tensorflow Using Batch Normalization
Parameters
contrib.layers.batch_norm params | Remarks |
|---|---|
beta | python bool type. Whether or not to center the moving_mean and moving_variance |
| ------ | ------ |
gamma | python bool type. Whether or not to scale the moving_mean and moving_variance |
| ------ | ------ |
is_training | Accepts python bool or TensorFlow tf.palceholder(tf.bool) |
| ------ | ------ |
decay | The default setting is decay=0.999. A smaller value (i.e. decay=0.9) is better for smaller dataset and/or less training steps. |
Remarks
Here is a screen shot of the result of the working example above.
The code and a jupyter notebook version of this working example can be found at the author's repository
