Adding binary sensitivity and specificity loss functions.

CusFns.py

@@ -0,0 +1,42 @@
+from keras import backend as K
+
+def binary_sensitivity(y_true, y_pred):
+    import tensorflow as tf
+    y_true = tf.minimum(tf.maximum(y_true - 0.499, 0) * 10000, 1)
+    y_pred = tf.minimum(tf.maximum(y_pred - 0.499, 0) * 10000, 1)
+
+    cp = K.sum(y_true)
+    tp_tensor = y_true * y_pred
+    tp = K.sum(tp_tensor)
+    return tp/cp
+
+def binary_sensitivity_loss(y_true, y_pred):
+    import tensorflow as tf
+    y_true = tf.minimum(tf.maximum(y_true - 0.499, 0) * 10000, 1)
+    y_pred = tf.minimum(tf.maximum(y_pred - 0.499, 0) * 10000, 1)
+    cp = K.sum(y_true)
+    tp_tensor = y_true * y_pred
+    tp = K.sum(tp_tensor)
+    return 1 - tp / cp
+
+def binary_specificity(y_true, y_pred):
+    import tensorflow as tf
+    y_true = tf.minimum(tf.maximum(y_true - 0.499, 0) * 10000, 1)
+    y_pred = tf.minimum(tf.maximum(y_pred - 0.499, 0) * 10000, 1)
+    ones_tensor = tf.ones_like(y_true)
+    cn_tensor = ones_tensor - y_true
+    cn = K.sum(cn_tensor)
+    y_pred_n = ones_tensor - y_pred
+    tn = K.sum(cn_tensor * y_pred_n)
+    return tn/cn
+
+def binary_specificity_loss(y_true, y_pred):
+    import tensorflow as tf
+    y_true = tf.minimum(tf.maximum(y_true - 0.499, 0) * 10000, 1)
+    y_pred = tf.minimum(tf.maximum(y_pred - 0.499, 0) * 10000, 1)
+    ones_tensor = tf.ones_like(y_true)
+    cn_tensor = ones_tensor - y_true
+    cn = K.sum(cn_tensor)
+    y_pred_n = ones_tensor - y_pred
+    tn = K.sum(cn_tensor * y_pred_n)
+    return 1 - tn/cn