Commit 3ed53b423bf127a27f1005baec835e1a1af878e5
1 parent
0774ae544e
Exists in
master
just fixed and error
Showing 1 changed file with 1 additions and 1 deletions Inline Diff
volia/clustering_modules/kmeans_multidistance.py
| 1 | 1 | ||
| 2 | import pickle | 2 | import pickle |
| 3 | from abstract_clustering import AbstractClustering | 3 | from abstract_clustering import AbstractClustering |
| 4 | from KMeans_Multidistance.KMeans_Class import KMeans | 4 | from KMeans_Multidistance.KMeans_Class import KMeans |
| 5 | from random import seed | 5 | from random import seed |
| 6 | from random import randint | 6 | from random import randint |
| 7 | import numpy as np | 7 | import numpy as np |
| 8 | from sklearn.metrics import pairwise_distances | 8 | from sklearn.metrics import pairwise_distances |
| 9 | 9 | ||
| 10 | class kmeansMultidistance(): | 10 | class kmeansMultidistance(): |
| 11 | def __init__(self, distance="cosine"): | 11 | def __init__(self, distance="cosine"): |
| 12 | self.kmeans_model = None # Best model | 12 | self.kmeans_model = None # Best model |
| 13 | self.centroids = None | 13 | self.centroids = None |
| 14 | self.distance = distance | 14 | self.distance = distance |
| 15 | self.seed = None # Seed of the best | 15 | self.seed = None # Seed of the best |
| 16 | self.seeds = None | 16 | self.seeds = None |
| 17 | 17 | ||
| 18 | def predict(self, features): | 18 | def predict(self, features): |
| 19 | """ | 19 | """ |
| 20 | 20 | ||
| 21 | @param features: | 21 | @param features: |
| 22 | @return: | 22 | @return: |
| 23 | """ | 23 | """ |
| 24 | return self.kmeans_model.assign_clusters(data=features, centroids=self.centroids, distance=self.kmeans_model.distance) | 24 | return self.kmeans_model.assign_clusters(data=features, centroids=self.centroids, distance=self.kmeans_model.distance) |
| 25 | 25 | ||
| 26 | def load(self, model_path: str): | 26 | def load(self, model_path: str): |
| 27 | """ | 27 | """ |
| 28 | 28 | ||
| 29 | @param model_path: | 29 | @param model_path: |
| 30 | @return: | 30 | @return: |
| 31 | """ | 31 | """ |
| 32 | with open(model_path, "rb") as f: | 32 | with open(model_path, "rb") as f: |
| 33 | data = pickle.load(f) | 33 | data = pickle.load(f) |
| 34 | self.kmeans_model = data["kmeans_model"] | 34 | self.kmeans_model = data["kmeans_model"] |
| 35 | self.centroids = data["centroids"] | 35 | self.centroids = data["centroids"] |
| 36 | self.distance = self.kmeans_model.distance | 36 | self.distance = self.kmeans_model.distance |
| 37 | 37 | ||
| 38 | def save(self, model_path: str): | 38 | def save(self, model_path: str): |
| 39 | """ | 39 | """ |
| 40 | 40 | ||
| 41 | @param model_path: | 41 | @param model_path: |
| 42 | @return: | 42 | @return: |
| 43 | """ | 43 | """ |
| 44 | with open(model_path, "wb") as f: | 44 | with open(model_path, "wb") as f: |
| 45 | pickle.dump({ | 45 | pickle.dump({ |
| 46 | "kmeans_model": self.kmeans_model, | 46 | "kmeans_model": self.kmeans_model, |
| 47 | "centroids": self.centroids | 47 | "centroids": self.centroids |
| 48 | }, f) | 48 | }, f) |
| 49 | 49 | ||
| 50 | def fit(self, features, k: int, tol: float, ninit: int, maxiter: int=300, debug: bool=False): | 50 | def fit(self, features, k: int, tol: float, ninit: int, maxiter: int=300, debug: bool=False): |
| 51 | """ | 51 | """ |
| 52 | 52 | ||
| 53 | @param features: | 53 | @param features: |
| 54 | @param k: | 54 | @param k: |
| 55 | @return: | 55 | @return: |
| 56 | """ | 56 | """ |
| 57 | # Initialization | 57 | # Initialization |
| 58 | self.kmeans_model = None | 58 | self.kmeans_model = None |
| 59 | self.centroids = None | 59 | self.centroids = None |
| 60 | self.seed = None | 60 | self.seed = None |
| 61 | 61 | ||
| 62 | # Compute seeds before using seeds | 62 | # Compute seeds before using seeds |
| 63 | seed() | 63 | seed() |
| 64 | self.seeds = [randint(1, 100000) for i in range(ninit)] | 64 | self.seeds = [randint(1, 100000) for i in range(ninit)] |
| 65 | 65 | ||
| 66 | # Learning k-means model | 66 | # Learning k-means model |
| 67 | results = [] | 67 | results = [] |
| 68 | for i in range(ninit): | 68 | for i in range(ninit): |
| 69 | model = KMeans(k=5, | 69 | model = KMeans(k=k, |
| 70 | maxiter=maxiter, | 70 | maxiter=maxiter, |
| 71 | distance=self.distance, | 71 | distance=self.distance, |
| 72 | record_heterogeneity=[], | 72 | record_heterogeneity=[], |
| 73 | verbose=debug, | 73 | verbose=debug, |
| 74 | seed=self.seeds[i]) | 74 | seed=self.seeds[i]) |
| 75 | centroids, closest_cluster = model.fit(features) | 75 | centroids, closest_cluster = model.fit(features) |
| 76 | 76 | ||
| 77 | # Compute distance | 77 | # Compute distance |
| 78 | kwds = {} | 78 | kwds = {} |
| 79 | if self.distance == "mahalanobis": | 79 | if self.distance == "mahalanobis": |
| 80 | VI = np.linalg.pinv(np.cov(features.T)).T | 80 | VI = np.linalg.pinv(np.cov(features.T)).T |
| 81 | kwds = {"VI": VI} | 81 | kwds = {"VI": VI} |
| 82 | distances = pairwise_distances(features, centroids, metric=self.distance, **kwds) | 82 | distances = pairwise_distances(features, centroids, metric=self.distance, **kwds) |
| 83 | 83 | ||
| 84 | # Then compute the loss | 84 | # Then compute the loss |
| 85 | loss = np.sum(distances[np.arange(len(distances)), closest_cluster]) | 85 | loss = np.sum(distances[np.arange(len(distances)), closest_cluster]) |
| 86 | 86 | ||
| 87 | results.append({ | 87 | results.append({ |
| 88 | "model": model, | 88 | "model": model, |
| 89 | "centroids": centroids, | 89 | "centroids": centroids, |
| 90 | "seed": self.seeds[i], | 90 | "seed": self.seeds[i], |
| 91 | "loss": loss | 91 | "loss": loss |
| 92 | }) | 92 | }) |
| 93 | losses = [result["loss"] for result in results] | 93 | losses = [result["loss"] for result in results] |
| 94 | best = results[losses.index(min(losses))] | 94 | best = results[losses.index(min(losses))] |
| 95 | self.kmeans_model = best["model"] | 95 | self.kmeans_model = best["model"] |
| 96 | self.centroids = best["centroids"] | 96 | self.centroids = best["centroids"] |
| 97 | self.seed = best["seed"] | 97 | self.seed = best["seed"] |
| 98 | 98 |