From 2457c5635a96de614538c965948601ba7fdfb81b Mon Sep 17 00:00:00 2001 From: Haibin Lin Date: Thu, 15 Nov 2018 23:13:25 -0800 Subject: [PATCH] Update basic_layers.py --- python/mxnet/gluon/nn/basic_layers.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/python/mxnet/gluon/nn/basic_layers.py b/python/mxnet/gluon/nn/basic_layers.py index d26841977ac2..697eed1ee03a 100644 --- a/python/mxnet/gluon/nn/basic_layers.py +++ b/python/mxnet/gluon/nn/basic_layers.py @@ -160,9 +160,9 @@ class Dense(HybridBlock): Activation function to use. See help on `Activation` layer. If you don't specify anything, no activation is applied (ie. "linear" activation: `a(x) = x`). - use_bias : bool + use_bias : bool, default True Whether the layer uses a bias vector. - flatten: bool + flatten: bool, default True Whether the input tensor should be flattened. If true, all but the first axis of input data are collapsed together. If false, all but the last axis of input data are kept the same, and the transformation