unsupervised labels no works with labels directly but assumes K=num_c…

…lasses.

utils.py

@@ -1,7 +1,6 @@
 import os
 import pickle
 import numpy as np
-from scipy import stats
 from scipy.optimize import linear_sum_assignment
 
 
@@ -39,56 +38,30 @@ def save_performance(perf, epoch, save_path):
     assert str(perf) == str(perf_load), 'performance saving failed'
 
 
-def unsupervised_labels(alpha, y, mdl, loss_type):
+def unsupervised_labels(y, y_hat, num_classes, num_clusters):
     """
-    :param alpha: concentration parameter
     :param y: true label
-    :param mdl: the model object
-    :param loss_type: name used for printing updates
+    :param y_hat: concentration parameter
+    :param num_classes: number of classes (determined by data)
+    :param num_clusters: number of clusters (determined by model)
     :return: classification error rate
     """
-    # same number of classes as labels?
-    if mdl.K == mdl.num_classes:
+    assert num_classes == num_clusters
 
-        # construct y-hat
-        y_hat = np.argmax(alpha, axis=1)
+    # initialize count matrix
+    cnt_mtx = np.zeros([num_classes, num_classes])
 
-        # initialize count matrix
-        cnt_mtx = np.zeros([mdl.K, mdl.K])
+    # fill in matrix
+    for i in range(len(y)):
+        cnt_mtx[int(y_hat[i]), int(y[i])] += 1
 
-        # fill in matrix
-        for i in range(len(y)):
-            cnt_mtx[int(y_hat[i]), int(y[i])] += 1
+    # find optimal permutation
+    row_ind, col_ind = linear_sum_assignment(-cnt_mtx)
 
-        # find optimal permutation
-        row_ind, col_ind = linear_sum_assignment(-cnt_mtx)
-
-        # compute error
-        error = 1 - cnt_mtx[row_ind, col_ind].sum() / cnt_mtx.sum()
-
-    # different number of classes than labels
-    else:
-
-        # initialize y-hat
-        y_hat = -np.ones(y.shape)
-
-        # loop over the number of latent clusters
-        for i in range(mdl.K):
-
-            # find the real label corresponding to the largest concentration for this cluster
-            i_sort = np.argsort(alpha[:, i])[-100:]
-            y_real = stats.mode(y[i_sort])[0]
-
-            # assign that label to all points where its concentration is maximal
-            y_hat[np.argmax(alpha, axis=1) == i] = y_real
-
-        # make sure we handled everyone
-        assert np.sum(y_hat < 0) == 0
-
-        # compute the error
-        error = np.mean(y != y_hat)
+    # compute error
+    error = 1 - cnt_mtx[row_ind, col_ind].sum() / cnt_mtx.sum()
 
     # print results
-    print('Classification error for ' + loss_type + ' data = {:.4f}'.format(error))
+    print('Classification error = {:.4f}'.format(error))
 
     return error