Remove dependency on Tensorflow by fixing clipping bounds suggested in …

…#3
titu1994 · Mar 11, 2019 · 5ce819b · 5ce819b
1 parent a0344d1
commit 5ce819b
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 9 deletions.
diff --git a/README.md b/README.md
@@ -48,10 +48,6 @@ Weights are available inside the [Releases tab](https://github.com/titu1994/kera
 
 <img src="https://github.com/titu1994/keras-adabound/blob/master/images/val_loss.PNG?raw=true" height=50% width=100%>
 
-# Issue with clipping
-
-Currently dependent on Tensorflow backend for `tf.clip_by_value`. Will be backend independent after next release of Keras.
-
 # Requirements
 - Keras 2.2.4+ & Tensorflow 1.12+ (Only supports TF backend for now).
 - Numpy
diff --git a/adabound.py b/adabound.py
@@ -1,4 +1,3 @@
-import tensorflow as tf
 from keras import backend as K
 from keras.optimizers import Optimizer
 
@@ -100,10 +99,8 @@ def get_updates(self, loss, params):
             # Compute the bounds
             step_size_p = step_size * K.ones_like(denom)
             step_size_p_bound = step_size_p / denom
-            # TODO: Replace with K.clip after releast of Keras > 2.2.4
-            bounded_lr_t = m_t * tf.clip_by_value(step_size_p_bound,
-                                                  lower_bound,
-                                                  upper_bound)
+            bounded_lr_t = m_t * K.minimum(K.maximum(step_size_p_bound,
+                                                     lower_bound), upper_bound)
 
             p_t = p - bounded_lr_t