def filter(self, run_counts, criteria): wrong_confidence = criteria['wrong_confidence'] below_t = wrong_confidence <= self.t filtered_counts = deep_copy(run_counts) for key in filtered_counts: filtered_counts[key] = filtered_counts[key][below_t] return filtered_counts
def filter(self, run_counts, criteria): correctness = criteria['correctness'] assert correctness.dtype == np.bool filtered_counts = deep_copy(run_counts) for key in filtered_counts: filtered_counts[key] = filtered_counts[key][correctness] return filtered_counts
def start(self, run_counts): _logger.info("Started working on a MaxConfidence goal") _logger.info("Threshold: " + str(self.t)) if self.new_work_goal is None: if self.t >= 1.: _logger.info("This goal will run forever") else: _logger.info("This goal will run until all examples have confidence" + " greater than " + str(self.t) + ", which may never" + " happen.") self.work_before = deep_copy(run_counts)
def start(self, run_counts): for key in run_counts: value = run_counts[key] assert value.ndim == 1 _logger.info("Started working on a Misclassify goal") self.work_before = deep_copy(run_counts)