def __get_version(version):
# matching 1.6.1, and 1.6.1rc, 1.6.1.dev
version_regex = '^\d+\.\d+\.\d+'
version = _re.search(version_regex, str(version)).group(0)
return _StrictVersion(version)
version_regex = '^\d+\.\d+\.\d+'
version = _re.search(version_regex, str(version)).group(0)
return _StrictVersion(version)
# ---------------------------------------------------------------------------------------
HAS_SKLEARN = True
SKLEARN_MIN_VERSION = '0.15'
def __get_sklearn_version(version):
# matching 0.15b, 0.16bf, etc
version_regex = '^\d+\.\d+'
version = _re.search(version_regex, str(version)).group(0)
return _StrictVersion(version)
try:
import sklearn
if __get_sklearn_version(sklearn.__version__) < _StrictVersion(SKLEARN_MIN_VERSION):
HAS_SKLEARN = False
_logging.warn(('scikit-learn version %s is not supported. Minimum required version: %s. '
'Disabling scikit-learn conversion API.')
% (sklearn.__version__, SKLEARN_MIN_VERSION) )
except:
HAS_SKLEARN = False
# ---------------------------------------------------------------------------------------
HAS_LIBSVM = True
try:
import svm
except:
HAS_LIBSVM = False
# ---------------------------------------------------------------------------------------
HAS_SKLEARN = True
SKLEARN_VERSION = None
SKLEARN_MIN_VERSION = '0.17'
def __get_sklearn_version(version):
# matching 0.15b, 0.16bf, etc
version_regex = '^\d+\.\d+'
version = _re.search(version_regex, str(version)).group(0)
return _StrictVersion(version)
try:
import sklearn
SKLEARN_VERSION = __get_sklearn_version(sklearn.__version__)
if SKLEARN_VERSION < _StrictVersion(SKLEARN_MIN_VERSION):
HAS_SKLEARN = False
_logging.warn((
'scikit-learn version %s is not supported. Minimum required version: %s. '
'Disabling scikit-learn conversion API.') %
(sklearn.__version__, SKLEARN_MIN_VERSION))
except:
HAS_SKLEARN = False
# ---------------------------------------------------------------------------------------
HAS_LIBSVM = True
try:
import svm
except:
HAS_LIBSVM = False
_SKLEARN_MIN_VERSION = "0.17"
_SKLEARN_MAX_VERSION = "0.19.2"
def __get_sklearn_version(version):
# matching 0.15b, 0.16bf, etc
version_regex = r"^\d+\.\d+"
version = _re.search(version_regex, str(version)).group(0)
return _StrictVersion(version)
try:
import sklearn
_SKLEARN_VERSION = __get_sklearn_version(sklearn.__version__)
if _SKLEARN_VERSION < _StrictVersion(
_SKLEARN_MIN_VERSION) or _SKLEARN_VERSION > _StrictVersion(
_SKLEARN_MAX_VERSION):
_HAS_SKLEARN = False
_logging.warning((
"scikit-learn version %s is not supported. Minimum required version: %s. "
"Maximum required version: %s. "
"Disabling scikit-learn conversion API.") %
(sklearn.__version__, _SKLEARN_MIN_VERSION,
_SKLEARN_MAX_VERSION))
except:
_HAS_SKLEARN = False
MSG_SKLEARN_NOT_FOUND = "Sklearn not found."
# ---------------------------------------------------------------------------------------
_HAS_LIBSVM = True
try:
_keras.layers.Concatenate: _layers2.convert_merge,
_keras.layers.Dot: _layers2.convert_merge,
_keras.layers.core.Flatten: _layers2.convert_flatten,
_keras.layers.core.Permute: _layers2.convert_permute,
_keras.layers.core.Reshape: _layers2.convert_reshape,
_keras.layers.embeddings.Embedding: _layers2.convert_embedding,
_keras.layers.core.RepeatVector: _layers2.convert_repeat_vector,
_keras.layers.core.Dropout: _layers2.default_skip,
_keras.layers.core.SpatialDropout2D: _layers2.default_skip,
_keras.layers.core.SpatialDropout1D: _layers2.default_skip,
_keras.layers.wrappers.TimeDistributed: _layers2.default_skip,
}
from distutils.version import StrictVersion as _StrictVersion
## 2.2 Version check
if _keras.__version__ >= _StrictVersion("2.2.0"):
_KERAS_LAYER_REGISTRY[
_keras.layers.DepthwiseConv2D
] = _layers2.convert_convolution
_KERAS_LAYER_REGISTRY[
_keras.engine.input_layer.InputLayer
] = _layers2.default_skip
if _keras.__version__ >= _StrictVersion("2.2.1"):
_KERAS_LAYER_REGISTRY[
_keras.layers.advanced_activations.ReLU
] = _layers2.convert_advanced_relu
else:
_KERAS_LAYER_REGISTRY[
_keras.applications.mobilenet.DepthwiseConv2D
] = _layers2.convert_convolution
_KERAS_LAYER_REGISTRY[_keras.engine.topology.InputLayer] = _layers2.default_skip
def get_game_version(vstring):
return _StrictVersion(vstring)
def _from_saved_model(saved_model_dir):
from tensorflow.python.tools import freeze_graph
# must import here as tf.contrib is only available on TF 1.x
from tensorflow.contrib.saved_model.python.saved_model import reader
saved_model_tags = reader.get_saved_model_tag_sets(saved_model_dir)[0]
if not saved_model_tags:
msg = "Unsupported SavedModel directory format: no tag_sets available"
raise NotImplementedError(msg)
# get model outputs
output_node_names = []
if _get_version(tf.__version__) < _StrictVersion("1.13.1"):
sess = tf.Session()
else:
sess = tf.compat.v1.Session()
metagraph = tf.saved_model.loader.load(sess, saved_model_tags,
saved_model_dir)
for sd in metagraph.signature_def.values():
output_node_names += [
o.name.split(":")[0] for o in sd.outputs.values()
]
sess.close()
# get frozen graph
output_graph = mktemp()
tf.compat.v1.reset_default_graph() if _get_version(
tf.__version__) >= _StrictVersion(
"1.13.1") else tf.reset_default_graph()
freeze_graph.freeze_graph(
input_graph=None,
input_saver=None,
input_binary=None,
input_checkpoint=None,
output_node_names=",".join(output_node_names),
restore_op_name=None,
filename_tensor_name=None,
output_graph=output_graph,
clear_devices=True,
initializer_nodes="",
variable_names_whitelist="",
variable_names_blacklist="",
input_meta_graph=None,
input_saved_model_dir=saved_model_dir,
saved_model_tags=",".join(saved_model_tags),
)
if _get_version(tf.__version__) < _StrictVersion("1.13.1"):
graph_def = tf.GraphDef()
with open(output_graph, "rb") as f:
graph_def.ParseFromString(f.read())
graph_def = tf.graph_util.remove_training_nodes(graph_def)
else:
graph_def = tf.compat.v1.GraphDef()
with open(output_graph, "rb") as f:
graph_def.ParseFromString(f.read())
graph_def = tf.compat.v1.graph_util.remove_training_nodes(
graph_def)
with tf.Graph().as_default() as graph:
tf.graph_util.import_graph_def(graph_def, name="")
return graph.as_graph_def(add_shapes=True)
def global_game_version():
global _GameVersion
if _GameVersion is None:
_GameVersion = _StrictVersion(C.Game.Version)
return _GameVersion
def get_game_version(vstring):
return _StrictVersion(vstring)
_keras.layers.core.Flatten: _layers2.convert_flatten,
_keras.layers.core.Permute:_layers2.convert_permute,
_keras.layers.core.Reshape:_layers2.convert_reshape,
_keras.layers.embeddings.Embedding:_layers2.convert_embedding,
_keras.layers.core.RepeatVector:_layers2.convert_repeat_vector,
_keras.layers.core.Dropout:_layers2.default_skip,
_keras.layers.core.SpatialDropout2D:_layers2.default_skip,
_keras.layers.core.SpatialDropout1D:_layers2.default_skip,
_keras.layers.wrappers.TimeDistributed:_layers2.default_skip,
}
from distutils.version import StrictVersion as _StrictVersion
## 2.2 Version check
keras_version = _keras.__version__.rstrip('-tf')
if keras_version >= _StrictVersion('2.2.0'):
_KERAS_LAYER_REGISTRY[_keras.layers.DepthwiseConv2D] = (
_layers2.convert_convolution
)
_KERAS_LAYER_REGISTRY[_keras.engine.input_layer.InputLayer] = (
_layers2.default_skip
)
if keras_version >= _StrictVersion('2.2.1'):
_KERAS_LAYER_REGISTRY[_keras.layers.advanced_activations.ReLU] = (
_layers2.convert_advanced_relu
)
else:
_KERAS_LAYER_REGISTRY[_keras.applications.mobilenet.DepthwiseConv2D] = (
_layers2.convert_convolution
)
_KERAS_LAYER_REGISTRY[_keras.engine.topology.InputLayer] = (
def _constant_propagation(fn, new_graph, constant_nodes, constant_node_num_outputs):
try:
if len(constant_nodes) > 0:
with tf.Graph().as_default() as graph:
tf.import_graph_def(new_graph, name="")
# We're only making one call to `sess.run()` in order to compute constant values.
# In this context, the default optimization settings make everything dramatically
# slower and more memory-intensive.
if tf.__version__ < _StrictVersion("1.13.1"):
session_config = tf.ConfigProto()
session_config.graph_options.optimizer_options.opt_level = (
tf.OptimizerOptions.L0
)
sess = tf.Session(graph=graph, config=session_config)
else:
session_config = tf.compat.v1.ConfigProto()
session_config.graph_options.optimizer_options.opt_level = (
tf.compat.v1.OptimizerOptions.L0
)
session_config.graph_options.rewrite_options.disable_meta_optimizer = (
True
)
sess = tf.compat.v1.Session(graph=graph, config=session_config)
query_list = list()
control_flow_ops = list()
for c in constant_nodes:
for j in range(constant_node_num_outputs[c]):
query = c + ":" + str(j)
lower_query = query.lower()
if "switch" in lower_query or "cond" in lower_query:
control_flow_ops.append(query)
else:
query_list.append(query)
result_list = sess.run(query_list)
result = {
query_list[i]: result_list[i] for i in range(len(query_list))
}
# propagate switch one by one
for op in control_flow_ops:
try:
res = sess.run([op])
result.update({op: res[0]})
except:
logging.warning(
'[Constant Propagation] Skip "dead" tensor: {}'.format(
op
)
)
result.update({op: None})
sess.close()
for k, v in fn.graph.items():
if k in constant_node_num_outputs:
if constant_node_num_outputs[k] == 1:
result_entry = k + ":0"
try:
v.value, v.datatype = numpy_val_to_builtin_val(
result[result_entry]
)
except:
logging.error(result_entry)
logging.error(result[result_entry])
else:
values = [
result[k + ":" + str(i)]
for i in range(constant_node_num_outputs[k])
]
try:
npval = [numpy_val_to_builtin_val(i) for i in values]
v.datatype = types.tuple(tuple([val[1] for val in npval]))
v.value = v.datatype()
for idx, val in enumerate(npval):
v.value.val[idx] = val[0]
except:
logging.error(values)
for k, v in fn.graph.items():
if v.op == "get_tuple":
inp = fn.graph[v.inputs[0]]
idx = v.attr["index"]
if inp.value is not None:
v.value = inp.value.val[idx]
v.datatype = inp.datatype.T[idx]
except Exception as e:
logging.exception("Constant Propagation pass failed: {}".format(e))
def test_conv(
self,
use_cpu_only,
backend,
op,
padding,
data_format,
spatial_dim_and_ks,
strides,
dilations,
batch_size,
groups,
):
# tensorflow supports groupwise convolution only for version > tf.2.5.0-rc3
if _get_version(
_tf.__version__) < _StrictVersion("2.5.0") and groups != 1:
return
if op == tf.keras.layers.Conv3D and groups != 1:
pytest.xfail(
"rdar://81629932 (Conv3d with group > 1 tests failing in TF2.0 converter)"
)
# TF does not support strides > 1 in conjunction with dilation_rate > 1
for i, stride in enumerate(strides):
if stride > 1 and dilations[i] > 1:
return
# Dilations with Conv3D not supported yet, since SpaceToBatchND is only supported for ranks 3 or 4
for d in dilations:
if d > 1 and op == tf.keras.layers.Conv3D:
return
s1, s2, s3, k1, k2, k3 = spatial_dim_and_ks
c_in, c_out = 2, 4
input_shape = None
kernel_size = None
if op == tf.keras.layers.Conv1D:
input_shape = (batch_size, s3, c_in)
kernel_size = k3
strides = strides[2]
dilations = dilations[2]
elif op == tf.keras.layers.Conv2D:
input_shape = (batch_size, s2, s3, c_in)
kernel_size = (k2, k3)
strides = (strides[1], strides[2])
dilations = dilations[1:]
elif op == tf.keras.layers.Conv3D:
input_shape = (batch_size, s1, s2, s3, c_in)
kernel_size = (k1, k2, k3)
model = tf.keras.Sequential([
op(
batch_input_shape=input_shape,
filters=c_out,
kernel_size=kernel_size,
strides=strides,
padding=padding.upper(),
data_format=data_format,
dilation_rate=dilations,
groups=groups,
)
])
TensorFlowBaseTest.run_compare_tf_keras(
model,
[random_gen(input_shape, rand_min=-10, rand_max=10)],
use_cpu_only=use_cpu_only,
backend=backend,
)
_keras.layers.Multiply,
_keras.layers.Average,
_keras.layers.Maximum,
_keras.layers.Concatenate,
_keras.layers.Dot,
]
_KERAS_SKIP_LAYERS = [
_keras.layers.core.Dropout,
_keras.layers.core.SpatialDropout1D,
_keras.layers.core.SpatialDropout2D,
]
from distutils.version import StrictVersion as _StrictVersion
if _keras.__version__.rstrip('-tf') >= _StrictVersion('2.2.0'):
from tensorflow.python.keras.engine.input_layer import InputLayer
else:
from keras.engine.topology import InputLayer
def _to_list(x):
if type(x) is not list:
return [x]
else:
return x
def _insert_to_dict(d, key, e):
# d is a dict where key maps to a list
if key not in d:
# ---------------------------------------------------------------------------------------
HAS_SKLEARN = True
SKLEARN_MIN_VERSION = '0.15'
def __get_sklearn_version(version):
# matching 0.15b, 0.16bf, etc
version_regex = '^\d+\.\d+'
version = _re.search(version_regex, str(version)).group(0)
return _StrictVersion(version)
try:
import sklearn
if __get_sklearn_version(
sklearn.__version__) < _StrictVersion(SKLEARN_MIN_VERSION):
HAS_SKLEARN = False
_logging.warn((
'scikit-learn version %s is not supported. Minimum required version: %s. '
'Disabling scikit-learn conversion API.') %
(sklearn.__version__, SKLEARN_MIN_VERSION))
except:
HAS_SKLEARN = False
# ---------------------------------------------------------------------------------------
HAS_LIBSVM = True
try:
import svm
except:
HAS_LIBSVM = False
def __get_sklearn_version(version):
# matching 0.15b, 0.16bf, etc
version_regex = '^\d+\.\d+'
version = _re.search(version_regex, str(version)).group(0)
return _StrictVersion(version)
from turicreate._deps.minimal_package import is_minimal_pkg
def __get_version(version):
# matching 1.6.1, and 1.6.1rc, 1.6.1.dev
version_regex = "^\d+\.\d+\.\d+"
version = _re.search(version_regex, str(version)).group(0)
return _StrictVersion(version)
HAS_PANDAS = True
PANDAS_MIN_VERSION = "0.13.0"
try:
import pandas
if __get_version(pandas.__version__) < _StrictVersion(PANDAS_MIN_VERSION):
HAS_PANDAS = False
_logging.warn((
"Pandas version %s is not supported. Minimum required version: %s. "
"Pandas support will be disabled.") %
(pandas.__version__, PANDAS_MIN_VERSION))
except:
HAS_PANDAS = False
from . import pandas_mock as pandas
HAS_NUMPY = True
NUMPY_MIN_VERSION = "1.8.0"
try:
import numpy
if __get_version(numpy.__version__) < _StrictVersion(NUMPY_MIN_VERSION):
_keras.layers.Multiply,
_keras.layers.Average,
_keras.layers.Maximum,
_keras.layers.Concatenate,
_keras.layers.Dot,
]
_KERAS_SKIP_LAYERS = [
_keras.layers.core.Dropout,
_keras.layers.core.SpatialDropout1D,
_keras.layers.core.SpatialDropout2D,
]
from distutils.version import StrictVersion as _StrictVersion
if _keras.__version__ >= _StrictVersion('2.2.0'):
from keras.engine.input_layer import InputLayer
else:
from keras.engine.topology import InputLayer
def _to_list(x):
if type(x) is not list:
return [x]
else:
return x
def _insert_to_dict(d, key, e):
# d is a dict where key maps to a list
if key not in d:
def global_game_version():
global _GameVersion
if _GameVersion is None:
_GameVersion = _StrictVersion(C.Game.Version)
return _GameVersion
