Merge pull request #68 from AnjaSei/master

Global Average Pooling

deeplift.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: deeplift
-Version: 0.6.7.1
+Version: 0.6.8.0
 Summary: DeepLIFT (Deep Learning Important FeaTures)
 Home-page: https://github.com/kundajelab/deeplift
 License: UNKNOWN

deeplift/conversion/kerasapi_conversion.py

@@ -262,6 +262,12 @@ def maxpool1d_conversion(config, name, verbose,
              **pool1d_kwargs)]
 
 
+def globalavgpooling1d_conversion(config, name, verbose, **kwargs):
+    return [layers.GlobalAvgPool1D(
+             name=name,
+             verbose=verbose)]
+
+
 def avgpool1d_conversion(config, name, verbose, **kwargs):
     pool1d_kwargs = prep_pool1d_kwargs(
                         config=config,
@@ -305,6 +311,7 @@ def layer_name_to_conversion_function(layer_name):
         'maxpooling1d': maxpool1d_conversion,
         'globalmaxpooling1d': globalmaxpooling1d_conversion,
         'averagepooling1d': avgpool1d_conversion,
+        'globalaveragepooling1d': globalavgpooling1d_conversion,
 
         'conv2d': conv2d_conversion,
         'maxpooling2d': maxpool2d_conversion,

deeplift/layers/pooling.py

@@ -171,6 +171,37 @@ def _get_mxts_increments_for_inputs(self):
         return pos_mxts_increments, neg_mxts_increments 
 
 
+class GlobalAvgPool1D(SingleInputMixin, Node):
+
+    def __init__(self, **kwargs):
+        super(GlobalAvgPool1D, self).__init__(**kwargs)
+
+    def _compute_shape(self, input_shape):
+        assert len(input_shape)==3
+        shape_to_return = [None, input_shape[-1]]
+        return shape_to_return
+
+    def _build_activation_vars(self, input_act_vars):
+        return tf.reduce_mean(input_act_vars, axis=1)
+
+    def _build_pos_and_neg_contribs(self):
+        inp_pos_contribs, inp_neg_contribs =\
+            self._get_input_pos_and_neg_contribs()
+        pos_contribs = self._build_activation_vars(inp_pos_contribs)
+        neg_contribs = self._build_activation_vars(inp_neg_contribs)
+        return pos_contribs, neg_contribs
+
+    def _grad_op(self, out_grad):
+        width = self._get_input_activation_vars().get_shape().as_list()[1]
+        mask = tf.ones_like(self._get_input_activation_vars()) / float(width)
+        return tf.multiply(tf.expand_dims(out_grad, axis=1), mask)
+
+    def _get_mxts_increments_for_inputs(self):
+        pos_mxts_increments = self._grad_op(self.get_pos_mxts())
+        neg_mxts_increments = self._grad_op(self.get_neg_mxts())
+        return pos_mxts_increments, neg_mxts_increments
+
+
 class Pool2D(SingleInputMixin, Node):
 
     def __init__(self, pool_size, strides, padding, data_format, **kwargs):

setup.py

@@ -8,7 +8,7 @@
 Implements the methods in "Learning Important Features Through Propagating Activation Differences" by Shrikumar, Greenside & Kundaje, as well as other commonly-used methods such as gradients, guided backprop and integrated gradients. See https://github.com/kundajelab/deeplift for documentation and FAQ.
           """,
           url='https://github.com/kundajelab/deeplift',
-          version='0.6.7.1',
+          version='0.6.8.0',
           packages=['deeplift',
                     'deeplift.layers', 'deeplift.visualization',
                     'deeplift.conversion'],