dacapo.compute_context

Submodules

• dacapo.compute_context.bsub
• dacapo.compute_context.compute_context
• dacapo.compute_context.local_torch

Classes

ComputeContext	Helper class that provides a standard way to create an ABC using
LocalTorch	The LocalTorch class is a subclass of the ComputeContext class.
Bsub	Helper class that provides a standard way to create an ABC using

Functions

create_compute_context(→ ComputeContext)

Create a compute context based on the global DaCapo options.

Package Contents

class dacapo.compute_context.ComputeContext

Helper class that provides a standard way to create an ABC using inheritance.

distribute_workers: bool | None

The ComputeContext class is an abstract base class for defining the context in which computations are to be done. It is inherited from the built-in class ABC (Abstract Base Classes). Other classes can inherit this class to define their own specific variations of the context. It requires to implement several property methods, and also includes additional methods related to the context design.

device

The device on which computations are to be done.

_wrap_command(command)

Wraps a command in the context specific command.

wrap_command(command)

Wraps a command in the context specific command and returns it.

execute(command)

Runs a command in the context specific way.

Note

The class is abstract and requires to implement the abstract methods.

abstract property device

Abstract property method to define the device on which computations are to be done.

A device can be a CPU, GPU, TPU, etc. It is used to specify the context in which computations are to be done.

Returns:

The device on which computations are to be done.

Return type:

str

Raises:

NotImplementedError – If the method is not implemented in the derived class.

Examples

>>> context = ComputeContext()
>>> device = context.device

Note

The method should be implemented in the derived class.

wrap_command(command)

A method to wrap a command in the context specific command and return it.

Parameters:

command (List[str]) – The command to be wrapped.

Returns:

The wrapped command.

Return type:

List[str]

Raises:

NotImplementedError – If the method is not implemented in the derived class.

Examples

>>> context = ComputeContext()
>>> command = ["python", "script.py"]
>>> wrapped_command = context.wrap_command(command)

Note

The method should be implemented in the derived class.

execute(command)

A method to run a command in the context specific way.

Parameters:

command (List[str]) – The command to be executed.

Raises:

NotImplementedError – If the method is not implemented in the derived class.

Examples

>>> context = ComputeContext()
>>> command = ["python", "script.py"]
>>> context.execute(command)

Note

The method should be implemented in the derived class.

dacapo.compute_context.create_compute_context() → ComputeContext

Create a compute context based on the global DaCapo options.

Returns:

The compute context object.

Return type:

ComputeContext

Raises:

ValueError – If the store type is unknown.

Examples

>>> context = create_compute_context()

Note

The method is implemented in the module.

class dacapo.compute_context.LocalTorch

The LocalTorch class is a subclass of the ComputeContext class. It is used to specify the context in which computations are to be done. LocalTorch is used to specify that computations are to be done on the local machine using PyTorch.

_device

This stores the type of device on which torch computations are to be done. It can

Type:

Optional[str]

take "cuda" for GPU or "cpu" for CPU. None value results in automatic detection of device type.

oom_limit

The out of GPU memory to leave free in GB. If the free memory is below

Type:

Optional[float | int]

this limit, we will fall back on CPU.

device()

Returns the torch device object.

Note

The class is a subclass of the ComputeContext class.

distribute_workers: bool | None

The ComputeContext class is an abstract base class for defining the context in which computations are to be done. It is inherited from the built-in class ABC (Abstract Base Classes). Other classes can inherit this class to define their own specific variations of the context. It requires to implement several property methods, and also includes additional methods related to the context design.

device

The device on which computations are to be done.

_wrap_command(command)

Wraps a command in the context specific command.

wrap_command(command)

Wraps a command in the context specific command and returns it.

execute(command)

Runs a command in the context specific way.

Note

The class is abstract and requires to implement the abstract methods.

oom_limit: float | int | None

property device

A property method that returns the torch device object. It automatically detects and uses "cuda" (GPU) if

available, else it falls back on using "cpu".

Returns:

The torch device object.

Return type:

torch.device

class dacapo.compute_context.Bsub

Helper class that provides a standard way to create an ABC using inheritance.

distribute_workers: bool | None

The Bsub class is a subclass of the ComputeContext class. It is used to specify the context in which computations are to be done. Bsub is used to specify that computations are to be done on a cluster using LSF.

queue

The queue to run on.

Type:

str

num_gpus

The number of gpus to train on. Currently only 1 gpu can be used.

Type:

int

num_cpus

The number of cpus to use to generate training data.

Type:

int

billing

Project name that will be paying for this Job.

Type:

Optional[str]

device()

Returns the device on which computations are to be done.

_wrap_command(command)

Wraps a command in the context specific command.

Note

The class is a subclass of the ComputeContext class.

queue: str

num_gpus: int

num_cpus: int

billing: str | None

property device

A property method that returns the device on which computations are to be done.

A device can be a CPU, GPU, TPU, etc. It is used to specify the context in which computations are to be done.

Returns:

The device on which computations are to be done.

Return type:

str

Examples

>>> context = Bsub()
>>> device = context.device