dSalmon.scalers

Scalers for streaming data.

Classes

SWQuantileScaler(window, quantile[, float_type])

Performs normalization so that the p-quantile of the current sliding window is mapped to 0 and the (1-p)-quantile is mapped to 1.

SWScaler([float_type])

Base class for sliding window scalers.

SWZScoreScaler(window[, float_type])

Performs z-score normalization of samples based on mean and standard deviation observed in a sliding window of length window.

class dSalmon.scalers.SWQuantileScaler(window, quantile, float_type=<class 'numpy.float64'>)[source]

Performs normalization so that the p-quantile of the current sliding window is mapped to 0 and the (1-p)-quantile is mapped to 1. If quantile==0, performs minmax normalization. Note that due to its lacking robustness, minmax normalization is likely to result in unstable results for stream data.

Parameters
  • window (float) – Window length after which samples will be pruned.

  • quantile (float with 0 <= quantile < 0.5) – The quantile value for computing reference values.

  • float_type (np.float32 or np.float64) – The floating point type to use for internal processing.

transform(X, times=None)

Transform the next chunk of data.

Parameters
  • X (ndarray, shape (n_samples, n_features)) – The input data.

  • times (ndarray, shape (n_samples,), optional) – Timestamps for input data. If None, timestamps are linearly increased for each sample.

Returns

X_tr – Transformed input data.

Return type

ndarray, shape (n_samples, n_features)

transform_inplace(X, times=None)

Transform the next chunk of data in-place. Requires X to be a C-style contiguous ndarray.

Parameters
  • X (ndarray, shape (n_samples, n_features)) – The input data.

  • times (ndarray, shape (n_samples,), optional) – Timestamps for input data. If None, timestamps are linearly increased for each sample.

Returns

X_tr – Transformed input data. Equal to X.

Return type

ndarray, shape (n_samples, n_features)

class dSalmon.scalers.SWScaler(float_type=<class 'numpy.float64'>)[source]

Base class for sliding window scalers.

transform(X, times=None)[source]

Transform the next chunk of data.

Parameters
  • X (ndarray, shape (n_samples, n_features)) – The input data.

  • times (ndarray, shape (n_samples,), optional) – Timestamps for input data. If None, timestamps are linearly increased for each sample.

Returns

X_tr – Transformed input data.

Return type

ndarray, shape (n_samples, n_features)

transform_inplace(X, times=None)[source]

Transform the next chunk of data in-place. Requires X to be a C-style contiguous ndarray.

Parameters
  • X (ndarray, shape (n_samples, n_features)) – The input data.

  • times (ndarray, shape (n_samples,), optional) – Timestamps for input data. If None, timestamps are linearly increased for each sample.

Returns

X_tr – Transformed input data. Equal to X.

Return type

ndarray, shape (n_samples, n_features)

class dSalmon.scalers.SWZScoreScaler(window, float_type=<class 'numpy.float64'>)[source]

Performs z-score normalization of samples based on mean and standard deviation observed in a sliding window of length window.

Parameters
  • window (float) – Window length after which samples will be pruned.

  • float_type (np.float32 or np.float64) – The floating point type to use for internal processing.

transform(X, times=None)

Transform the next chunk of data.

Parameters
  • X (ndarray, shape (n_samples, n_features)) – The input data.

  • times (ndarray, shape (n_samples,), optional) – Timestamps for input data. If None, timestamps are linearly increased for each sample.

Returns

X_tr – Transformed input data.

Return type

ndarray, shape (n_samples, n_features)

transform_inplace(X, times=None)

Transform the next chunk of data in-place. Requires X to be a C-style contiguous ndarray.

Parameters
  • X (ndarray, shape (n_samples, n_features)) – The input data.

  • times (ndarray, shape (n_samples,), optional) – Timestamps for input data. If None, timestamps are linearly increased for each sample.

Returns

X_tr – Transformed input data. Equal to X.

Return type

ndarray, shape (n_samples, n_features)