Pipelines

class stride.optimisation.pipelines.pipeline.Pipeline(steps=None, **kwargs)

Bases: stride.core.Operator

A pipeline represents a series of processing steps that will be applied in order to a series of inputs. Pipelines encode pre-processing or post-processing steps such as filtering time traces or smoothing a gradient.

Parameters

steps (list, optional) – List of steps that form the pipeline. Steps can be callable or strings pointing to a default, pre-defined step.

async adjoint(*args, **kwargs)

Method defining the adjoint behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method will be called with positional arguments comprised of: the gradients of every output of the forward operation, followed by the arguments originally given when calling the forward method.

The adjoint method needs to return a gradient for each of its Variable inputs (or None if the variable does not needs_grad).

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

property comms

async forward(*args, **kwargs)

Apply all steps in the pipeline in order.

property head

classmethod local(*args, uid=None, **kwargs)

classmethod local_parameter(*args, uid=None, **kwargs)

property logger

property loop

property monitor

property node

classmethod parameter(*args, uid=None, **kwargs)

classmethod remote(*args, **kwargs)

property runtime

async classmethod select_worker(runtime=None)

Select an available worker.

Parameters

runtime (str or Runtime, optional) – If a valid runtime is given, this will be selected as the target worker.

Returns

UID of the target worker.

Return type

str

property worker

property zmq_context

Default pipelines

class stride.optimisation.pipelines.default_pipelines.ProcessGlobalGradient(steps=None, no_grad=True, **kwargs)

Bases: stride.optimisation.pipelines.pipeline.Pipeline

Default pipeline to process the global gradient before updating the variable.

Default steps:

• mask

• smooth_field

• norm_field

property comms

property head

classmethod local(*args, uid=None, **kwargs)

classmethod local_parameter(*args, uid=None, **kwargs)

property logger

property loop

property monitor

property node

classmethod parameter(*args, uid=None, **kwargs)

classmethod remote(*args, **kwargs)

property runtime

async classmethod select_worker(runtime=None)

Select an available worker.

Parameters

runtime (str or Runtime, optional) – If a valid runtime is given, this will be selected as the target worker.

Returns

UID of the target worker.

Return type

str

property worker

property zmq_context

class stride.optimisation.pipelines.default_pipelines.ProcessModelIteration(steps=None, no_grad=True, **kwargs)

Bases: stride.optimisation.pipelines.pipeline.Pipeline

Default pipeline to process the model after each iteration.

Default steps:

• clip

property comms

property head

classmethod local(*args, uid=None, **kwargs)

classmethod local_parameter(*args, uid=None, **kwargs)

property logger

property loop

property monitor

property node

classmethod parameter(*args, uid=None, **kwargs)

classmethod remote(*args, **kwargs)

property runtime

async classmethod select_worker(runtime=None)

Select an available worker.

Parameters

runtime (str or Runtime, optional) – If a valid runtime is given, this will be selected as the target worker.

Returns

UID of the target worker.

Return type

str

property worker

property zmq_context

class stride.optimisation.pipelines.default_pipelines.ProcessTraces(steps=None, no_grad=False, **kwargs)

Bases: stride.optimisation.pipelines.pipeline.Pipeline

Default pipeline to process modelled and observed before running the functional.

Default steps:

• mute_traces

• filter_traces

• norm_per_shot

property comms

property head

classmethod local(*args, uid=None, **kwargs)

classmethod local_parameter(*args, uid=None, **kwargs)

property logger

property loop

property monitor

property node

classmethod parameter(*args, uid=None, **kwargs)

classmethod remote(*args, **kwargs)

property runtime

async classmethod select_worker(runtime=None)

Select an available worker.

Parameters

runtime (str or Runtime, optional) – If a valid runtime is given, this will be selected as the target worker.

Returns

UID of the target worker.

Return type

str

property worker

property zmq_context

class stride.optimisation.pipelines.default_pipelines.ProcessWavelets(steps=None, no_grad=False, **kwargs)

Bases: stride.optimisation.pipelines.pipeline.Pipeline

Default pipeline to process wavelets before running the forward problem.

Default steps:

• filter_wavelets

property comms

property head

classmethod local(*args, uid=None, **kwargs)

classmethod local_parameter(*args, uid=None, **kwargs)

property logger

property loop

property monitor

property node

classmethod parameter(*args, uid=None, **kwargs)

classmethod remote(*args, **kwargs)

property runtime

async classmethod select_worker(runtime=None)

Select an available worker.

Parameters

runtime (str or Runtime, optional) – If a valid runtime is given, this will be selected as the target worker.

Returns

UID of the target worker.

Return type

str

property worker

property zmq_context

Steps

class stride.optimisation.pipelines.steps.filter_traces.FilterTraces(**kwargs)

Bases: stride.core.Operator

Filter a set of time traces.

Parameters

• f_min (float, optional) – Lower value for the frequency filter, defaults to None (no lower filtering).

• f_max (float, optional) – Upper value for the frequency filter, defaults to None (no upper filtering).

• filter_type (str, optional) – Type of filter to apply, from butterworth (default for band pass and high pass), fir, or cos (default for low pass).

adjoint(*d_traces, **kwargs)

Method defining the adjoint behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method will be called with positional arguments comprised of: the gradients of every output of the forward operation, followed by the arguments originally given when calling the forward method.

The adjoint method needs to return a gradient for each of its Variable inputs (or None if the variable does not needs_grad).

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

forward(*traces, **kwargs)

Method defining the forward behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method can take multiple inputs and produce multiple outputs. Outputs of this method should be of type Variable.

Positional and keyword arguments to forward are processed so that present variables are tracked.

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

class stride.optimisation.pipelines.steps.mute_traces.MuteTraces(**kwargs)

Bases: stride.core.Operator

Mute traces with respect to each other.

adjoint(d_modelled, d_observed, modelled, observed, **kwargs)

Method defining the adjoint behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method will be called with positional arguments comprised of: the gradients of every output of the forward operation, followed by the arguments originally given when calling the forward method.

The adjoint method needs to return a gradient for each of its Variable inputs (or None if the variable does not needs_grad).

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

forward(modelled, observed, **kwargs)

Method defining the forward behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method can take multiple inputs and produce multiple outputs. Outputs of this method should be of type Variable.

Positional and keyword arguments to forward are processed so that present variables are tracked.

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

class stride.optimisation.pipelines.steps.filter_wavelets.FilterWavelets(**kwargs)

Bases: stride.core.Operator

Filter wavelets to 3/4 of the set frequencies.

Parameters

• f_min (float, optional) – Lower value for the frequency filter, defaults to None (no lower filtering).

• f_max (float, optional) – Upper value for the frequency filter, defaults to None (no upper filtering).

• filter_type (str, optional) – Type of filter to apply, from butterworth (default for band pass and high pass), fir, or cos (default for low pass).

adjoint(d_wavelets, wavelets, **kwargs)

Method defining the adjoint behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method will be called with positional arguments comprised of: the gradients of every output of the forward operation, followed by the arguments originally given when calling the forward method.

The adjoint method needs to return a gradient for each of its Variable inputs (or None if the variable does not needs_grad).

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

forward(wavelets, **kwargs)

Method defining the forward behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method can take multiple inputs and produce multiple outputs. Outputs of this method should be of type Variable.

Positional and keyword arguments to forward are processed so that present variables are tracked.

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

class stride.optimisation.pipelines.steps.norm_per_shot.NormPerShot(**kwargs)

Bases: stride.core.Operator

Normalised a series of time traces to the maximum value of the set.

Parameters

amplitude (bool, optional) – Whether to keep the true amplitude of the modelled after normalisation.

adjoint(*d_traces, **kwargs)

Method defining the adjoint behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method will be called with positional arguments comprised of: the gradients of every output of the forward operation, followed by the arguments originally given when calling the forward method.

The adjoint method needs to return a gradient for each of its Variable inputs (or None if the variable does not needs_grad).

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

forward(*traces, **kwargs)

Method defining the forward behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method can take multiple inputs and produce multiple outputs. Outputs of this method should be of type Variable.

Positional and keyword arguments to forward are processed so that present variables are tracked.

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

class stride.optimisation.pipelines.steps.norm_per_trace.NormPerTrace(**kwargs)

Bases: stride.core.Operator

Normalised a series of time traces individually.

Parameters

amplitude (bool, optional) – Whether to keep the true amplitude of the modelled after normalisation.

adjoint(*d_traces, **kwargs)

Method defining the adjoint behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method will be called with positional arguments comprised of: the gradients of every output of the forward operation, followed by the arguments originally given when calling the forward method.

The adjoint method needs to return a gradient for each of its Variable inputs (or None if the variable does not needs_grad).

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

forward(*traces, **kwargs)

Method defining the forward behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method can take multiple inputs and produce multiple outputs. Outputs of this method should be of type Variable.

Positional and keyword arguments to forward are processed so that present variables are tracked.

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

class stride.optimisation.pipelines.steps.smooth_field.SmoothField(**kwargs)

Bases: stride.core.Operator

Apply Gaussian smoothing to a StructuredData object.

Parameters

sigma (float, optional) – Standard deviation of the Gaussian kernel, defaults to 0.25 (25% of a grid point).

adjoint(d_field, field, **kwargs)

Method defining the adjoint behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method will be called with positional arguments comprised of: the gradients of every output of the forward operation, followed by the arguments originally given when calling the forward method.

The adjoint method needs to return a gradient for each of its Variable inputs (or None if the variable does not needs_grad).

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

forward(field, **kwargs)

Method defining the forward behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method can take multiple inputs and produce multiple outputs. Outputs of this method should be of type Variable.

Positional and keyword arguments to forward are processed so that present variables are tracked.

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

class stride.optimisation.pipelines.steps.norm_field.NormField(**kwargs)

Bases: stride.core.Operator

Normalise a StructuredData object between -1 and +1.

adjoint(d_field, field, **kwargs)

Method defining the adjoint behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method will be called with positional arguments comprised of: the gradients of every output of the forward operation, followed by the arguments originally given when calling the forward method.

The adjoint method needs to return a gradient for each of its Variable inputs (or None if the variable does not needs_grad).

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

forward(field, **kwargs)

Method defining the forward behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method can take multiple inputs and produce multiple outputs. Outputs of this method should be of type Variable.

Positional and keyword arguments to forward are processed so that present variables are tracked.

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

class stride.optimisation.pipelines.steps.clip.Clip(**kwargs)

Bases: stride.core.Operator

Clip data between two extreme values.

Parameters

• min (float, optional) – Lower value for the clipping, defaults to None (no lower clipping).

• max (float, optional) – Upper value for the clipping, defaults to None (no upper clipping).

adjoint(d_field, field, **kwargs)

Method defining the adjoint behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method will be called with positional arguments comprised of: the gradients of every output of the forward operation, followed by the arguments originally given when calling the forward method.

The adjoint method needs to return a gradient for each of its Variable inputs (or None if the variable does not needs_grad).

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

forward(field, **kwargs)

Method defining the forward behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method can take multiple inputs and produce multiple outputs. Outputs of this method should be of type Variable.

Positional and keyword arguments to forward are processed so that present variables are tracked.

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

class stride.optimisation.pipelines.steps.mask.Mask(**kwargs)

Bases: stride.core.Operator

Mask a StructuredData object to remove values outside inner domain.

adjoint(d_field, field, **kwargs)

Method defining the adjoint behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method will be called with positional arguments comprised of: the gradients of every output of the forward operation, followed by the arguments originally given when calling the forward method.

The adjoint method needs to return a gradient for each of its Variable inputs (or None if the variable does not needs_grad).

This method should not be called directly from user code.

Parameters

• args –

• kwargs –

forward(field, **kwargs)

Method defining the forward behaviour of the operator. This method needs to be defined by classes inheriting from the operator.

The method can take multiple inputs and produce multiple outputs. Outputs of this method should be of type Variable.

Positional and keyword arguments to forward are processed so that present variables are tracked.

This method should not be called directly from user code.

Parameters

• args –

• kwargs –