def plot_op(fn, inputs=[], outputs=[]):
"""
User-exposed api method for constructing a python_node
Args:
fn: python function that computes some np.ndarrays given np.ndarrays as inputs. it can have arbitrary side effects.
inputs: array of tf.Tensors (optional). These are where fn derives its values from
outputs: tf.Placeholder nodes (optional). These are constructed by the user (which allows the user to
plug them into other ht.Ops or tf.Ops). The outputs of fn are mapped to each of the output placeholders.
raises an Error if fn cannot map
"""
global COUNT, ht
# check outputs
if not isinstance(outputs,list):
outputs=[outputs]
for tensor in outputs:
if tensor.op.type is not 'Placeholder':
raise Error('Output nodes must be Placeholders')
op=PlotOp(fn, COUNT, inputs, outputs)
op_store.add_op(op)
COUNT+=1
# if node has output, return value for python_op is the first output (placeholder) tensor
# otherwise, return the op
if outputs:
return outputs[0]
else:
return op
def python_op(fn, inputs=None, outputs=None):
"""
User-exposed api method for constructing a python_node
Args:
fn: python function that computes some np.ndarrays given np.ndarrays as inputs. it can have arbitrary side effects.
inputs: array of tf.Tensors (optional). These are where fn derives its values from
outputs: tf.Placeholder nodes (optional). These are constructed by the user (which allows the user to
plug them into other ht.Ops or tf.Ops). The outputs of fn are mapped to each of the output placeholders.
raises an Error if fn cannot map
"""
# construct a PythonOp and return its TensorNode outputs, if it has one
global COUNT
# check outputs
if not isinstance(outputs, list):
outputs = [outputs]
for tensor in outputs:
if tensor.op.type != 'Placeholder':
raise TypeError('Output nodes must be Placeholders')
op = PythonOp('Python', fn, COUNT, inputs, outputs)
op_store.add_op(op)
COUNT += 1
if outputs:
return outputs[0]
else:
return op
def plot_op(fn, inputs=[], outputs=[]):
"""
User-exposed api method for constructing a python_node
Args:
fn: python function that computes some np.ndarrays given np.ndarrays as inputs. it can have arbitrary side effects.
inputs: array of tf.Tensors (optional). These are where fn derives its values from
outputs: tf.Placeholder nodes (optional). These are constructed by the user (which allows the user to
plug them into other ht.Ops or tf.Ops). The outputs of fn are mapped to each of the output placeholders.
raises an Error if fn cannot map
"""
global COUNT, ht
# check outputs
if not isinstance(outputs, list):
outputs = [outputs]
for tensor in outputs:
if tensor.op.type is not 'Placeholder':
raise Error('Output nodes must be Placeholders')
op = PlotOp(fn, COUNT, inputs, outputs)
op_store.add_op(op)
COUNT += 1
# if node has output, return value for python_op is the first output (placeholder) tensor
# otherwise, return the op
if outputs:
return outputs[0]
else:
return op
def python_op(fn, inputs=None, outputs=None):
"""
User-exposed api method for constructing a python_node
Args:
fn: python function that computes some np.ndarrays given np.ndarrays as inputs. it can have arbitrary side effects.
inputs: array of tf.Tensors (optional). These are where fn derives its values from
outputs: tf.Placeholder nodes (optional). These are constructed by the user (which allows the user to
plug them into other ht.Ops or tf.Ops). The outputs of fn are mapped to each of the output placeholders.
raises an Error if fn cannot map
"""
# construct a PythonOp and return its TensorNode outputs, if it has one
global COUNT
# check outputs
if not isinstance(outputs,list):
outputs=[outputs]
for tensor in outputs:
if tensor.op.type != 'Placeholder':
raise TypeError('Output nodes must be Placeholders')
op=PythonOp('Python', fn, COUNT, inputs, outputs)
op_store.add_op(op)
COUNT+=1
if outputs:
return outputs[0]
else:
return op