Reexpress MCP Server

# Copyright Reexpress AI, Inc. All rights reserved. import constants import numpy as np from typing import Optional class UncertaintyStatistics: """ Global statistics across iterations of training and data splitting This collects the min valid similarity values across training iterations, which can be useful for analysis purposes. For example, it is a useful indicator to know if one of those values is inf, which suggests the alpha value may be too high to achieve with the given model and/or data. """ def __init__(self, globalUncertaintyModelUUID: str, numberOfClasses: int, min_rescaled_similarity_across_iterations: Optional[list[float]] = None): self.globalUncertaintyModelUUID = globalUncertaintyModelUUID self.numberOfClasses = numberOfClasses if min_rescaled_similarity_across_iterations is None: self.min_rescaled_similarity_across_iterations = [] else: self.min_rescaled_similarity_across_iterations = min_rescaled_similarity_across_iterations def update_min_rescaled_similarity_to_determine_high_reliability_region( self, min_rescaled_similarity_to_determine_high_reliability_region: float): self.min_rescaled_similarity_across_iterations.append( min_rescaled_similarity_to_determine_high_reliability_region) @staticmethod def get_median_absolute_deviation_around_the_median(list_of_floats: list[float]) -> float: """ Median absolute deviation (around the median) Parameters ---------- list_of_floats Returns ------- """ median_val = np.median(list_of_floats) return np.median(np.abs(np.array(list_of_floats) - median_val)) def _get_min_valid_rescaled_similarity_mad(self) -> float: if len(self.min_rescaled_similarity_across_iterations) > 0: min_q_bin = UncertaintyStatistics.get_median_absolute_deviation_around_the_median( self.min_rescaled_similarity_across_iterations) if np.isfinite(min_q_bin): return min_q_bin return np.inf def validate_min_rescaled_similarities(self): count_non_finite = 0 for rescaled_similarity in self.min_rescaled_similarity_across_iterations: if not np.isfinite(rescaled_similarity): count_non_finite += 1 if count_non_finite > 0: print(f"WARNING: In {count_non_finite} training iterations out of " f"{len(self.min_rescaled_similarity_across_iterations)}, a suitable threshold was not found at the " f"given alpha value. The model and/or data may be too weak to reliably determine the High " f"Reliability region.") else: print(f"Thresholds were found at the given alpha value for all " f"{len(self.min_rescaled_similarity_across_iterations)} training iterations.") print(f"Across iterations, the median absolute deviation around the median for the " f"rescaled similarity (q') to determine the high reliability region is: " f"{self._get_min_valid_rescaled_similarity_mad()}") def export_properties_to_dict(self): json_dict = {constants.STORAGE_KEY_version: constants.ProgramIdentifiers_version, constants.STORAGE_KEY_globalUncertaintyModelUUID: self.globalUncertaintyModelUUID, constants.STORAGE_KEY_numberOfClasses: self.numberOfClasses, constants.STORAGE_KEY_min_rescaled_similarity_across_iterations: self.min_rescaled_similarity_across_iterations, } return json_dict