gluonts.mx.representation.global_relative_binning module

class gluonts.mx.representation.global_relative_binning.GlobalRelativeBinning(num_bins: int = 1024, is_quantile: bool = True, linear_scaling_limit: int = 10, quantile_scaling_limit: float = 0.99, *args, **kwargs)[source]

Bases: gluonts.mx.representation.representation.Representation

A class representing a global relative binning approach. This binning first rescales all input series by their respective mean (relative) and then performs one binning across all series (global).

Parameters
  • num_bins – The number of discrete bins/buckets that we want values to be mapped to. (default: 1024)

  • is_quantile – Whether the binning is quantile or linear. Quantile binning allocated bins based on the cumulative distribution function, while linear binning allocates evenly spaced bins. (default: True, i.e. quantile binning)

  • linear_scaling_limit – The linear scaling limit. Values which are larger than linear_scaling_limit times the mean will be capped at linear_scaling_limit. (default: 10)

  • quantile_scaling_limit – The quantile scaling limit. Values which are larger than the quantile evaluated at quantile_scaling_limit will be capped at the quantile evaluated at quantile_scaling_limit. (default: 0.99)

hybrid_forward(F, data: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], observed_indicator: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], scale: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol, None], rep_params: List[Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol]], **kwargs) → Tuple[Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], List[Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol]]][source]

Transform the data into the desired representation.

Parameters
  • F

  • data – Target data.

  • observed_indicator – Target observed indicator.

  • scale – Pre-computed scale.

  • rep_params – Additional pre-computed representation parameters.

  • **kwargs, – Additional block-specfic parameters.

Returns

Tuple consisting of the transformed data, the computed scale, and additional parameters to be passed to post_transform.

Return type

Tuple[Tensor, Tensor, List[Tensor]]

initialize_from_array(input_array: numpy.ndarray, ctx: mxnet.context.Context = gpu(0))[source]

Initialize the representation based on a numpy array.

Parameters
  • input_array – Numpy array.

  • ctx – MXNet context.

initialize_from_dataset(input_dataset: Iterable[Dict[str, Any]], ctx: mxnet.context.Context = gpu(0))[source]

Initialize the representation based on an entire dataset.

Parameters
  • input_dataset – GluonTS dataset.

  • ctx – MXNet context.

post_transform(F, samples: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], scale: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], rep_params: List[Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol]]) → Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol][source]

Transform samples back to the original representation.

Parameters
  • samples – Samples from a distribution.

  • scale – The scale of the samples.

  • rep_params – Additional representation-specific parameters used during post transformation.

Returns

Post-transformed samples.

Return type

Tensor