def get_reduction_of_data(aggregation_name, tensor_data, tensor_name, abs=False):
reduction_name = aggregation_name
# If tensor_data is of np.ndarray type invoke np operators.
if isinstance(tensor_data, np.ndarray):
return get_numpy_reduction(reduction_name, tensor_data, abs)
if abs:
tensor_data = mx.ndarray.abs(tensor_data)
if reduction_name in ALLOWED_REDUCTIONS:
if hasattr(mx.ndarray, aggregation_name):
f = getattr(mx.ndarray, aggregation_name)
op = f(tensor_data, name=tensor_name)
else:
# If aggregation is not supported by mxnet, we convert data into np.ndarray and invoke numpy operators
tensor_data_np = make_numpy_array(tensor_data)
op = get_numpy_reduction(aggregation_name, numpy_data=tensor_data_np, abs=abs)
return op
elif reduction_name in ALLOWED_NORMS:
if reduction_name in ["l1", "l2"]:
op = mx.ndarray.norm(data=tensor_data, ord=int(reduction_name[1]))
return op
else:
raise RuntimeError(
"Invalid normalization operation {0} for mx.NDArray".format(reduction_name)
)
elif hasattr(mx, reduction_name):
f = getattr(mx, reduction_name)
op = f(tensor_data, name=tensor_name)
return op
elif hasattr(np, aggregation_name):
f = getattr(np, aggregation_name)
op = f(tensor_data)
return op
raise RuntimeError("Invalid aggregation_name {0} for mx.NDArray".format(aggregation_name))
def histogram_summary(tag, values, bins, max_bins=None):
"""Outputs a `Summary` protocol buffer with a histogram.
Adding a histogram summary makes it possible to visualize the data's distribution in
TensorBoard. See detailed explanation of the TensorBoard histogram dashboard at
https://www.tensorflow.org/get_started/tensorboard_histograms
This op reports an `InvalidArgument` error if any value is not finite.
Adapted from the TensorFlow function `histogram()` at
https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/summary/summary.py
and Pytorch's function
https://github.com/pytorch/pytorch/blob/2655b2710c8f0b3253fe2cfe0d2674f5d3592d22/torch/utils/tensorboard/summary.py#L232
Parameters
----------
tag : str
A name for the summary of the histogram. Will also serve as a series name in
TensorBoard.
values : `numpy.ndarray`
Values for building the histogram.
Returns
-------
A `Summary` protobuf of the histogram.
"""
tag = _clean_tag(tag)
values = make_numpy_array(values)
hist = _make_histogram(values.astype(float), bins, max_bins)
return Summary(value=[Summary.Value(tag=tag, histo=hist)])
def test_make_numpy_array():
simple_numpy_array = np.ndarray(shape=(2, 2), dtype=float, order="F")
# Check support for ndarray
try:
x = make_numpy_array(simple_numpy_array)
assert x.all() == simple_numpy_array.all()
except:
assert False
# Check support for scalar
simple_scalar = "foo"
try:
x = make_numpy_array(simple_scalar)
assert x == np.array([simple_scalar])
except:
assert False
# Check support for tuple
simple_tuple = (0.5, 0.7)
try:
x = make_numpy_array(simple_tuple)
assert x.all() == np.array(simple_tuple).all()
except:
assert False
# Check support for list
simple_list = [0.5, 0.7]
try:
x = make_numpy_array(simple_list)
assert x.all() == np.array(simple_list).all()
except:
assert False
# Check support for dict
simple_dict = {"a": 0.5, "b": 0.7}
try:
x = make_numpy_array(simple_dict)
assert x == np.array(simple_dict)
except:
assert False
def _make_numpy_array(tensor_value):
"""
Convert the tensor value into a numpy array.
Here it's already numpy array
"""
if is_tf_version_2x() and tf.executing_eagerly():
if (isinstance(tensor_value, tf.Variable)
or isinstance(tensor_value, tf.Tensor)) and hasattr(
tensor_value, "numpy"):
# TF 2.X eager mode
return tensor_value.numpy()
return make_numpy_array(tensor_value)
def scalar_summary(tag, scalar):
"""Outputs a `Summary` protocol buffer containing a single scalar value.
The generated Summary has a Tensor.proto containing the input Tensor.
Adapted from the TensorFlow function `scalar()` at
https://github.com/tensorflow/tensorflow/blob/r1.6/tensorflow/python/summary/summary.py
Parameters
----------
tag : str
A name for the generated summary. Will also serve as the series name in TensorBoard.
scalar : int, MXNet `NDArray`, or `numpy.ndarray`
A scalar value or an ndarray of shape (1,).
Returns
-------
A `Summary` protobuf of the `scalar` value.
Raises
------
ValueError: If the scalar has the wrong shape or type.
"""
tag = _clean_tag(tag)
scalar = make_numpy_array(scalar)
assert scalar.squeeze().ndim == 0, "scalar should be 0D"
scalar = float(scalar)
return Summary(value=[Summary.Value(tag=tag, simple_value=scalar)])
def _make_numpy_array(tensor_value):
"""
Convert the tensor value into a numpy array.
Here it's already numpy array
"""
return make_numpy_array(tensor_value)
def _make_numpy_array(tensor_value):
if isinstance(tensor_value, torch.Tensor):
return tensor_value.to(torch.device("cpu")).data.numpy()
return make_numpy_array(tensor_value)
def _make_numpy_array(tensor_value):
return make_numpy_array(tensor_value)
