]))
outfile.write('\n\n')
outfile.write('\tppc["gen"] = ')
outfile.write(np.array_repr(ppc["gen"]))
outfile.write('\n\n')
outfile.write('\tppc["branch"] = ')
outfile.write(np.array_repr(ppc["branch"]))
outfile.write('\n\n')
outfile.write('\treturn ppc')
outfile.close()
return True
def __repr__(self):
# if numpy.all(self == 0):
# # Bin-only output
# return "{}(bins={})".format(type(self).__name__, numpy.array_repr(self._bins))
# else:
if self.ndim == 1:
return "{}({}, data={})".format(type(self).__name__,
numpy.array_repr(self._bins)[len("array("):-1],
numpy.array_repr(self)[len(type(self).__name__)+1:-1])
else:
return "{}(({}), data={})".format(type(self).__name__,
",".join([numpy.array_repr(x)[6:-1] for x in self._bins]),
numpy.array_repr(self)[len(type(self).__name__)+1:-1])
def test():
from sklearn.metrics import mean_squared_error
import Surrogates.DataExtraction.pcs_parser as pcs_parser
sp = pcs_parser.read(file("/home/eggenspk/Surrogates/Data_extraction/Experiments2014/hpnnet/smac_2_06_01-dev/nips2011.pcs"))
# Read data from csv
header, data = read_csv("/home/eggenspk/Surrogates/Data_extraction/hpnnet_nocv_convex_all/hpnnet_nocv_convex_all_fastrf_results.csv",
has_header=True, num_header_rows=3)
para_header = header[0][:-2]
type_header = header[1]
cond_header = header[2]
#print data.shape
checkpoint = hash(numpy.array_repr(data))
assert checkpoint == 246450380584980815
model = GaussianProcess(sp=sp, encode=False, rng=1, debug=True)
x_train_data = data[:100, :-2]
y_train_data = data[:100, -1]
x_test_data = data[100:, :-2]
y_test_data = data[100:, -1]
model.train(x=x_train_data, y=y_train_data, param_names=para_header)
y = model.predict(x=x_train_data[1, :])
print "Is: %100.70f, Should: %f" % (y, y_train_data[1])
assert y[0] == 0.470745153514900149804844886602950282394886016845703125
print "Predict whole data"
y_whole = model.predict(x=x_test_data)
mse = mean_squared_error(y_true=y_test_data, y_pred=y_whole)
print "MSE: %100.70f" % mse
assert mse == 0.006257598609004190459703664828339242376387119293212890625
print "Soweit so gut"
# Try the same with encoded features
model = GaussianProcess(sp=sp, encode=True, rng=1, debug=True)
#print data[:10, :-2]
model.train(x=x_train_data, y=y_train_data, param_names=para_header)
y = model.predict(x=x_train_data[1, :])
print "Is: %100.70f, Should: %f" % (y, y_train_data[1])
assert y[0] == 0.464671665294324409689608046392095275223255157470703125
print "Predict whole data"
y_whole = model.predict(x=x_test_data)
mse = mean_squared_error(y_true=y_test_data, y_pred=y_whole)
print "MSE: %100.70f" % mse
assert mse == 0.00919265128042330570412588031103950925171375274658203125
assert hash(numpy.array_repr(data)) == checkpoint
def myf(x):
header = 'MaxStepX, TorsoWy, TorsoWx, StepHeight, Stiffness, MaxStepTheta, MaxStepY, MaxStepFrequency'
print header
print np.array_repr(x[0]).replace('\n', '').replace('\t', '')
speed = input('Input 1/average_speed = ')
output = np.array(float(speed))
with open("data_new.py",'a') as f:
np.savetxt(f, x, delimiter=",")
# for item in x:
# f.write("%s\n" % str(np.array_repr(item).replace('\n', '').replace('\t', '')))
with open("readings_new.py",'a') as f:
f.write("%s\n" % str(output))
return output
def test():
from sklearn.metrics import mean_squared_error
import Surrogates.DataExtraction.pcs_parser as pcs_parser
sp = pcs_parser.read(file("/home/eggenspk/Surrogates/Data_extraction/Experiments2014/hpnnet/smac_2_06_01-dev/nips2011.pcs"))
# Read data from csv
header, data = read_csv("/home/eggenspk/Surrogates/Data_extraction/hpnnet_nocv_convex_all/hpnnet_nocv_convex_all_fastrf_results.csv",
has_header=True, num_header_rows=3)
para_header = header[0][:-2]
type_header = header[1]
cond_header = header[2]
#print data.shape
checkpoint = hash(numpy.array_repr(data))
assert checkpoint == 246450380584980815
model = GradientBoosting(sp=sp, encode=False, debug=True)
x_train_data = data[:1000, :-2]
y_train_data = data[:1000, -1]
x_test_data = data[1000:, :-2]
y_test_data = data[1000:, -1]
model.train(x=x_train_data, y=y_train_data, param_names=para_header, rng=1)
y = model.predict(x=x_train_data[1, :])
print "Is: %100.70f, Should: %f" % (y, y_train_data[1])
assert y[0] == 0.45366000254662230961599789225147105753421783447265625
print "Predict whole data"
y_whole = model.predict(x=x_test_data)
mse = mean_squared_error(y_true=y_test_data, y_pred=y_whole)
print "MSE: %100.70f" % mse
assert mse == 0.00188246958253847243396073007914992558653466403484344482421875
print "Soweit so gut"
# Try the same with encoded features
model = GradientBoosting(sp=sp, encode=True, debug=True)
#print data[:10, :-2]
model.train(x=x_train_data, y=y_train_data, param_names=para_header, rng=1)
y = model.predict(x=x_train_data[1, :])
print "Is: %100.70f, Should: %f" % (y, y_train_data[1])
assert y[0] == 0.460818965082699205648708584703854285180568695068359375
print "Predict whole data"
y_whole = model.predict(x=x_test_data)
mse = mean_squared_error(y_true=y_test_data, y_pred=y_whole)
print "MSE: %100.70f" % mse
assert mse == 0.002064362783199560034963493393433964229188859462738037109375
assert hash(numpy.array_repr(data)) == checkpoint
def testTransformerFeedForwardLayer(self):
with self.session(use_gpu=True) as sess:
tf.set_random_seed(3980847392)
inputs = tf.random_normal([5, 2, 3], seed=948387483)
paddings = tf.zeros([5, 2])
p = layers_with_attention.TransformerFeedForwardLayer.Params()
p.name = 'transformer_fflayer'
p.input_dim = 3
p.hidden_dim = 7
transformer_fflayer = layers_with_attention.TransformerFeedForwardLayer(p)
h = transformer_fflayer.FPropDefaultTheta(inputs, paddings)
tf.global_variables_initializer().run()
actual_layer_output = sess.run(h)
# pylint: disable=bad-whitespace
# pyformat: disable
expected_output = [
[[-0.88366592, -0.05049637, 0.01003706],
[-0.10550675, 1.68050027, 2.29110384]],
[[-1.30083609, -0.40521634, 0.1911681 ],
[ 1.2597878 , 1.45850968, 1.58734488]],
[[ 0.10373873, -0.2716777 , 0.2314173 ],
[ 0.46293864, -0.06359965, 1.20189023]],
[[ 0.3673597 , -0.1691664 , 0.78656065],
[-1.51081395, -0.70281881, -0.9093715 ]],
[[-1.04800868, -0.70610946, -0.35321558],
[-1.92480004, 0.08361804, 0.62713993]]]
# pyformat: enable
# pylint: enable=bad-whitespace
print(np.array_repr(actual_layer_output))
self.assertAllClose(actual_layer_output, expected_output)
def __str__(self):
""" prints compressed_data object content
"""
output = 'Data: \n'
output += array_repr(self.data[:], precision=3, suppress_small=True)
output += '\n Compression level ' + str(self._compression_level) + '\n'
return output
def setUp(self):
self._sp = pcs_parser.read(file(os.path.join(os.path.dirname(os.path.realpath(__file__)), "Testdata/nips2011.pcs")))
# Read data from csv
header, self._data = read_csv(os.path.join(os.path.dirname(os.path.realpath(__file__)), "Testdata/hpnnet_nocv_convex_all_fastrf_results.csv"),
has_header=True, num_header_rows=3)
self._para_header = header[0][:-2]
self._checkpoint = hash(numpy.array_repr(self._data))
def testSpectrumAugmenterWithFreqWarping(self):
with self.session(use_gpu=False, graph=tf.Graph()):
tf.random.set_seed(1234)
inputs = tf.broadcast_to(
tf.cast(tf.range(8), dtype=tf.float32), (5, 1, 8))
inputs = tf.expand_dims(inputs, -1)
paddings = tf.zeros([3, 2])
p = spectrum_augmenter.SpectrumAugmenter.Params()
p.name = 'specAug_layers'
p.freq_mask_max_bins = 0
p.time_mask_max_frames = 0
p.freq_warp_max_bins = 4
p.time_warp_max_frames = 0
p.random_seed = 345678
specaug_layer = p.Instantiate()
# pyformat: disable
# pylint: disable=bad-whitespace,bad-continuation
expected_output = np.array(
[[[0.0, 4.0, 4.5714283, 5.142857, 5.714286, 6.285714, 6.8571434,
3.999998]],
[[0.0, 0.8, 1.6, 2.4, 3.2, 4.0, 5.3333335, 6.6666665]],
[[0.0, 0.6666667, 1.3333334, 2.0, 3.2, 4.4, 5.6000004, 6.8]],
[[0.0, 1.3333334, 2.6666667, 4.0, 4.8, 5.6000004, 6.3999996,
5.5999947]],
[[0.0, 2.0, 2.857143, 3.7142859, 4.571429, 5.4285717, 6.2857146,
5.999997]]])
# pylint: enable=bad-whitespace,bad-continuation
# pyformat: enable
h, _ = specaug_layer.FPropDefaultTheta(inputs, paddings)
actual_layer_output = self.evaluate(tf.squeeze(h, -1))
print(np.array_repr(actual_layer_output))
self.assertAllClose(actual_layer_output, expected_output)
def testSpectrumAugmenterWarpMatrixConstructor(self):
with self.session(use_gpu=False, graph=tf.Graph()):
inputs = tf.broadcast_to(tf.cast(tf.range(10), dtype=tf.float32), (4, 10))
origin = tf.cast([2, 4, 4, 5], dtype=tf.float32)
destination = tf.cast([3, 2, 6, 8], dtype=tf.float32)
choose_range = tf.cast([4, 8, 8, 10], dtype=tf.float32)
p = spectrum_augmenter.SpectrumAugmenter.Params()
p.name = 'specAug_layers'
specaug_layer = p.Instantiate()
# pyformat: disable
# pylint: disable=bad-whitespace,bad-continuation
expected_output = np.array(
[[0.0000000, 0.6666667, 1.3333333, 2.0000000, 4.0000000,
5.0000000, 6.0000000, 7.0000000, 8.0000000, 9.0000000],
[0.0000000, 2.0000000, 4.0000000, 4.6666667, 5.3333333,
6.0000000, 6.6666667, 7.3333333, 8.0000000, 9.0000000],
[0.0000000, 0.6666667, 1.3333333, 2.0000000, 2.6666667,
3.3333333, 4.0000000, 6.0000000, 8.0000000, 9.0000000],
[0.0000000, 0.6250000, 1.2500000, 1.8750000, 2.5000000,
3.1250000, 3.7500000, 4.3750000, 5.0000000, 7.5000000]])
# pylint: enable=bad-whitespace,bad-continuation
# pyformat: enable
warp_matrix = specaug_layer._ConstructWarpMatrix(
batch_size=4,
matrix_size=10,
origin=origin,
destination=destination,
choose_range=choose_range,
dtype=tf.float32)
outputs = tf.einsum('bij,bj->bi', warp_matrix, inputs)
actual_layer_output = self.evaluate(outputs)
print(np.array_repr(actual_layer_output))
self.assertAllClose(actual_layer_output, expected_output)
def __init__(self,q,x,f,qdesc="q",xdesc="x",fdesc="f"):
"""
q ... 1D-array of shape (q.size)
x ... 2D-array of shape (q.size,x.size)
f ... 2D-array of shape (q.size,x.size)
*desc... description string for q,x,f
"""
self.f = np.asarray(f,dtype='float');
self.x = np.asarray(x,dtype='float');
self.q = np.asarray(q,dtype='float');
self.fdesc = fdesc;
self.xdesc = xdesc;
self.qdesc = qdesc;
# complete input data (if constant for different q)
if (len(self.x.shape)==1):
self.x=np.tile(self.x,(len(self.q),1));
if (len(self.f.shape)==1):
self.f=np.tile(self.f,(len(self.q),1));
# test shape of input data
if (self.q.shape[0] <> self.x.shape[0]) or \
(self.x.shape <> self.f.shape):
raise ValueError("Invalid shape of arguments.");
# test for double parameters
if (np.unique(self.q).size < self.q.size):
raise ValueError("Parameters are not unique: \n " + np.array_repr(np.sort(self.q)));
def write(self):
outdict = {'model' : self.model,
'data_out' : {'q' : json.dumps(self.q_vals_out),
'i' : json.dumps(self.i_vals_out),
'units' : 'A^-1'},
'run' : {'command' : self.command,
'date' : str(datetime.date.today()),
'time' : str(datetime.time())},
'parameters_in' : self.parameters_in}
if self.dataset:
outdict['dataset'] = {'q_in' : json.dumps(self.datain.q),
'i_in' : json.dumps(self.datain.i)}
if (self.fitsuccess and (self.command == 'fit')):
outdict['fit'] = {'chi^2' : self.chisqr,
'cov_x' : np.array_repr(self.cov_x),
'parameters_out' : self.parameters}
path, filename = os.path.split(self.outpath)
if filename == '':
filename = self.outfile
if not os.path.exists(path):
os.mkdir(path)
if self.xml:
self.write_cml(path, filename)
else:
f = open(os.path.join(path, filename), 'w')
json.dump(outdict, f)
f.close()
def testFProp(self, dtype=tf.float32, fprop_dtype=tf.float32):
with self.session() as sess:
tf.set_random_seed(_TF_RANDOM_SEED)
p = self._testParams()
p.dtype = dtype
if fprop_dtype:
p.fprop_dtype = fprop_dtype
p.input.dtype = fprop_dtype
mdl = p.cls(p)
mdl.FProp(mdl.theta)
loss = mdl.loss
logp = mdl.eval_metrics['log_pplx'][0]
tf.global_variables_initializer().run()
vals = []
for _ in range(5):
vals += [sess.run((loss, logp))]
print('actual vals = %s' % np.array_repr(np.array(vals)))
self.assertAllClose(vals, [(189.22296, 10.368382),
(282.57202, 10.369616),
(142.55638, 10.367737),
(139.9939, 10.369918),
(293.08011, 10.374517)],
atol=1e-6,
rtol=1e-6)
def testSpectrumAugmenterWithFrequencyMask(self):
with self.session(use_gpu=False, graph=tf.Graph()) as sess:
tf.compat.v1.set_random_seed(1234)
inputs = tf.ones([3, 5, 4, 2], dtype=tf.float32)
paddings = tf.zeros([3, 5])
p = spectrum_augmenter.SpectrumAugmenter.Params()
p.name = 'specAug_layers'
p.freq_mask_max_bins = 6
p.time_mask_max_frames = 0
specaug_layer = p.Instantiate()
expected_output = np.array(
[[[[1., 1.], [1., 1.], [1., 1.], [1., 1.]],
[[1., 1.], [1., 1.], [1., 1.], [1., 1.]],
[[1., 1.], [1., 1.], [1., 1.], [1., 1.]],
[[1., 1.], [1., 1.], [1., 1.], [1., 1.]],
[[1., 1.], [1., 1.], [1., 1.], [1., 1.]]],
[[[1., 1.], [0., 0.], [0., 0.], [0., 0.]],
[[1., 1.], [0., 0.], [0., 0.], [0., 0.]],
[[1., 1.], [0., 0.], [0., 0.], [0., 0.]],
[[1., 1.], [0., 0.], [0., 0.], [0., 0.]],
[[1., 1.], [0., 0.], [0., 0.], [0., 0.]]],
[[[0., 0.], [0., 0.], [0., 0.], [1., 1.]],
[[0., 0.], [0., 0.], [0., 0.], [1., 1.]],
[[0., 0.], [0., 0.], [0., 0.], [1., 1.]],
[[0., 0.], [0., 0.], [0., 0.], [1., 1.]],
[[0., 0.], [0., 0.], [0., 0.], [1., 1.]]]])
h, _ = specaug_layer.FPropDefaultTheta(inputs, paddings)
actual_layer_output = sess.run(h)
print(np.array_repr(actual_layer_output))
self.assertAllClose(actual_layer_output, expected_output)
def __repr__(self):
"""representation a mdf_skeleton class data strucutre
Returns:
------------
string of mdf class ordered as below
master_channel_name
channel_name description
numpy_array unit
"""
output = 'file name : ' + self.fileName + '\n'
for m in self.file_metadata.keys():
output += m + ' : ' + str(self.file_metadata[m]) + '\n'
output += '\nchannels listed by data groups:\n'
for d in self.masterChannelList.keys():
if d is not None:
output += d + '\n'
for c in self.masterChannelList[d]:
output += ' ' + c + ' : '
desc = self.getChannelDesc(c)
if desc is not None:
output += str(desc)
output += '\n '
data = self.getChannelData(c)
if data.dtype.kind != 'V': # not byte, impossible to represent
output += array_repr(data, \
precision=3, suppress_small=True)
unit = self.getChannelUnit(c)
output += ' ' + unit + '\n'
return output
def testDecoderSampleTargetSequences(self):
p = self._DecoderParams(
vn_config=py_utils.VariationalNoiseParams(None, False, False),
num_classes=8)
p.target_seq_len = 5
p.random_seed = 1
config = tf.ConfigProto(
graph_options=tf.GraphOptions(
optimizer_options=tf.OptimizerOptions(do_function_inlining=False)))
with self.session(use_gpu=False, config=config) as sess:
tf.set_random_seed(8372740)
np.random.seed(35315)
dec = p.Instantiate()
source_sequence_length = 5
batch_size = 4
source_encodings = tf.constant(
np.random.normal(
size=[source_sequence_length, batch_size, p.source_dim]),
dtype=tf.float32)
source_encoding_padding = tf.constant(
[[0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 1.0], [0.0, 1.0, 1.0, 1.0],
[0.0, 1.0, 1.0, 1.0], [0.0, 1.0, 1.0, 1.0]],
dtype=tf.float32)
encoder_outputs = py_utils.NestedMap(
encoded=source_encodings, padding=source_encoding_padding)
sampled_sequences = dec.SampleTargetSequences(
dec.theta, encoder_outputs, random_seed=tf.to_int32(123))
self.assertAllEqual([batch_size, p.target_seq_len],
sampled_sequences.ids.shape)
tf.global_variables_initializer().run()
decoder_output = sess.run(sampled_sequences)
print('ids=%s' % np.array_repr(decoder_output.ids))
lens = np.sum(1 - decoder_output.paddings, axis=1)
print('lens=%s' % lens)
# pyformat: disable
# pylint: disable=bad-whitespace,bad-continuation
expected_ids = [[6, 2, 2, 2, 2],
[0, 0, 7, 5, 1],
[6, 1, 5, 1, 5],
[6, 7, 7, 4, 4]]
# pylint: enable=bad-whitespace,bad-continuation
# pyformat: enable
expected_lens = [2, 5, 5, 5]
self.assertAllEqual(expected_lens, lens)
self.assertAllEqual(expected_ids, decoder_output.ids)
# Sample again with the same random seed.
decoder_output2 = sess.run(
dec.SampleTargetSequences(
dec.theta, encoder_outputs, random_seed=tf.to_int32(123)))
# Get the same output.
self.assertAllEqual(decoder_output.ids, decoder_output2.ids)
self.assertAllEqual(decoder_output.paddings, decoder_output2.paddings)
# Sample again with a different random seed.
decoder_output3 = sess.run(
dec.SampleTargetSequences(
dec.theta, encoder_outputs, random_seed=tf.to_int32(123456)))
# Get different sequences.
self.assertNotAllClose(expected_ids, decoder_output3.ids)
def __repr__(self) -> str:
name_class = type(self).__name__
repr_point = np.array_repr(self.point)
return f"{name_class}(point={repr_point}, radius={self.radius})"
def testBeamSearchHelper(self):
with self.session(use_gpu=False) as sess:
topk_ids, topk_lens, topk_scores = GetBeamSearchHelperResults(
sess, num_hyps_per_beam=3)
print(np.array_repr(topk_ids))
print(np.array_repr(topk_lens))
print(np.array_repr(topk_scores))
expected_topk_ids = [[4, 3, 4, 3, 2, 0, 0], [4, 3, 11, 2, 0, 0, 0],
[4, 3, 6, 2, 0, 0, 0], [6, 0, 4, 6, 6, 11, 2],
[6, 0, 4, 6, 1, 2, 0], [6, 0, 4, 6, 6, 2, 0]]
expected_topk_lens = [5, 4, 4, 7, 6, 6]
expected_topk_scores = [[8.27340603, 6.26949024, 5.59490776],
[9.74691486, 8.46679497, 7.14809656]]
self.assertAllEqual(expected_topk_ids, topk_ids.tolist())
self.assertAllEqual(expected_topk_lens, topk_lens.tolist())
self.assertAllClose(expected_topk_scores, topk_scores)
def __init__(self, q, x, f, qdesc="q", xdesc="x", fdesc="f"):
"""
q ... 1D-array of shape (q.size)
x ... 2D-array of shape (q.size,x.size)
f ... 2D-array of shape (q.size,x.size)
*desc... description string for q,x,f
"""
self.f = np.asarray(f, dtype='float')
self.x = np.asarray(x, dtype='float')
self.q = np.asarray(q, dtype='float')
self.fdesc = fdesc
self.xdesc = xdesc
self.qdesc = qdesc
# complete input data (if constant for different q)
if (len(self.x.shape) == 1):
self.x = np.tile(self.x, (len(self.q), 1))
if (len(self.f.shape) == 1):
self.f = np.tile(self.f, (len(self.q), 1))
# test shape of input data
if (self.q.shape[0] <> self.x.shape[0]) or \
(self.x.shape <> self.f.shape):
raise ValueError("Invalid shape of arguments.")
# test for double parameters
if (np.unique(self.q).size < self.q.size):
raise ValueError("Parameters are not unique: \n " +
np.array_repr(np.sort(self.q)))
def plot(points, step, h, w):
new_p = []
for p in points:
new_p.append({
"x": p["x"] + (p["vel_x"] * step),
"y": p["y"] + (p["vel_y"] * step)
})
max_x = max(p['x'] for p in new_p)
max_y = max(p['y'] for p in new_p)
min_x = min(p["x"] for p in new_p)
min_y = min(p["y"] for p in new_p)
res = False
# If it fis between the bounds, print it out
if max_y - min_y <= h and max_x - min_x <= w:
array = numpy.zeros((h, w))
res = True
for p in new_p:
array[p['y'] - min_y][p['x'] - min_x] = 1
# Print nicely
for ar in array:
print(numpy.array_repr(ar).replace('\n', '').replace('\t', '').replace(" ", "").replace("0.", " "))
return res
def _testLayerHelper(self,
test_case,
sess,
p,
expected=None,
not_expected=None,
global_step=-1):
tf.set_random_seed(398847392)
np.random.seed(12345)
p.name = 'proj'
p.input_dim = 3
p.output_dim = 4
p.params_init = py_utils.WeightInit.Gaussian(0.1)
l = p.cls(p)
in_padding = tf.zeros([2, 4, 1], dtype=tf.float32)
in_padding = tf.constant([[[0], [0], [1], [0]], [[1], [1], [0], [0]]],
dtype=tf.float32)
inputs = tf.constant(np.random.normal(0.1, 0.5, [2, 4, 3]),
dtype=tf.float32)
output = l.FPropDefaultTheta(inputs, in_padding)
tf.global_variables_initializer().run()
if global_step >= 0:
sess.run(
tf.assign(py_utils.GetOrCreateGlobalStepVar(), global_step))
output = output.eval()
print('QuantizableLayerTest output', test_case, ':\n',
np.array_repr(output))
if expected is not None:
self.assertAllClose(output, expected)
if not_expected is not None:
self.assertNotAllClose(output, not_expected)
return l
def FindResultFromList(result, expected_results):
"""Find the given result from a list of expected results.
Args:
result: A MassOutput tuple, from running ops.mass().
expected_results: A list of MassOutput. The test asserts `result` is equal
to at least one result from `expected_results`.
Returns:
The index of first match found, or None for not found.
We use this when the specific output from ops.mass() is not stable across
different platforms. Specifically, the implementation currently uses
std::shuffle(), which have different implementations between libc++ and
stdlibc++.
"""
for idx, expected in enumerate(expected_results):
match = True
for attr in MassOutput._fields:
if not np.array_equal(getattr(result, attr), getattr(expected, attr)):
match = False
break
if match:
return idx
tf.logging.error('Found unexpected output from op.mass that fails to match'
' any expected result.')
for attr in MassOutput._fields:
tf.logging.info('%s = %s', attr, np.array_repr(getattr(result, attr)))
return None
def testTransformerFeedForwardLayerSpecOutDim(self):
with self.session(use_gpu=True) as sess:
tf.set_random_seed(3980847392)
inputs = tf.random_normal([5, 2, 3], seed=948387483)
paddings = tf.zeros([5, 2])
p = layers_with_attention.TransformerFeedForwardLayer.Params()
p.name = 'transformer_fflayer'
p.input_dim = 3
p.output_dim = 5
p.hidden_dim = 7
transformer_fflayer = layers_with_attention.TransformerFeedForwardLayer(p)
h = transformer_fflayer.FPropDefaultTheta(inputs, paddings)
tf.global_variables_initializer().run()
actual_layer_output = sess.run(h)
# pylint: disable=bad-whitespace
# pyformat: disable
expected_output = [
[[ 1.42697251, 0.79269135, -0.85500956, -0.8122285 , -1.56555367],
[-1.7876718 , 0.26025945, -3.18244219, 1.34756351, 0.25739765]],
[[ 1.27962363, 0.88677615, -1.23556185, -1.06855559, -1.27293301],
[ 0.89336467, 2.46229172, 0.11302143, 1.19385004, -2.37805009]],
[[ 2.80146003, -0.66912627, 1.50160134, -2.30645609, -1.18872762],
[ 1.61967182, -0.51639485, 0.24441491, -1.0871532 , -0.95539457]],
[[ 2.03333473, -0.78205228, 0.71245927, -1.63276744, -0.91654319],
[ 1.54542768, -0.30343491, 0.10666496, -1.67965126, -0.15671858]],
[[ 1.60873222, -1.88402128, 0.79040933, -1.97199082, 0.4778356 ],
[-0.13516766, -0.42583361, -1.86275542, -1.09650302, 0.83263111]]]
# pyformat: enable
# pylint: enable=bad-whitespace
print(np.array_repr(actual_layer_output))
self.assertAllClose(actual_layer_output, expected_output)
def showProjectionDialog(self):
"""Get and set OpenGL ModelView matrix and focus.
Useful for setting two different instances to the exact same projection"""
dlg = uic.loadUi("multilineinputdialog.ui")
dlg.setWindowTitle("Get and set OpenGL ModelView matrix and focus")
precision = 8 # use default precision
MV_repr = np.array_repr(self.MV, precision=precision)
focus_repr = np.array_repr(self.focus, precision=precision)
txt = "self.MV = \\\n" "%s\n\n" "self.focus = %s" % (MV_repr, focus_repr)
dlg.plainTextEdit.insertPlainText(txt)
dlg.plainTextEdit.selectAll()
if dlg.exec_(): # returns 1 if OK, 0 if Cancel
txt = str(dlg.plainTextEdit.toPlainText())
from numpy import array, float32 # required for exec()
exec(txt) # update self.MV and self.focus, with hopefully no maliciousness
def testBiEncoderForwardPassWithDropout(self):
with self.session(use_gpu=False):
tf.set_random_seed(8372749040)
p = self._BiEncoderParams()
p.dropout_prob = 0.5
mt_enc = encoder.MTEncoderBiRNN(p)
batch = py_utils.NestedMap()
batch.ids = tf.transpose(tf.reshape(tf.range(0, 8, 1), [4, 2]))
batch.paddings = tf.zeros([2, 4])
enc_out = mt_enc.FPropDefaultTheta(batch).encoded
tf.global_variables_initializer().run()
actual_enc_out = enc_out.eval()
print('bi_enc_actual_enc_out_with_dropout',
np.array_repr(actual_enc_out))
# pylint: disable=bad-whitespace,bad-continuation
# pyformat: disable
expected_enc_out = [[[-2.25614094e-05, 1.19781353e-05],
[-2.74532852e-07, 8.17993077e-06]],
[[2.66865045e-05, 1.02941645e-04],
[1.51371260e-05, 3.78371587e-05]],
[[3.50117516e-05, 7.65562072e-06],
[-1.30227636e-05, 3.01171349e-06]],
[[2.27566215e-06, 1.42354111e-07],
[1.04521234e-06, 2.50320113e-06]]]
# pyformat: enable
# pylint: enable=bad-whitespace,bad-continuation
self.assertAllClose(expected_enc_out, actual_enc_out)
def testFProp(self, dtype=tf.float32, fprop_dtype=tf.float32):
with self.session() as sess:
tf.set_random_seed(_TF_RANDOM_SEED)
p = self._testParams()
p.dtype = dtype
if fprop_dtype:
p.fprop_dtype = fprop_dtype
p.input.dtype = fprop_dtype
mdl = p.Instantiate()
input_batch = mdl.GetInputBatch()
mdl.FProp(mdl.theta, input_batch)
loss = mdl.loss
logp = mdl.eval_metrics['log_pplx'][0]
tf.global_variables_initializer().run()
vals = []
for _ in range(5):
vals += [sess.run((loss, logp))]
print('actual vals = %s' % np.array_repr(np.array(vals)))
self.assertAllClose(vals, [
[233.337143, 10.370541],
[235.853119, 10.367168],
[217.87796, 10.375141],
[217.822205, 10.372487],
[159.483185, 10.37289],
])
def testEvolvedTransformerDecoderBranchedConvsLayer(self):
layer = layers_with_attention.EvolvedTransformerDecoderBranchedConvsLayer
with self.session(use_gpu=True) as sess:
tf.set_random_seed(3980847392)
inputs = tf.random_normal([5, 2, 3], seed=948387483)
paddings = tf.zeros([5, 2])
p = layer.Params()
p.name = 'et_decoder_branched_convs'
p.input_dim = 3
et_branched_convs = layer(p)
h = et_branched_convs.FPropDefaultTheta(inputs, paddings)
tf.global_variables_initializer().run()
actual_layer_output = sess.run(h)
# pylint: disable=bad-whitespace
# pyformat: disable
expected_output = [
[[-0.31987068, -0.65715098, 0.90350437],
[ 0.00773269, 1.07779562, 4.11094666]],
[[-0.84862059, -0.93186408, 1.16371167],
[ 1.31467259, 0.03560367, 2.36822462]],
[[ 0.02183507, -0.0799394 , -1.68870354],
[ 0.77921551, 1.30145741, -0.86353606]],
[[ 0.31672907, 0.50000876, -0.93973017],
[-0.54707348, 0.19211179, -1.45307386]],
[[-0.46405494, 0.65833056, -1.09345317],
[-1.17221224, -0.08027397, 0.84021652]]]
# pyformat: enable
# pylint: enable=bad-whitespace
print(np.array_repr(actual_layer_output))
self.assertAllClose(actual_layer_output, expected_output)
def __repr__(self):
'''
this does not print parent. mode is printed as ha.modes['name']
'''
return "Star(HylaaSettings({}, {}), {}, {}, {}, None, ha.modes['{}'], extra_init=({}, {}, {}))".format(
self.settings.step,
self.settings.step * self.settings.num_steps,
array_repr(self.center),
array_repr(self.basis_matrix),
self.constraint_list,
self.mode.name,
array_repr(self.start_basis_matrix) if self.start_basis_matrix is not None else None,
self.total_steps,
self.fast_forward_steps
)
def testEvolvedTransformerEncoderLayerFProp(self):
with self.session(use_gpu=True) as sess:
np.random.seed(6348575)
depth = 4
p = GPipeEvolvedTransformerEncoderLayer.Params()
p.name = 'gpipe_evolved_transformer_encoder'
p.source_dim = depth
p.transformer_tpl.tr_fflayer_tpl.hidden_dim = 7
p.transformer_tpl.tr_atten_tpl.num_attention_heads = 2
transformer = GPipeEvolvedTransformerEncoderLayer(p)
(source_vecs, source_padding, _, _) = self._testInputs(depth=depth)
h = transformer.FPropDefaultTheta(source_vecs, source_padding, None)[0]
tf.global_variables_initializer().run()
actual_layer_output = sess.run([h])[0]
tf.logging.info(np.array_repr(actual_layer_output))
# pylint: disable=bad-whitespace
# pyformat: disable
expected_layer_output = [
[[-2.03854632, -1.07184005, -0.28417355, 0.17936069],
[-0.74067241, -1.48318326, 0.26369774, 0.62173623]],
[[-2.12831736, -0.86353737, -0.54453588, 0.13070297],
[-0.76326936, -0.04828247, -0.49510449, 1.20852029]],
[[ 0.85539216, -1.21577334, -1.28910851, -0.15619087],
[-1.45574117, -1.11208296, 0.71455258, 0.91494167]],
[[-1.21304905, -1.37239563, 0.7022025 , 0.16537377],
[ 3.07106829, 1.35782909, -0.9944036 , -2.28987551]],
[[-0.13129801, -1.70681071, -0.42324018, 1.32114363],
[-1.53065133, 0.18422687, -0.93387115, 1.37142754]]]
# pyformat: enable
# pylint: enable=bad-whitespace
self.assertAllClose(expected_layer_output, actual_layer_output)
def store_img(self,
img,
label,
poly,
keypoints=None,
processed_img=None,
rel_path=None):
'''
img: the image
poly: polygon points
path: path ending with name (not extension) which will be used for prepending pickle, processed, etc
label: name
processed_img: the processed image. Will be saved for reference
'''
if rel_path is None:
rel_path = hashlib.sha256(np.array_repr(img).encode()).hexdigest()
path = os.path.join(self.template_dir, rel_path)
if len(path.split('.jpg')) > 1:
path = path.split('.jpg')[0]
img = ImageDB.Image(path, img, label, poly, keypoints)
self.images.append(img)
if processed_img is not None:
cv2.imwrite(path + '_processes.jpg', processed_img)
# store pickle
with open(path + '.pickle', 'wb') as f:
pickle.dump(img, f)
def testhomos(self):
"""Are the H**O indices equal to 34 and 33 (one more alpha electron
than beta electron)?
"""
msg = "%s != array([34, 33], 'i')" % numpy.array_repr(self.data.homos)
numpy.testing.assert_array_equal(self.data.homos,
numpy.array([34, 33], "i"), msg)
def repr_ndarray(x, _helper):
# noinspection PyPackageRequirements
import numpy as np
dims = len(x.shape)
if (
# Too many dimensions to be concise
dims > 6 or
# There's a bug with array_repr and matrices
isinstance(x, np.matrix)
and np.lib.NumpyVersion(np.__version__) < '1.14.0' or
# and with masked arrays...
isinstance(x, np.ma.MaskedArray)):
name = type_name(x)
if name == 'ndarray':
name = 'array'
return '%s(%r, shape=%r)' % (name, x.dtype, x.shape)
edgeitems = repr_ndarray.maxparts // 2
if dims == 3:
edgeitems = min(edgeitems, 2)
elif dims > 3:
edgeitems = 1
opts = np.get_printoptions()
try:
np.set_printoptions(threshold=repr_ndarray.maxparts,
edgeitems=edgeitems)
return np.array_repr(x)
finally:
np.set_printoptions(**opts)
def testEvolvedTransformerEncoderBranchedConvsLayer(self):
layer = layers_with_attention.EvolvedTransformerEncoderBranchedConvsLayer
with self.session(use_gpu=True) as sess:
tf.set_random_seed(3980847392)
inputs = tf.random_normal([5, 2, 3], seed=948387483)
paddings = tf.zeros([5, 2])
p = layer.Params()
p.name = 'et_encoder_branched_convs'
p.input_dim = 3
et_branched_convs = layer(p)
h = et_branched_convs.FPropDefaultTheta(inputs, paddings)
tf.global_variables_initializer().run()
actual_layer_output = sess.run(h)
# pylint: disable=bad-whitespace
# pyformat: disable
expected_output = [
[[-0.13232423, -0.46060669, 0.72598207],
[ 0.6725747 , 1.58664441, 2.64087844]],
[[-0.21702465, -0.68267912, 1.20886588],
[ 1.69793618, 0.53306532, 1.02958691]],
[[-0.46037287, -0.42950529, -1.68443251],
[ 0.21459752, 0.42246291, -0.01271994]],
[[-0.23293658, 0.15300342, -0.83518255],
[-0.48914853, -0.44239512, -0.2328119 ]],
[[-0.57934833, 0.24165238, -1.05392623],
[-0.8292231 , 0.06175411, 1.28672981]]]
# pyformat: enable
# pylint: enable=bad-whitespace
print(np.array_repr(actual_layer_output))
self.assertAllClose(actual_layer_output, expected_output)
def testForwardPassWithStackingAfterMiddleLayer(self):
with self.session(use_gpu=False):
vn_config = py_utils.VariationalNoiseParams(None, False, False)
p = self._EncoderParams(vn_config)
p.stacking_layer_tpl.left_context = 1
p.stacking_layer_tpl.right_context = 0
p.stacking_layer_tpl.stride = 2
p.layer_index_before_stacking = 0
enc_out = self._ForwardPass(p).encoded
enc_out_sum = tf.reduce_sum(enc_out, 0)
tf.global_variables_initializer().run()
# pyformat: disable
# pylint: disable=bad-whitespace
expected_enc_out = [
[0.00102275, -0.02697385, 0.01709868, -0.00939053, -0.01576837,
0.0070826, -0.00626193, 0.01143604, -0.01742513, -0.00529445,
0.00284249, -0.01362027, -0.00490865, 0.0216262, -0.01344598,
-0.00460993, -0.01329017, 0.01379208, -0.00850593, 0.0193335,
0.01134925, -0.00131254, 0.00375953, -0.00588882, 0.01347932,
-0.00252493, 0.01274828, 0.01027388, 0.02657663, 0.02644286,
0.0286899, -0.00833998],
[-0.01801126, 0.0115137, 0.01355767, 0.00113954, 0.00986663,
-0.0128988, 0.00794239, -0.00524312, 0.00246279, -0.00575782,
-0.00213567, -0.01528412, 0.00186096, 0.00253562, -0.00411006,
-0.00390748, -0.01001569, -0.00344393, -0.01211706, 0.00387725,
0.02194905, 0.02578988, -0.00255773, 0.00690117, 0.00976908,
0.01935913, 0.01131854, 0.0013859, -0.01567556, 0.01858256,
0.02251371, -0.0185001]]
# pylint: enable=bad-whitespace
# pyformat: enable
enc_out_sum_val = enc_out_sum.eval()
print('enc_out_sum_val', np.array_repr(enc_out_sum_val))
self.assertAllClose(expected_enc_out, enc_out_sum_val)
def testTransformerAttentionLayerCase2(self):
with self.session(use_gpu=True) as sess:
depth = 4
p = layers_with_attention.TransformerAttentionLayer.Params()
p.name = 'transformer_atten'
p.source_dim = depth
p.is_masked = True
p.num_attention_heads = 2
transformer_atten = layers_with_attention.TransformerAttentionLayer(
p)
(source_vecs, source_padding, _,
_) = self._testTransformerAttentionLayerInputs(depth=depth)
ctx, probs = transformer_atten.FPropDefaultTheta(
source_vecs, source_padding)
tf.global_variables_initializer().run()
actual_ctx, actual_probs = sess.run([ctx, probs])
tf.logging.info(np.array_repr(actual_ctx))
tf.logging.info(np.array_repr(actual_probs))
# pylint: disable=bad-whitespace
# pyformat: disable
expected_ctx = [
[[-0.14429152, 1.15510106, 1.11930299, -1.19245839],
[-0.69580591, -0.47006619, 0.82592297, 0.69593251]],
[[0.24164687, 0.53328454, -1.02119482, -1.49412084],
[-0.82601064, 0.024203, -1.11880171, 1.80784416]],
[[1.7644347, -0.53346401, -1.1461122, -1.42797422],
[-0.95326459, 0.39580142, 0.39262164, 0.67513674]],
[[-0.28252155, -0.95237327, 2.08757687, -0.21231559],
[1.4362365, 0.46009994, -1.45436597, -1.90602148]],
[[-0.51681399, -0.70075679, -0.48352116, 1.93754733],
[-1.44486678, 0.81801879, -1.03079689, 1.86697066]]
]
expected_probs = [[[1., 0., 0., 0., 0.], [0.2, 0.2, 0.2, 0.2,
0.2]],
[[0.3966811, 0.60331887, 0., 0., 0.],
[0., 1., 0., 0., 0.]],
[[0.41050252, 0.58949745, 0., 0., 0.],
[0., 0.5245893, 0.4754107, 0., 0.]],
[[0.58882225, 0.41117775, 0., 0., 0.],
[0., 0.31849149, 0.28174096, 0.39976761, 0.]],
[[0.16272782, 0.15781289, 0., 0., 0.67945927],
[0., 0.55003977, 0.26049581, 0.18946445, 0.]]]
# pyformat: enable
# pylint: enable=bad-whitespace
self.assertAllClose(expected_ctx, actual_ctx)
self.assertAllClose(expected_probs, actual_probs)
def testTransformerAttentionLayerCase3(self):
with self.session(use_gpu=True) as sess:
depth = 4
p = layers_with_attention.TransformerAttentionLayer.Params()
p.name = 'transformer_atten'
p.source_dim = depth
p.is_masked = False
p.num_attention_heads = 2
transformer_atten = layers_with_attention.TransformerAttentionLayer(p)
(query_vec, _, aux_vecs,
aux_paddings) = self._testTransformerAttentionLayerInputs(depth=depth)
ctx, probs = transformer_atten.FPropDefaultTheta(query_vec, aux_paddings,
aux_vecs)
tf.global_variables_initializer().run()
actual_ctx, actual_probs = sess.run([ctx, probs])
tf.logging.info(np.array_repr(actual_ctx))
tf.logging.info(np.array_repr(actual_probs))
# pylint: disable=bad-whitespace
# pyformat: disable
expected_ctx = [
[[-1.42420077, 1.19024372, 1.35146523, 0.85896158],
[-0.44974625, -1.00108492, 1.63387251, 1.678146 ]],
[[ 0.1134335 , 1.97617495, -0.35918081, 0.26396495],
[-0.19688171, -0.71197301, 0.0659425 , 2.5417304 ]],
[[ 1.58169425, 0.81259179, -0.58948535, 0.20254248],
[-0.84438968, -0.65845209, 1.45584249, 1.87587976]],
[[-1.01532316, -0.05166581, 2.07901478, 0.97540361],
[ 2.08563352, 0.34328598, -0.23240227, -0.19035631]],
[[-0.53881919, -0.60117185, 0.29170275, 2.6474514 ],
[-0.88318163, 0.37149727, -0.16098523, 2.3810885 ]]]
expected_probs = [
[[ 0.32392544, 0., 0.27218491, 0., 0.19574419, 0., 0.20814547],
[ 0., 0.273045 , 0., 0.43572819, 0., 0.2912268 , 0.]],
[[ 0.24094662, 0., 0.23919827, 0., 0.26563686, 0., 0.25421822],
[ 0., 0.21680018, 0., 0.33962148, 0.,0.44357836 , 0.]],
[[ 0.20083594, 0., 0.20683075, 0., 0.28931937, 0., 0.30301392],
[ 0., 0.24710922, 0., 0.453915 , 0.,0.29897571 , 0.]],
[[ 0.32845193, 0., 0.26491433, 0., 0.18304622, 0., 0.22358747],
[ 0., 0.39426237, 0., 0.19774443, 0.,0.4079932 , 0.]],
[[ 0.23542665, 0., 0.27910906, 0., 0.30036426, 0., 0.18510005],
[ 0., 0.20147586, 0., 0.37759233, 0., 0.42093182, 0.]]]
# pyformat: enable
# pylint: enable=bad-whitespace
self.assertAllClose(expected_ctx, actual_ctx, rtol=1e-05, atol=1e-05)
self.assertAllClose(expected_probs, actual_probs, rtol=1e-05, atol=1e-05)
def _testBeamSearchOpHelper(self,
b_size,
num_beams,
seq_len,
init_best_score,
probs,
init_atten_probs,
atten_probs,
best_scores_expected,
cum_scores_expected,
scores_expected,
hyps_expected,
prev_hyps_expected,
atten_probs_expected,
force_eos_in_last_step=False):
(best_scores, cumulative_scores, scores, hyps, prev_hyps, done_hyps,
atten_probs, done, scores, atten_probs) = self._runBeamSearchOpHelper(
b_size,
num_beams,
seq_len,
init_best_score,
probs,
init_atten_probs,
atten_probs,
force_eos_in_last_step=force_eos_in_last_step)
tf.logging.info(np.array_repr(best_scores))
tf.logging.info(np.array_repr(cumulative_scores))
tf.logging.info(np.array_repr(scores))
tf.logging.info(np.array_repr(hyps))
tf.logging.info(np.array_repr(prev_hyps))
tf.logging.info(np.array_repr(done_hyps))
tf.logging.info(np.array_repr(atten_probs))
tf.logging.info(np.array_repr(done))
tf.logging.info(np.array_repr(scores))
tf.logging.info(np.array_repr(atten_probs))
self.assertAllClose(best_scores_expected, best_scores)
self.assertAllClose(cum_scores_expected, cumulative_scores)
self.assertAllClose(scores_expected, scores)
self.assertAllClose(hyps_expected, hyps)
self.assertAllClose(prev_hyps_expected, prev_hyps)
self.assertAllClose(atten_probs_expected, atten_probs)
self.assertEqual(False, done)
return done_hyps
def show(self, count):
hist = self.collect_percent(count)
hist_str = np.array_repr(hist,
max_line_width=1024,
precision=2,
suppress_small=True)
print("Histogram for vertex {}\n{}".format(
self.vert_id, hist_str))
def execute(self, v):
print("\n\nMonitorLayer %s at execution %d:" % (self.tag, self.i))
o = np.get_printoptions()
np.set_printoptions(threshold=np.nan)
print(np.array_repr(v))
np.set_printoptions(**o)
self.i += 1
return v
def showProjectionDialog(self):
"""Get and set OpenGL ModelView matrix and focus.
Useful for setting two different instances to the exact same projection"""
dlg = uic.loadUi('multilineinputdialog.ui')
dlg.setWindowTitle('Get and set OpenGL ModelView matrix and focus')
precision = 8 # use default precision
MV_repr = np.array_repr(self.MV, precision=precision)
focus_repr = np.array_repr(self.focus, precision=precision)
txt = ("self.MV = \\\n"
"%s\n\n"
"self.focus = %s" % (MV_repr, focus_repr))
dlg.plainTextEdit.insertPlainText(txt)
dlg.plainTextEdit.selectAll()
if dlg.exec_(): # returns 1 if OK, 0 if Cancel
txt = str(dlg.plainTextEdit.toPlainText())
from numpy import array, float32 # required for exec()
exec(txt) # update self.MV and self.focus, with hopefully no maliciousness
def test_train(self):
model = Surrogates.RegressionModels.RidgeRegression.RidgeRegression(sp=self._sp, encode=False, rng=1, debug=True)
x_train_data = self._data[:1000, :-2]
y_train_data = self._data[:1000, -1]
x_test_data = self._data[1000:, :-2]
y_test_data = self._data[1000:, -1]
self.assertEqual(hash(numpy.array_repr(x_train_data)), -4233919799601849470)
self.assertEqual(hash(numpy.array_repr(y_train_data)), -5203961977442829493)
model.train(x=x_train_data, y=y_train_data, param_names=self._para_header)
lower, upper = model._scale_info
should_be_lower = [None, -29.6210089736, 0.201346561323, 0, -20.6929600285, 0, 0, 0, 4.60517018599, 0,
2.77258872224, 0, 0, 0.502038871605, -17.2269829469]
should_be_upper = [None, -7.33342451433, 1.99996215592, 1, -6.92778489957, 2, 1, 1, 9.20883924585, 1,
6.9314718056, 3, 1, 0.998243871085, 4.72337617503]
for idx in range(x_train_data.shape[1]):
self.assertEqual(lower[idx], should_be_lower[idx])
self.assertEqual(upper[idx], should_be_upper[idx])
y = model.predict(x=x_train_data[1, :])
print "Is: %100.70f, Should: %f" % (y, y_train_data[1])
self.assertAlmostEqual(y[0], 0.337919078549359763741222195676527917385101318359375, msg=y[0])
print "Predict whole data"
y_whole = model.predict(x=x_test_data)
mse = mean_squared_error(y_true=y_test_data, y_pred=y_whole)
print "MSE: %100.70f" % mse
self.assertAlmostEqual(mse, 0.009198484147153261625273756862)
# Try the same with encoded features
model = Surrogates.RegressionModels.RidgeRegression.RidgeRegression(sp=self._sp, encode=True, rng=1, debug=True)
#print data[:10, :-2]
model.train(x=x_train_data, y=y_train_data, param_names=self._para_header, rng=1)
y = model.predict(x=self._data[1, :-2])
print "Is: %100.70f, Should: %f" % (y, self._data[1, -2])
self.assertAlmostEqual(y[0], 0.337619548171, msg="%f" % y[0])
print "Predict whole data"
y_whole = model.predict(x=x_test_data)
mse = mean_squared_error(y_true=y_test_data, y_pred=y_whole)
print "MSE: %100.70f" % mse
self.assertAlmostEqual(mse, 0.0092026737874672301)
def __str__(self):
""" prints compressed_data object content
"""
output = 'Data: \n'
output += array_repr(self.data[:],
precision=3,
suppress_small=True)
output += '\n Compression level ' + str(self._compression_level) + '\n'
return output
def _repr_footer(self):
levheader = "Levels (%d): " % len(self.levels)
# TODO: should max_line_width respect a setting?
levstring = np.array_repr(self.levels, max_line_width=60)
indent = " " * (levstring.find("[") + len(levheader) + 1)
lines = levstring.split("\n")
levstring = "\n".join([lines[0]] + [indent + x.lstrip() for x in lines[1:]])
namestr = "Name: %s, " % self.name if self.name is not None else ""
return u("%s\n%sLength: %d" % (levheader + levstring, namestr, len(self)))
def __repr__(self):
temp = "Categorical: %s\n%s\n%s"
values = np.asarray(self)
levheader = "Levels (%d): " % len(self.levels)
levstring = np.array_repr(self.levels, max_line_width=60)
indent = " " * (levstring.find("[") + len(levheader) + 1)
lines = levstring.split("\n")
levstring = "\n".join([lines[0]] + [indent + x.lstrip() for x in lines[1:]])
return temp % ("" if self.name is None else self.name, repr(values), levheader + levstring)
def main():
file1 = open(str(sys.argv[1]),'r')
file2 = open(str(sys.argv[2]),'r')
cell_1 = UnitCell(file1)
cell_2 = UnitCell(file2)
disp = displacements(cell_1,cell_2,conv='C')
disp = disp.flatten()
if len(sys.argv) > 3:
in_name = str(sys.argv[3])
else:
in_name = "phonons.out"
if len(sys.argv) > 4:
mass_name = str(sys.argv[4])
else:
mass_name = "apos.dat"
freqs,normal_modes = read_normal_modes(in_name)
mass_vec = read_masses(mass_name)
normal_disp = calc_atomic_displacements(normal_modes,mass_vec)
weights = decompose_displacements(normal_disp,disp)
print np.array_repr(weights,precision=8,suppress_small=True)
print np.linalg.norm(weights)
def __repr__(self):
output = 'file name : ' + self.fileName + '\n'
for m in self.file_metadata.keys():
output += m + ' : ' + str(self.file_metadata[m]) + '\n'
output += '\nchannels listed by data groups:\n'
for d in self.masterChannelList.keys():
if d is not None:
output += d + '\n'
for c in self.masterChannelList[d]:
output += ' ' + c + ' : ' + str(self[c]['description']) + '\n'
output += ' ' + array_repr(self[c]['data'], precision=3, suppress_small=True) \
+ ' ' + self[c]['unit'] + '\n'
return output
def __repr__(self):
temp = 'Categorical: %s\n%s\n%s'
values = np.asarray(self)
levheader = 'Levels (%d): ' % len(self.levels)
levstring = np.array_repr(self.levels,
max_line_width=60)
indent = ' ' * (levstring.find('[') + len(levheader) + 1)
lines = levstring.split('\n')
levstring = '\n'.join([lines[0]] + [indent + x.lstrip() for x in lines[1:]])
return temp % ('' if self.name is None else self.name,
repr(values), levheader + levstring)
def area_precision_recall(detector_response, actual):
if type(actual) is str:
actual_img = cv2.imread(actual, flags=-1)
else:
actual_img = actual
assert actual_img.shape == detector_response.shape, \
"Actual shape: {0} Got: {1}".format(actual_img.shape, detector_response.shape)
assert all_binary(detector_response), np.array_repr(np.unique(detector_response))
assert all_binary(actual_img), actual
pr = area_precision(detector_response, actual_img)
rec = area_recall(detector_response, actual_img)
return pr, rec
def compare_response_to_truth(detector_response, actual, cascade=False, thresh=10):
if type(actual) is str:
actual_img = cv2.imread(actual, flags=-1)
else:
actual_img = actual
assert actual_img.shape == detector_response.shape, \
"Actual shape: {0} Got: {1}".format(actual_img.shape, detector_response.shape)
assert all_binary(detector_response), np.array_repr(np.unique(detector_response))
assert all_binary(actual_img), actual
tp = true_positive(detector_response, actual_img, cascade, thresh)
fp = false_postive(detector_response, actual_img, cascade, thresh)
fn = false_negative(detector_response, actual_img, cascade, thresh)
return tp, fp, fn
def save_colormap(self, event):
import textwrap
template = textwrap.dedent('''
from matplotlib.colors import LinearSegmentedColormap
from numpy import nan, inf
# Used to reconstruct the colormap in pycam02ucs.cm.viscm
parameters = {{'xp': {xp},
'yp': {yp},
'min_Jp': {min_Jp},
'max_Jp': {max_Jp}}}
cm_data = {array_list}
test_cm = LinearSegmentedColormap.from_list(__file__, cm_data)
if __name__ == "__main__":
import matplotlib.pyplot as plt
import numpy as np
try:
from pycam02ucs.cm.viscm import viscm
viscm(test_cm)
except ImportError:
print("pycam02ucs not found, falling back on simple display")
plt.imshow(np.linspace(0, 100, 256)[None, :], aspect='auto',
cmap=test_cm)
plt.show()
''')
rgb, _ = self.cmap_model.get_sRGB(num=256)
with open('/tmp/new_cm.py', 'w') as f:
array_list = np.array_repr(rgb, max_line_width=78)
array_list = array_list.replace('array(', '')[:-1]
xp, yp = self.cmap_model.bezier_model.get_control_points()
data = dict(array_list=array_list,
xp=xp,
yp=yp,
min_Jp=self.cmap_model.min_Jp,
max_Jp=self.cmap_model.max_Jp)
f.write(template.format(**data))
print("*" * 50)
print("Saved colormap to /tmp/new_cm.py")
print("*" * 50)
def __str__(self):
"""representation a mdf_skeleton class data strucutre
Returns:
------------
string of mdf class ordered as below
master_channel_name
channel_name description
numpy_array unit
"""
if self.fileName is not None:
output = 'file name : ' + self.fileName + '\n'
else:
output = ''
for m in self.file_metadata.keys():
if self.file_metadata[m] is not None:
output += m + ' : ' + str(self.file_metadata[m]) + '\n'
if not self._pandasframe:
output += '\nchannels listed by data groups:\n'
for d in self.masterChannelList.keys():
if d is not None:
output += d + '\n'
for c in self.masterChannelList[d]:
output += ' ' + c + ' : '
desc = self.getChannelDesc(c)
if desc is not None:
try:
output += str(desc)
except:
pass
output += '\n '
data = self.getChannelData(c)
# not byte, impossible to represent
if data.dtype.kind != 'V':
output += array_repr(data[:],
precision=3, suppress_small=True)
unit = self.getChannelUnit(c)
if unit is not None:
output += ' ' + unit + '\n'
return output
else:
set_option('max_rows', 3)
set_option('expand_frame_repr', True)
set_option('max_colwidth', 6)
for master in self.masterGroups:
output += master
output += str(self[master])
return output
def format_molecule_for_numpy(self, npobj=True):
"""Returns a NumPy array of the non-dummy atoms of the geometry
in Cartesian coordinates in Angstroms with element encoded as
atomic number. If *npobj* is False, returns representation of
NumPy array.
"""
import numpy as np
factor = 1.0 if self.PYunits == 'Angstrom' else psi_bohr2angstroms
self.update_geometry()
# TODO fn title is format_mol... but return args not compatible
geo = []
for i in range(self.natom()):
[x, y, z] = self.atoms[i].compute()
geo.append([self.Z(i), x * factor, y * factor, z * factor])
nparr = np.array(geo)
return nparr if npobj else np.array_repr(nparr)
def count_posneg(img_file):
if type(img_file) is str:
image = cv2.imread(img_file, flags=-1)
else:
image = img_file
dims = image.shape
if len(dims) is 2:
h, w = dims
elif len(dims) is 3:
h, w, d = dims
else:
raise Exception("Incorrect image shape")
pixels = h * w
positives = np.sum(np.sum(image))
negatives = pixels - positives
assert np.array_equal(np.unique(image), np.array([0, 1])) or \
np.array_equal(np.unique(image), np.array([0])), np.array_repr(np.unique(image))
return positives, negatives
def array_repr(arr, max_line_width=None, precision=None, suppress_small=None):
"""Returns the string representation of an array.
Args:
arr (array_like): Input array. It should be able to feed to
:func:`cupy.asnumpy`.
max_line_width (int): The maximum number of line lengths.
precision (int): Floating point precision. It uses the current printing
precision of NumPy.
suppress_small (bool): If True, very small numbers are printed as
zeros
Returns:
str: The string representation of ``arr``.
.. seealso:: :func:`numpy.array_repr`
"""
return numpy.array_repr(cupy.asnumpy(arr), max_line_width, precision,
suppress_small)
def get_vec_str(vec):
assert vec.ndim==1
# use numpy for printing (suppresses small values when others are large)
tmp = np.array_repr(vec, precision=5, suppress_small=True)
assert tmp.startswith('array([')
tmp = tmp[7:]
assert tmp.endswith('])')
tmp = tmp[:-2]
tmp = tmp.strip()
tmps = [t.strip() for t in tmp.split(',')]
# convert -0 to 0
tmps2 = []
for t in tmps:
if t=='-0.':
tmps2.append('0.')
else:
tmps2.append(t)
tmps = tmps2
tmps = ['% 8s'%(t,) for t in tmps ]
result = ', '.join( tmps )
return result
def __str__(self):
str = "Cost as + \ = \ -\n"
for k in range(0, 21):
str += np.array_repr(self.Table_Cost[k]) + " " + self.statsName[k] + "\n"
str += "\nRFD as + \ = \ -\n"
for k in range(0, 21):
str += np.array_repr(self.Table_RFD[k]) + " " + self.statsName[k] + "\n"
str += "\nEPISILON as + \ = \ -\n"
for k in range(0, 21):
str += np.array_repr(self.Table_EPSILON[k]) + " " + self.statsName[k] + "\n"
str += "\nIGD as + \ = \ -\n"
for k in range(0, 21):
str += np.array_repr(self.Table_IGD[k]) + " " + self.statsName[k] + "\n"
str += "\nFt as + \ = \ -\n"
for k in range(0, 21):
str += np.array_repr(self.Table_Ft[k]) + " " + self.statsName[k] + "\n"
str += "\nRFDc as + \ = \ -\n"
for k in range(0, 21):
str += np.array_repr(self.Table_RFDc[k]) + " " + self.statsName[k] + "\n"
return str
def writeStats(self):
str = "# Stats Cost + / = / -\n"
for k in range(0, 21):
str += np.array_repr(self.exp_Cost[k]) + " " + self.statsName[k] + "\n"
str += "\n# Stats RFD + / = / -\n"
for k in range(0, 21):
str += np.array_repr(self.exp_RFD[k]) + " " + self.statsName[k] + "\n"
str += "\n# Stats EPSILON + / = / -\n"
for k in range(0, 21):
str += np.array_repr(self.exp_EPSILON[k]) + " " + self.statsName[k] + "\n"
str += "\n# Stats IGD + / = / -\n"
for k in range(0, 21):
str += np.array_repr(self.exp_IGD[k]) + " " + self.statsName[k] + "\n"
str += "\n# Stats Ft + / = / -\n"
for k in range(0, 21):
str += np.array_repr(self.exp_Ft[k]) + " " + self.statsName[k] + "\n"
str += "\n# Stats RFDc + / = / -\n"
for k in range(0, 21):
str += np.array_repr(self.exp_RFDc[k]) + " " + self.statsName[k] + "\n"
f = open("stats.data", "w")
f.write(str)
f.close()
def cosSim(word1, word2, par):
"""
Args:
contexts1: contexts of word to be compared with contexts of another word
contexts2: contexts of word to be compared with contexts of another word
Returns:
cosine distributional probability of two words
also outputs contexts arrays to serialized.txt
"""
contexts1 = getContext(word1, CORPUS, WINDOW_SIZE)
contexts2 = getContext(word2, CORPUS, WINDOW_SIZE)
keys1 = np.unique(np.array((contexts1, contexts2)).T[0])
keys2 = np.unique(np.array((contexts1, contexts2)).T[1])
# all_keys=None
# for key in range(0, len(keys2)):
# if keys2[key] not in all_keys:
# np.append(all_keys, keys2)
all_keys = np.sort(np.unique(np.append(keys1, keys2)))
# print all_keys
if par is "ppmi":
array1 = np.array([[i, ppmi(getContext(i, CORPUS, WINDOW_SIZE), contexts1, i, word1)] for i in all_keys])
array2 = np.array([[i, ppmi(getContext(i, CORPUS, WINDOW_SIZE), contexts2, i, word2)] for i in all_keys])
elif par is "freq":
array1 = np.array([[i, contexts1.get(i, 0)] for i in all_keys])
array2 = np.array([[i, contexts2.get(i, 0)] for i in all_keys])
elif par is "bin":
array1 = np.array([[i, contexts1.get(i, 0)] for i in all_keys])
for i in range(0, len(array1)):
if array1[i][1].astype(int) > 1:
array1[i][1] = 1
array2 = np.array([[i, contexts2.get(i, 0)] for i in all_keys])
for i in range(0, len(array2)):
if array2[i][1].astype(int) > 1:
array2[i][1] = 1
else:
return "ERROR"
# print contexts1, "\n", contexts2
# print array1, "\n", array2
array1_i = np.array([i[1] for i in array1], dtype=float)
array2_i = np.array([i[1] for i in array2], dtype=float)
# print array1_i, "\n", array2_i
out = (np.dot(array1_i, array2_i)) / (np.linalg.norm(array1_i) * np.linalg.norm(array2_i))
file1 = open("serialized.txt", "w+")
file1.write("Word 1 Frequencies:\n" + np.array_repr(array1) + "\nVector:\n" + np.array_repr(array1_i))
file1.write("\n\nWord 2 Frequencies:\n" + np.array_repr(array2) + "\nVector:\n" + np.array_repr(array2_i))
file1.write("\n\n" + np.array_repr(out))
file1.close()
return out
def testhomos(self):
"""Is the index of the H**O equal to 34?"""
numpy.testing.assert_array_equal(self.data.homos, numpy.array([34],"i"), "%s != array([34],'i')" % numpy.array_repr(self.data.homos))
aggregate_confusions[i][r[-2]][cl] += 1
for i in map(lambda x:x/10., range(12)):
cl = r[-4] >= i
aggregate_score_confusions[i][r[-2]][cl] += 1
if opts.full:
csvfull = csv.writer(flouts)
csvfull.writerow(['trueprobs','sum']+header + ['average_score','uid','link','office_level'])
for file_counter in range(num_full):
print len(zip(*full_predictions[file_counter]))
fulldf_avgs = map(lambda a:sum(a)/len(a),zip(*full_df[file_counter]))
print len(fulldf_avgs)
full_predictions[file_counter].append(fulldf_avgs)
full_predictions[file_counter].append(data_full[file_counter].link)
full_predictions[file_counter].append(data_full[file_counter].data)
print len(zip(*full_predictions[file_counter]))
try:
for r in zip(*full_predictions[file_counter]):
cps = class_prob(conditional_probabilities, class_probabilities, r[:-3],fits, r[-3])
csvfull.writerow(['{0:1.5f}'.format(cps[1]),sum(map(lambda x: -1 if x==2 else x,r[:-3]))]+map(str,r[:-3]) + [r[-3],eval(r[-1])['uid'],r[-2],eval(r[-1])['office_level']])
except Exception as error:
import pdb;pdb.set_trace()
for k,v in aggregate_confusions.iteritems():
print k
print np.array_repr(v,precision=0,suppress_small=True)
keys = aggregate_score_confusions.keys()
keys.sort()
for k in keys:
v = aggregate_score_confusions[k]
print k
print np.array_repr(v,precision=0,suppress_small=True)
