ezmsg.learn.process.sgd

Classes

class SGDDecoder(*args, settings=None, **kwargs)

Bases: BaseAdaptiveTransformerUnit[SGDDecoderSettings, AxisArray, ClassifierMessage, SGDDecoderTransformer]

Parameters:

settings (Settings | None)

SETTINGS

alias of SGDDecoderSettings

class SGDDecoderSettings(alpha=1e-05, eta0=0.0003, loss='hinge', label_weights=None, settings_path=None)

Bases: Settings

Parameters:

• alpha (float)

• eta0 (float)

• loss (str)

• label_weights (dict[str, float] | None)

• settings_path (str | None)

alpha: float = 1e-05

eta0: float = 0.0003

loss: str = 'hinge'

label_weights: dict[str, float] | None = None

settings_path: str | None = None

__init__(alpha=1e-05, eta0=0.0003, loss='hinge', label_weights=None, settings_path=None)

Parameters:

• alpha (float)

• eta0 (float)

• loss (str)

• label_weights (dict[str, float] | None)

• settings_path (str | None)

Return type:

None

class SGDDecoderState

Bases: object

model: Any = None

b_first_train: bool = True

class SGDDecoderTransformer(*args, **kwargs)

Bases: BaseAdaptiveTransformer[SGDDecoderSettings, AxisArray, ClassifierMessage, SGDDecoderState]

STREAMING_DIMS: ClassVar[tuple[str, ...]] = ('win', 'time')

This decoder is fed windows, so a producer that declares no stream_dim is accumulating along win rather than time. The base default would guess time and flatten the windows into the feature vector.

partial_fit(message)

Return type:

None

Parameters:

message (AxisArray)