File size: 2,953 Bytes
fc0f7bd | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 | # Copyright (c) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License.
import logging
from ._constants import OUTPUT_SEPARATOR
logger = logging.getLogger(__name__)
class InterpolatedPredictor:
"""Predictor for computing predictions between two actual predictions.
The predictions are represented through the threshold rules operation0 and operation1.
:param p_ignore: p_ignore changes the interpolated prediction P to the desired
solution using the transformation p_ignore * prediction_constant + (1 - p_ignore) * P
:param prediction_constant: 0 if not required, otherwise the x value of the best
solution should be passed
:param p0: interpolation multiplier for prediction from the first predictor
:param operation0: threshold rule for the first predictor
:param p1: interpolation multiplier for prediction from the second predictor
:param operation1: threshold rule for the second predictor
:return: an anonymous function that scales the original prediction to the desired one
:rtype: lambda
"""
def __init__(self, p_ignore, prediction_constant, p0, operation0, p1, operation1):
self._operation0 = operation0
self._operation1 = operation1
self._p_ignore = p_ignore
self._prediction_constant = prediction_constant
self._p0 = p0
self._p1 = p1
logger.debug(OUTPUT_SEPARATOR)
logger.debug("p_ignore: %s", p_ignore)
logger.debug("prediction_constant: %s", prediction_constant)
logger.debug("p0: %s", p0)
logger.debug("operation0: %s", operation0)
logger.debug("p1: %s", p1)
logger.debug("operation1: %s", operation1)
logger.debug(OUTPUT_SEPARATOR)
def __repr__(self): # noqa: D105
return "[p_ignore: {}, prediction_constant: {}, " \
"p0: {}, operation0: {}, p1: {}, operation1: {}]" \
.format(self._p_ignore, self._prediction_constant, self._p0, self._operation0,
self._p1, self._operation1)
def predict(self, scores):
"""Create the interpolated prediction.
The interpolation is based on two threshold operations and the
transformation adjustment.
:param scores: the scores from an unconstrained predictor to which the threshold
operations are applied
:type scores: numpy.ndarray
:return: the interpolated prediction
:rtype: numpy.ndarray
"""
transformation_adjustment = self._p_ignore * self._prediction_constant
weighted_predictions0 = self._p0 * self._operation0.get_predictor_from_operation()(scores)
weighted_predictions1 = self._p1 * self._operation1.get_predictor_from_operation()(scores)
interpolated_predictions = (1 - self._p_ignore) * (weighted_predictions0 + weighted_predictions1) # noqa: E501
return transformation_adjustment + interpolated_predictions
|