| 1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889 |
- import numpy as np
- class LinearNormalizer(object):
- def __init__(self, vmin, vmax):
- super().__init__()
- self.vmin = vmin
- self.vmax = vmax
- def scale_with_revised_limits_to01(self, data_clipped, vmin2use, vmax2use):
- data_scaled = (data_clipped - vmin2use) / (vmax2use - vmin2use)
- return data_scaled
- def normalize(self, data):
- data_clipped = np.clip(data, self.vmin, self.vmax)
- return self.scale_with_revised_limits_to01(data_clipped, self.vmin, self.vmax)
- def get_data_limits(self):
- """
- Return the lower and upper limits of data as a tuple
- :return: tuple
- """
- return self.vmin, self.vmax
- class LinearNormalizerAcross0(LinearNormalizer):
- def __init__(self, vmin, vmax):
- assert (vmin < 0) & (vmax > 0), "LinearScaleAcross0 only works when vmin < 0 and vmax > 0"
- super().__init__(vmin, vmax)
- class BilinearNormalizerCentering0(LinearNormalizerAcross0):
- def __init__(self, vmin, vmax):
- super().__init__(vmin=vmin, vmax=vmax)
- def scale_with_revised_limits_to01(self, data_clipped, vmin2use, vmax2use):
- data_negative_masked_out = np.ma.MaskedArray(data=data_clipped, mask=data_clipped < 0, fill_value=0)
- data_positive_masked_out = np.ma.MaskedArray(data=data_clipped, mask=data_clipped >= 0, fill_value=0)
- data_negative_masked_out_scaled \
- = 0.5 + 0.5 * super().scale_with_revised_limits_to01(data_negative_masked_out, 0, vmax2use)
- data_positive_masked_out_scaled \
- = 0.5 * super().scale_with_revised_limits_to01(data_positive_masked_out, vmin2use, 0)
- return data_negative_masked_out_scaled.filled() + data_positive_masked_out_scaled.filled()
- class LinearNormalizerCentering0Symmetric(LinearNormalizerAcross0):
- def __init__(self, vmin=None, vmax=None):
- super().__init__(vmin=vmin, vmax=vmax)
- max_one_sided = max(np.abs(self.vmin), np.abs(self.vmax))
- self.vmin, self.vmax = -max_one_sided, max_one_sided
- def get_normalizer(mv_individualScale, vmin, vmax):
- scale_modifier = int(mv_individualScale / 10)
- if scale_modifier == 0 or not (vmin < 0 < vmax):
- scaler_class = LinearNormalizer
- elif scale_modifier == 1:
- scaler_class = BilinearNormalizerCentering0
- elif scale_modifier == 2:
- scaler_class = LinearNormalizerCentering0Symmetric
- else:
- raise NotImplementedError
- return scaler_class(vmin=vmin, vmax=vmax)
|
