Base_individuals

Nn_individual

class gaggle.population.base_individuals.nn_individual.NNIndividual(individual_args: IndividualArgs, sys_args: SysArgs = None, model: Module = None, *args, **kwargs)

Bases: Individual

An Individual whose initial parameters are a torch.nn.Module.

CONFIG_BASE_MODEL_STATE_DICT = 'base_model_state_dict'

debug(mode: bool = True) → None

debug_tick() → None

Clears plotting for debugs.

eval()

Sets the module in evaluation mode.

This has any effect only on certain modules. See documentations of particular modules for details of their behaviors in training/evaluation mode, if they are affected, e.g. Dropout, BatchNorm, etc.

This is equivalent with self.train(False).

See locally-disable-grad-doc for a comparison between .eval() and several similar mechanisms that may be confused with it.

Returns:

self

Return type:

Module

first_time(name) → bool

Checks if something has been invoked for the first time

forward(x)

By default we returned a flattened tensor of the model parameters.

Notes

We do not return the metadata that can be used to reconstruct the gene_pool dictionary as this should not be used to modify the parameters directly (the tensor does not link back to the parameters).

Parameters:

• *args –

• **kwargs –

Returns:

A flattened pytorch tensor of length self.genome_size.

generate_gene_pool()

Should return a dictionary of dictionaries where the inner dictionary has a “param” and a “size” key for each of the nn.Parameters in the dictionary

initialize()

Represents the initialization rule for an individual. It should modify the gene pool.

Parameters:

• *args –

• **kwargs –

Returns:

self

load(content=None, ckpt=None) → Module

parameters(recurse: bool = True) → Iterator[Parameter]

Returns an iterator over module parameters.

This is typically passed to an optimizer.

Parameters:

recurse (bool) – if True, then yields parameters of this module and all submodules. Otherwise, yields only parameters that are direct members of this module.

Yields:

Parameter – module parameter

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for param in model.parameters():
>>>     print(type(param), param.size())
<class 'torch.Tensor'> (20L,)
<class 'torch.Tensor'> (20L, 1L, 5L, 5L)

save(outdir_args: OutdirArgs = None) → dict

Saves the individual to a file in the path provided by outdir_args.

Parameters:

outdir_args –

Returns:

train(mode: bool = True)

Sets the module in training mode.

This has any effect only on certain modules. See documentations of particular modules for details of their behaviors in training/evaluation mode, if they are affected, e.g. Dropout, BatchNorm, etc.

Parameters:

mode (bool) – whether to set training mode (True) or evaluation mode (False). Default: True.

Returns:

self

Return type:

Module

Numpy_individual

class gaggle.population.base_individuals.numpy_individual.NumpyIndividual(np_params: dict[slice(<class 'int'>, <class 'numpy.ndarray'>, None)] = None, individual_args: ~gaggle.arguments.individual_args.IndividualArgs = None, sys_args: ~gaggle.arguments.sys_args.SysArgs = None)

Bases: Individual

An Individual whose initial parameters are numpy arrays.

forward(*args, **kwargs)

By default we returned a flattened tensor of the model parameters.

Notes

We do not return the metadata that can be used to reconstruct the gene_pool dictionary as this should not be used to modify the parameters directly (the tensor does not link back to the parameters).

Parameters:

• *args –

• **kwargs –

Returns:

A flattened pytorch tensor of length self.genome_size.

generate_gene_pool(np_params: dict[slice(<class 'int'>, <class 'numpy.ndarray'>, None)], *args, **kwargs) → dict[slice(<class 'int'>, dict[slice(<class 'str'>, <class 'torch.nn.parameter.Parameter'>, None), slice(<class 'str'>, <class 'int'>, None)], None)]

Should return a dictionary of dictionaries where the inner dictionary has a “param” and a “size” key for each of the nn.Parameters in the dictionary

initialize(*args, **kwargs) → Module

Represents the initialization rule for an individual. It should modify the gene pool.

Parameters:

• *args –

• **kwargs –

Returns:

self

Pytorch_individual

class gaggle.population.base_individuals.pytorch_individual.PytorchIndividual(tensors: dict[slice(<class 'int'>, <class 'torch.Tensor'>, None)] = None, individual_args: ~gaggle.arguments.individual_args.IndividualArgs = None, sys_args: ~gaggle.arguments.sys_args.SysArgs = None)

Bases: Individual

An Individual whose initial parameters are pytorch tensors.

forward(*args, **kwargs)

By default we returned a flattened tensor of the model parameters.

Notes

We do not return the metadata that can be used to reconstruct the gene_pool dictionary as this should not be used to modify the parameters directly (the tensor does not link back to the parameters).

Parameters:

• *args –

• **kwargs –

Returns:

A flattened pytorch tensor of length self.genome_size.

generate_gene_pool(tensors: dict[slice(<class 'int'>, <class 'torch.Tensor'>, None)], *args, **kwargs) → dict[slice(<class 'int'>, dict[slice(<class 'str'>, <class 'torch.nn.parameter.Parameter'>, None), slice(<class 'str'>, <class 'int'>, None)], None)]

Should return a dictionary of dictionaries where the inner dictionary has a “param” and a “size” key for each of the nn.Parameters in the dictionary

initialize(*args, **kwargs) → Module

Represents the initialization rule for an individual. It should modify the gene pool.

Parameters:

• *args –

• **kwargs –

Returns:

self

Rl_individual

class gaggle.population.base_individuals.rl_individual.RLIndividual(individual_args: IndividualArgs, sys_args: SysArgs = None, model: Module = None, *args, **kwargs)

Bases: NNIndividual

NNIndividual wrapper that adds an argmax to the prediction as RL problem usually require the output of the model to be an action rather than logits.

forward(x)

By default we returned a flattened tensor of the model parameters.

Notes

We do not return the metadata that can be used to reconstruct the gene_pool dictionary as this should not be used to modify the parameters directly (the tensor does not link back to the parameters).

Parameters:

• *args –

• **kwargs –

Returns:

A flattened pytorch tensor of length self.genome_size.