enchanter.utils

enchanter.utils.comet

class enchanter.utils.comet.TunerConfigGenerator(algorithm: str = 'bayes', metric: str = 'validate_avg_loss', objective: str = 'minimize', seed: Optional[int] = None, max_combo: int = 0, grid_size: int = 10, min_sample_size: int = 100, retry_limit: int = 20, retry_assign_limit: int = 0, name: Optional[str] = None, trials: int = 1)

Bases: object

See https://www.comet.ml/docs/python-sdk/introduction-optimizer/ for a more detailed explanation of each argument.

Parameters

• algorithm – Specifies the algorithm used for parameter tuning. The supported algorithms are ['grid','random','bayes'].

• metric – Specify the value to be minimized / maximized. By default,``validate_avg_loss`` is specified.

• objective – Specifies whether to maximize / minimize metrics. Specify with ['minimize','maximize'].

• seed – Set the seed value. Not specified by default.

• max_combo – Integer. Limit on the combination of parameters to try (default is 0, meaning no limit)

• grid_size – Integer. Number of bins per parameter when creating a grid (default is 10).

• min_sample_size – Integer. Number of samples to help find a suitable grid range (default is 100).

• retry_limit – integer. A limit that attempts to create a unique set of parameters before suspending (default is 20).

• retry_assign_limit –

• name – String. A personalizable name associated with this search instance. (option)

• trials – Specifies the number of trials in a single experiment.

Methods

export(filename)	It is a method to save the created Config as a file.
generate()	A method for generating a Config for comet_ml.Optimizer.
suggest_categorical(name, values)	A method for searching categorical variables.
suggest_linear(name, min_value, max_value[, …])	If Integer, independent distribution.
suggest_uniform(name, min_value, max_value)	This is a method for sampling and searching variables from a uniform distribution.
suggest_normal(name, min_value, max_value[, …])	This is a method for sampling and searching variables from a normal distribution.
suggest_lognormal(name, min_value, max_value)	A method for sampling and searching variables from a lognormal distribution.
suggest_loguniform(name, min_value, max_value)	This is a method for sampling and searching variables from a uniform distribution.
suggest_discrete(name, values)	This method is used to search for the specified numeric type variable.

export(filename: str) → None

It is a method to save the created Config as a file.

generate() → Dict[str, Any]

A method for generating a Config for comet_ml.Optimizer.

Examples

>>> cfg = TunerConfigGenerator()
>>> cfg.suggest_discrete("discrete", [10, 20, 30])
>>> cfg_dict = cfg.generate()

suggest_categorical(name: str, values: List[str])

A method for searching categorical variables.

Parameters

• name – Specify the name of the variable.

• values – I have a variable to search. Due to the specifications of comet.ml, it is necessary to give a list of strings.

Examples

>>> import comet_ml
>>> from enchanter.utils.comet import TunerConfigGenerator
>>> config = TunerConfigGenerator()
>>> config.suggest_categorical("activation", ["torch.relu", "torch.sigmoid", "torch.softmax"])
>>> opt = comet_ml.Optimizer(config.generate())
>>> for experiment in opt.get_experiments():
>>>     activation = experiment.get_parameter(eval("activation"))

suggest_discrete(name: str, values: List[Union[int, float]])

This method is used to search for the specified numeric type variable.

Parameters

• name – Variable name

• values – A list of numeric elements

Examples

>>> import comet_ml
>>> config = TunerConfigGenerator()
>>> config.suggest_discrete("discrete", [10, 20, 30])
>>> opt = comet_ml.Optimizer(config.generate())
>>> for experiment in opt.get_experiments():
>>>     discrete = experiment.get_parameter("discrete")

suggest_linear(name: str, min_value: Union[float, int], max_value: Union[float, int], dtype: Optional[type] = None)

If Integer, independent distribution. else if float the same as uniform.

Parameters

• name –

• min_value –

• max_value –

• dtype –

Returns:

suggest_lognormal(name: str, min_value: Union[float, int], max_value: Union[float, int], mu: float = 0.0, sigma: float = 1.0, dtype: Optional[type] = None)

A method for sampling and searching variables from a lognormal distribution.

Parameters

• name – Variable name

• min_value – minimum value

• max_value – Maximum value

• mu – Normal distribution μ

• sigma – Normal distribution σ

• dtype – Data type. If not specified, it is automatically estimated from the values of min_value and` max_value`.

Examples

>>> import comet_ml
>>> config = TunerConfigGenerator()
>>> config.suggest_lognormal("lognormal", 0.0, 1.0)
>>> opt = comet_ml.Optimizer(config.generate())
>>> for experiment in opt.get_experiments():
>>>     lognormal = experiment.get_parameter("lognormal")

suggest_loguniform(name: str, min_value: Union[float, int], max_value: Union[float, int], dtype: Optional[type] = None)

This is a method for sampling and searching variables from a uniform distribution.

Parameters

• name – Variable name

• min_value – minimum value

• max_value – Maximum value

• dtype – Data type. If not specified, it is automatically estimated from the values of min_value and` max_value`.

Examples

>>> import comet_ml
>>> config = TunerConfigGenerator()
>>> config.suggest_loguniform("loguniform", 0.0, 1.0)
>>> opt = comet_ml.Optimizer(config.generate())
>>> for experiment in opt.get_experiments():
>>>     loguniform = experiment.get_parameter("loguniform")

suggest_normal(name: str, min_value: Union[float, int], max_value: Union[float, int], mu: float = 0.0, sigma: float = 1.0, dtype: Optional[type] = None)

This is a method for sampling and searching variables from a normal distribution.

Parameters

• name – Variable name

• min_value – minimum value

• max_value – Maximum value

• mu – Normal distribution μ

• sigma – Normal distribution σ

• dtype – Data type. If not specified, it is automatically estimated from the values of min_value and` max_value`.

Examples

>>> import comet_ml
>>> config = TunerConfigGenerator()
>>> config.suggest_normal("normal", 0.0, 1.0)
>>> opt = comet_ml.Optimizer(config.generate())
>>> for experiment in opt.get_experiments():
>>>     normal = experiment.get_parameter("normal")

suggest_uniform(name: str, min_value: Union[float, int], max_value: Union[float, int], dtype: Optional[type] = None)

This is a method for sampling and searching variables from a uniform distribution.

Parameters

• name – Variable name

• min_value – minimum value

• max_value – Maximum value

• dtype – Data type. If not specified, it is automatically estimated from the values of min_value and` max_value`.

Examples

>>> import comet_ml
>>> config = TunerConfigGenerator()
>>> config.suggest_uniform("uniform", 0.0, 1.0)
>>> opt = comet_ml.Optimizer(config.generate())
>>> for experiment in opt.get_experiments():
>>>     uniform = experiment.get_parameter("uniform")

enchanter.utils.backend

enchanter.utils.backend.is_scalar(data: Union[numbers.Number, numpy.ndarray, torch.Tensor]) → bool

Returns True if the type of data is a scalar type.

Parameters

data (Union[Number, Union[np.ndarray, torch.Tensor]]) – Numerical value

Returns

True if data is a scalar type, False if it is not.

Examples

>>> a = torch.tensor([1.0])
>>> is_scalar(a)    # True
>>> a = torch.tensor(1.0)
>>> is_scalar(a)    # True
>>> a = torch.tensor([1, 2, 3])
>>> is_scalar(a)    # False
>>> a = 1.0
>>> is_scalar(a)    # True

enchanter.utils.backend.slice_axis(data: torch.Tensor, axis: int, begin: int, end: int) → torch.Tensor

Examples

>>> import torch
>>> x = torch.tensor([
>>>     [  1.,   2.,   3.,   4.],
>>>     [  5.,   6.,   7.,   8.],
>>>     [  9.,  10.,  11.,  12.]
>>> ])
>>>
>>> slice_axis(x, axis=0, begin=1, end=3)
>>> # [[  5.,   6.,   7.,   8.],
>>> # [  9.,  10.,  11.,  12.]]
>>>
>>> slice_axis(x, axis=1, begin=0, end=2)
>>> # [[  1.,   2.],
>>> # [  5.,   6.],
>>> # [  9.,  10.]]
>>>
>>> slice_axis(x, axis=1, begin=-3, end=-1)
>>> # [[  2.,   3.],
>>> # [  6.,   7.],
>>> # [ 10.,  11.]]

References

• Deep Graph Library

• mxnet.ndarray.slice_axis

Parameters

• data – Source input

• axis – Axis along which to be sliced

• begin – The beginning index along the axis to be sliced

• end – The ending index along the axis to be sliced

Returns

output - the output of this function.

enchanter.utils.visualize

enchanter.utils.visualize.with_netron(model: torch.nn.modules.module.Module, dummy: Tuple[torch.Tensor, …], backend: str = 'onnx', open_browser: bool = True, port: int = 8080, host: str = 'localhost')

Visualize the PyTorch graph in a browser.

Examples

>>> from enchanter.addons.layers import AutoEncoder
>>> x = torch.randn(1, 32)  # [N, in_features]
>>> model = AutoEncoder([32, 16, 8, 2])
>>> with_netron(model, (x, ))

Warning

Models that cannot be graphed using TorchScript or ONNX cannot be visualized.

Parameters

• model – PyTorch model

• dummy – dummy input for generate graph

• backend – specified graph format. [torchscript or onnx]. default onnx.

• open_browser – if True, open browser.

• port – port number. default: 8080.

• host – hostname. default: localhost.