project-codenet / model-experiments /token-based-similarity-classification /src /Clustering /NearestCluster.py
| """ | |
| """ | |
| import sys | |
| import os | |
| import math | |
| import numpy as np | |
| from VectorKMeans import VectorsClustered | |
| class NearestClusterClassifier(object): | |
| """ | |
| """ | |
| def __init__(self, train_samples, val_samples): | |
| """ | |
| """ | |
| self._val_samples = val_samples | |
| self._min_n_class_samples = min(map(len, train_samples)) | |
| print("_min_n_class_samples: ", self._min_n_class_samples) | |
| self._classes = [VectorsClustered(_samples) | |
| for _samples in train_samples] | |
| self.n_classes = len(self._classes) | |
| def nClustersTrain(self, n_clusters): | |
| """ | |
| """ | |
| for _class in self._classes: | |
| _class.kmeansCluster(n_clusters) | |
| def scaledClustersTrain(self, n_clusters): | |
| """ | |
| """ | |
| for _class in self._classes: | |
| _scale = _class.n_samples // self._min_n_class_samples | |
| print("Scale: ", _scale) | |
| _class.kmeansCluster(n_clusters * _scale) | |
| def classify(self, sample): | |
| """ | |
| """ | |
| _min_d = sys.maxsize | |
| _classification = None | |
| for _i, _class in enumerate(self._classes): | |
| _class_dist = _class.distanceToCenters(sample) | |
| if _class_dist < _min_d: | |
| _min_d = _class_dist | |
| _classification = _i | |
| return _classification | |
| def validate(self): | |
| """ | |
| """ | |
| return self.test(self._val_samples) | |
| def test(self, val_samples): | |
| """ | |
| """ | |
| _n_correct_list = [] | |
| _acc_list = [] | |
| _n_val_samples = sum(map(len, val_samples)) | |
| print(_n_val_samples) | |
| for _i, _val_class_samples in enumerate(val_samples): | |
| _n_correct = 0 | |
| for _sample in _val_class_samples: | |
| if _i == self.classify(_sample): | |
| _n_correct += 1 | |
| _n_correct_list.append(_n_correct) | |
| _acc_list.append(float(_n_correct) / | |
| float(len(_val_class_samples))) | |
| _n_correct = sum(_n_correct_list) | |
| print("Accuracy of classes: ", _acc_list) | |
| return float(_n_correct) / float(_n_val_samples) | |