dependent samples

bob/measure/dependent_samples_generation.py

@@ -19,28 +19,37 @@ def generate_groundtruth(first_sample_pb, keep_state_pb, nb_samples):
 def generate_ml(accuracy, groundtruth, nb_systems):
     ml = np.zeros([nb_systems, groundtruth.size])
     for system in range(nb_systems):
-        print(system)
+#         if (system % (nb_systems/10) == 0):
+            # print progress each 10 percent
+#             print((system / nb_systems) * 100, 'percent')
         for i in range(groundtruth.size):
             ml[system][i] = np.random.choice([groundtruth[i] ,1 - groundtruth[i]], p=[accuracy, 1 - accuracy])
     return ml
 
-def compute_accuracy(ml, gt):
-    return np.mean(ml == gt, axis=1)
-
-
-
-gt = generate_groundtruth(1, 1, 10000)
-ml = generate_ml(0.5, gt, 10000)
-acc = compute_accuracy(ml, gt)
-
-
-fig, ax = plt.subplots(nrows=2, ncols=2, tight_layout=True)
-n_bins = 10
-ax[0,0].hist(acc, bins=n_bins)
-n_bins = 20
-ax[0,1].hist(acc, bins=n_bins)
-n_bins = 50
-ax[1,0].hist(acc, bins=n_bins)
-n_bins = 100
-ax[1,1].hist(acc, bins=n_bins)
-plt.show()
+def compute_accuracy_k(ml, gt):
+    return (ml == gt).sum(axis = 1)
+
+
+for dependency_prob in np.linspace(0.1, 1, 9, endpoint=False):
+#     print("===== Dependency probability is ", "{:.1f}".format(dependency_prob), "=====")
+    gt = generate_groundtruth(0.5, dependency_prob, 1000)
+    for accuracy in np.linspace(0.1, 1, 9, endpoint=False):
+#         print("== Accuracy is ", "{:.1f}".format(accuracy), "==")
+        ml = generate_ml(accuracy, gt, 10000)
+        acc = compute_accuracy_k(ml, gt)
+        fig, ax = plt.subplots(nrows=2, ncols=2, tight_layout=True)
+        fig.suptitle('Dependency probability : ' + "{:.1f}".format(dependency_prob) + 'Accuracy : ' + "{:.1f}".format(accuracy), fontsize=16)
+        n_bins = 10
+        ax[0,0].hist(acc, bins=n_bins)
+        ax[0,0].set_title(str(n_bins) + ' bins')
+        n_bins = 20
+        ax[0,1].hist(acc, bins=n_bins)
+        ax[0,1].set_title(str(n_bins) + ' bins')
+        n_bins = 50
\ No newline at end of file
+        ax[1,0].hist(acc, bins=n_bins)
+        ax[1,0].set_title(str(n_bins) + ' bins')
+        n_bins = 100
+        ax[1,1].hist(acc, bins=n_bins)
+        ax[1,1].set_title(str(n_bins) + ' bins')
+        plt.savefig('/idiap/home/amorais/Desktop/Dependent_samples_figures/' + 'Depprob' + "{:.1f}".format(dependency_prob) + 'acc' + "{:.1f}".format(accuracy) + '.png')
+        plt.close()