add layer_dropout

Author: ewrfcas

KerasLayer/QAoutputBlock.py

@@ -0,0 +1,25 @@
+# ! -*- coding: utf-8 -*-
+from keras.engine.topology import Layer
+from keras.regularizers import *
+import tensorflow as tf
+import keras.backend as K
+
+class QAoutputBlock(Layer):
+    def __init__(self, ans_limit=30, **kwargs):
+        self.ans_limit = ans_limit
+        super(QAoutputBlock, self).__init__(**kwargs)
+
+    def build(self, input_shape):
+        super(QAoutputBlock, self).build(input_shape)
+
+    def call(self, x, mask=None):
+        x1 ,x2 = x
+        outer = tf.matmul(tf.expand_dims(x1, axis=2), tf.expand_dims(x2, axis=1))
+        outer = tf.matrix_band_part(outer, 0, self.ans_limit)
+        output1 = tf.reshape(tf.cast(tf.argmax(tf.reduce_max(outer, axis=2), axis=1), tf.float32),(-1,1))
+        output2 = tf.reshape(tf.cast(tf.argmax(tf.reduce_max(outer, axis=1), axis=1), tf.float32),(-1,1))
+
+        return [output1, output2]
+
+    def compute_output_shape(self, input_shape):
+        return [(input_shape[0][0],1), (input_shape[0][0],1)]

KerasLayer/layer_dropout.py

@@ -0,0 +1,22 @@
+# ! -*- coding: utf-8 -*-
+from keras.engine.topology import Layer
+import tensorflow as tf
+import keras.backend as K
+
+class LayerDropout(Layer):
+    def __init__(self, dropout = 0.0, **kwargs):
+        self.dropout = dropout
+        super(LayerDropout, self).__init__(**kwargs)
+
+    def build(self, input_shape):
+        super(LayerDropout, self).build(input_shape)
+
+    def call(self, x, mask=None, training=None):
+        x, residual = x
+        pred = tf.random_uniform([]) < self.dropout
+        x_train = tf.cond(pred, lambda: residual, lambda: tf.nn.dropout(x, 1.0 - self.dropout) + residual)
+        x_test = x + residual
+        return K.in_train_phase(x_train, x_test, training=training)
+
+    def compute_output_shape(self, input_shape):
+        return input_shape