Quillot Mathias / volia

Download as
Browse Code ยป

Commit 660d9960f95ade5bb2446df6177425097c9b71a9

Authored by quillotm
1 parent 78e6974959
Exists in master

Adding n init parameters

Showing 2 changed files with 53 additions and 28 deletions Side-by-side Diff

... ... @@ -88,6 +88,25 @@
88 88 @param output: output file if kmax not specified, else, output directory
89 89 @param mahalanobis: distance option of k-means.
90 90 """
  91 + json_content = locals().copy()
  92 +
  93 + def fit_model(k: int, output_file):
  94 + if debug:
  95 + print(f"Computing clustering with k={k}")
  96 + model = CLUSTERING_METHODS["k-means"]
  97 + if mahalanobis:
  98 + if debug:
  99 + print("Mahalanobis activated")
  100 + model = CLUSTERING_METHODS["k-means-mahalanobis"]
  101 + model.fit(X, k, tol, ninit, maxiter, debug)
  102 + model.save(output_file)
  103 + json_content["models"].append({
  104 + "model_file": output_file,
  105 + "k": k,
  106 + })
  107 +
  108 + json_content["models"] = []
  109 +
91 110 # -- READ FILES --
92 111 features_dict = read_features(features)
93 112 lst_dict = read_lst(lst)
... ... @@ -102,13 +121,7 @@
102 121  
103 122 # Mono value case
104 123 if kmax is None and klist is None:
105   - if debug:
106   - print(f"Computing clustering with k={k}")
107   - model = CLUSTERING_METHODS["k-means"]
108   - if mahalanobis:
109   - model = CLUSTERING_METHODS["k-means-mahalanobis"]
110   - model.fit(X, k, tol, maxiter, debug)
111   - model.save(output)
  124 + fit_model(k, output)
112 125  
113 126 # Multi values case with kmax
114 127 if kmax is not None:
... ... @@ -116,11 +129,7 @@
116 129 mkdir(output)
117 130 Ks = range(k, kmax + 1)
118 131 for i in Ks:
119   - model = CLUSTERING_METHODS["k-means"]
120   - if mahalanobis:
121   - model = CLUSTERING_METHODS["k-means-mahalanobis"]
122   - model.fit(X, i, tol, maxiter, debug)
123   - model.save(path.join(output, "clustering_" + str(i) + ".pkl"))
  132 + fit_model(i, path.join(output, "clustering_" + str(i) + ".pkl"))
124 133  
125 134 # Second multi values case with klist
126 135 if klist is not None:
127 136  
... ... @@ -128,16 +137,10 @@
128 137 mkdir(output)
129 138 for k in klist:
130 139 k = int(k)
131   - model = CLUSTERING_METHODS["k-means"]
132   - if mahalanobis:
133   - model = CLUSTERING_METHODS["k-means-mahalanobis"]
134   - model.fit(X, k, tol, maxiter, debug)
135   - model.save(path.join(output, "clustering_" + str(k) + ".pkl"))
  140 + fit_model(k, path.join(output, "clustering_" + str(i) + ".pkl"))
136 141  
137   - # TODO: Output json to explain the end parameters like number of iteration, tol reached and stoped the process ?
138   - # etc. (what distance, what parameters etc)
139   - # TODO: Move example data into a directory.
140   - # TODO: Add example receipts
  142 + print(json_content)
  143 + # TODO: compute loss with k-means mahalanobis.
141 144 # TODO: n_init have to be taken into account for mahalanobis case of k-means algorithm.
142 145  
143 146  
volia/clustering_modules/kmeans_mahalanobis.py
Diff comments   View file @ 660d996
... ... @@ -60,8 +60,17 @@
60 60 with open(modelpath, "wb") as f:
61 61 pickle.dump(data, f)
62 62  
63   - def fit(self, features, k: int, tol: float = 0.0001, maxiter: int=300, debug: bool=False):
64   - self._train(features, k, tol, maxiter, debug)
  63 + def fit(self, features, k: int, tol: float, ninit: int, maxiter: int=300, debug: bool=False):
  64 + results = []
  65 + for i in range(ninit):
  66 + results.append(self._train(features, k, tol, maxiter, debug))
  67 + losses = [v["loss"] for v in results]
  68 + best = results[losses.index(min(losses))]
  69 + if debug:
  70 + print(f"best: {best['loss']} loss")
  71 + self.C = best["C"]
  72 + self.L = best["L"]
  73 + self.K = best["K"]
65 74  
66 75 def _initialize_model(self, X, number_clusters):
67 76 d = X.shape[1]
... ... @@ -101,7 +110,6 @@
101 110 N = X.shape[0]
102 111 d = X.shape[1]
103 112  
104   - X_embedded = None
105 113 C, L = self._initialize_model(X, K)
106 114 self.C = C
107 115 self.L = L
108 116  
109 117  
110 118  
... ... @@ -114,13 +122,18 @@
114 122 print("Iteration: ", i)
115 123  
116 124 # Calcul matrix distance
117   - distances = np.zeros((N, K))
  125 + distances = np.zeros((N, self.K))
118 126  
119 127 for n in range(N):
120 128 for k in range(self.K):
121 129 distances[n][k] = self._dist(X[n], self.C[k], self.L[k])
  130 +
122 131 closest_cluster = np.argmin(distances, axis=1)
  132 + loss = np.sum(distances[np.arange(len(distances)), closest_cluster])
  133 + if debug:
  134 + print(f"loss {loss}")
123 135  
  136 +
124 137 # -- Debug tool ----------------------
125 138 if debug and i % 10 == 0:
126 139 # TSNE if needed
127 140  
128 141  
129 142  
... ... @@ -161,17 +174,26 @@
161 174 self.C[k] = C_new
162 175 self.L[k] = K_new
163 176  
  177 +
164 178 diff = np.sum(np.absolute((self.C - old_c) / old_c * 100))
165 179 if diff > tol:
166 180 end_algo = False
167 181 if debug:
168 182 print(f"{diff}")
169   - elif debug:
170   - print(f"Tolerance threshold {tol} reached with diff {diff}")
  183 + else:
  184 + if debug:
  185 + print(f"Tolerance threshold {tol} reached with diff {diff}")
171 186 end_algo = True
  187 +
172 188 i = i + 1
173 189 if i > maxiter:
174 190 end_algo = True
175 191 if debug:
176 192 print(f"Iteration {maxiter} reached")
  193 + return {
  194 + "loss": loss,
  195 + "C": self.C,
  196 + "K": self.K,
  197 + "L": self.L
  198 + }