Proper plotting of the ROC curves for the pair matching problem

1 files changed, 29 insertions, 27 deletions

diff --git a/data/pair-matching/roc.py b/data/pair-matching/roc.py
index ebf8f68..46e4d23 100755
--- a/data/pair-matching/roc.py
+++ b/data/pair-matching/roc.py
@@ -30,43 +30,45 @@ def gen_pairs(var,sk_data):
     m_pairs = zip(range(sk_data.shape[0]),range(sk_data.shape[0]))
     result = []
     for j in range(sk_data.shape[0]):
-            result += [(distance(sk1[m_pairs[j][0]],sk2[m_pairs[j][1]]), distance(sk1[u_pairs[j][0]],sk2[u_pairs[j][1]]))]
+            result += [(distance(sk1[m_pairs[j][0]],sk2[m_pairs[j][1]]),
+                        distance(sk1[u_pairs[j][0]],sk2[u_pairs[j][1]]))]
 
     return result
 
 if __name__ == "__main__":
-#    eg = np.loadtxt("eigenfaces.txt",delimiter=" ")
     ap = np.loadtxt("associatepredict.txt",delimiter=",")
-#    plt.plot(eg[:,0],eg[:,1])
-    plt.plot(ap[:,1],ap[:,0])
-    plt.xlabel("False positive %")
-    plt.ylabel("True positive %")
+    indices = [i for i in range(ap.shape[0]) if ap[i,1]<0.1]
+    ap_false = ap[:,1][indices]
+    ap_true = ap[:,0][indices]
+    plt.plot(ap_false,ap_true,label="Face detection")
+    plt.xlabel("False positive \%")
+    plt.ylabel("True positive \%")
     np.random.seed()
-    var = map(float,sys.argv[2].split(","))
+    std = map(float,sys.argv[2].split(","))
     sk_data  = np.loadtxt(sys.argv[1],comments="#",delimiter=",")
-    for v in var:
-        result = gen_pairs(v,sk_data)
-        thresholds = np.square(np.arange(0,10,0.01))
+    for s in std:
+        result = gen_pairs(s,sk_data)
+        thresholds = np.square(np.arange(0,10,0.001))
         true_pos = []
         false_pos = []
         for threshold in thresholds:
-            true_values = []
-            false_values = []
-            for i in range(4):
-                true = 0
-                false = 0
-                min_j = i*300
-                max_j = min(min_j+300,sk_data.shape[0])
-                for j in range(min_j,max_j):
-                    if result[j][0] < threshold:
-                        true += 1
-                    if result[j][1] < threshold:
-                        false += 1
-                true_values += [float(true)/(max_j-min_j)]
-                false_values += [float(false)/(max_j-min_j)]
-            true_pos += [sum(true_values)/4]
-            false_pos += [sum(false_values)/4]
-        plt.plot(false_pos,true_pos)
+            true = 0
+            false = 0
+            for j in range(len(result)):
+                if result[j][0] < threshold:
+                    true += 1
+                if result[j][1] < threshold:
+                    false += 1
+            true_pos += [float(true)/len(result)]
+            false_pos += [float(false)/len(result)]
+        indices = [i for i in range(len(false_pos)) if false_pos[i]<0.1]
+        false_pos = np.array(false_pos)
+        false_pos = false_pos[indices]
+        true_pos = np.array(true_pos)
+        true_pos = true_pos[indices]
+        plt.plot(false_pos,true_pos,label="$\sigma$ = "+str(s))
+    plt.legend(loc="lower right")
+    plt.savefig("roc.pdf")
     plt.show()