ThomasDelteil commented on a change in pull request #10511: add naming tutorial
URL: https://github.com/apache/incubator-mxnet/pull/10511#discussion_r180946437
##########
File path: docs/tutorials/gluon/naming.md
##########
@@ -0,0 +1,236 @@
+
+# Naming of Gluon Parameter and Blocks
+
+In gluon, each Parameter or Block has a name (and prefix). Parameter names are
specified by users and Block names can be either specified by users or
automatically created.
+
+In this tutorial we talk about the best practices on naming. First, let's
import MXNet and Gluon:
+
+
+```python
+from __future__ import print_function
+import mxnet as mx
+from mxnet import gluon
+```
+
+## Naming Blocks
+
+When creating a block, you can assign a prefix to it:
+
+
+```python
+mydense = gluon.nn.Dense(100, prefix='mydense_')
+print(mydense.prefix)
+```
+
+ mydense_
+
+
+When no prefix is given, Gluon will automatically generate one:
+
+
+```python
+dense0 = gluon.nn.Dense(100)
+print(dense0.prefix)
+```
+
+ dense0_
+
+
+When you create more Blocks of the same kind, they will be named differently
to avoid collision:
+
+
+```python
+dense1 = gluon.nn.Dense(100)
+print(dense1.prefix)
+```
+
+ dense1_
+
+
+## Naming Parameters
+
+Parameters within a Block will be named by prepending the prefix of the Block
to the name of the Parameter:
+
+
+```python
+print(dense0.collect_params())
+```
+
+ dense0_ (
+ Parameter dense0_weight (shape=(100, 0), dtype=<type 'numpy.float32'>)
+ Parameter dense0_bias (shape=(100,), dtype=<type 'numpy.float32'>)
+ )
+
+
+## Name scopes
+
+To manage the names of nested Blocks, each Block has a `name_scope` attached
to it. All Blocks created within a name scope will have its parent Block's
prefix prepended to its name.
+
+Let's demonstrate this by first define a simple neural net:
+
+
+```python
+class Model(gluon.Block):
+ def __init__(self, **kwargs):
+ super(Model, self).__init__(**kwargs)
+ with self.name_scope():
+ self.dense0 = gluon.nn.Dense(20)
+ self.dense1 = gluon.nn.Dense(20)
+ self.mydense = gluon.nn.Dense(20, prefix='mydense_')
+
+ def forward(self, x):
+ x = mx.nd.relu(self.dense0(x))
+ x = mx.nd.relu(self.dense1(x))
+ return mx.nd.relu(self.mydense(x))
+```
+
+Now let's instantiate our neural net.
+
+- Note that `model0.dense0` is named as `model0_dense0_` instead of `dense0_`.
+
+- Also note that although we specified `mydense_` as prefix for
`model.mydense`, its parent's prefix is automatically prepended to generate the
prefix `model0_mydense_`.
+
+
+```python
+model0 = Model()
+model0.initialize()
+model0(mx.nd.zeros((1, 20)))
+print(model0.prefix, model0.dense0.prefix, model0.dense1.prefix,
model0.mydense.prefix)
+```
+
+ model0_ model0_dense0_ model0_dense1_ model0_mydense_
+
+
+If we instantiate `Model` again, it will be given a different name like shown
before for `Dense`.
+
+- Note that `model1.dense0` is still named as `dense0_` instead of `dense2_`,
following dense layers in previously created `model0`. This is because each
instance of model's name scope is independent of each other.
+
+
+```python
+model1 = Model()
+print(model1.prefix, model1.dense0.prefix, model1.dense1.prefix,
model1.mydense.prefix)
+```
+
+ model1_ model1_dense0_ model1_dense1_ model1_mydense_
+
+
+**It is recommended that you manually specify prefix for the top level Block
(i.e. `model = Model(prefix='mymodel_')`) to avoid potential confusions in
naming**
+
+The same principle also applies to container blocks like Sequantial.
`name_scope` can be used inside `__init__` as well as out side of `__init__`:
+
+
+```python
+net = gluon.nn.Sequential()
+with net.name_scope():
+ net.add(gluon.nn.Dense(20))
+ net.add(gluon.nn.Dense(20))
+print(net.prefix, net[0].prefix, net[1].prefix)
+```
+
+ sequential0_ sequential0_dense0_ sequential0_dense1_
+
+
+`gluon.model_zoo` also behaves similarly:
+
+
+```python
+net = gluon.nn.Sequential()
+with net.name_scope():
+ net.add(gluon.model_zoo.vision.alexnet(pretrained=True))
+ net.add(gluon.model_zoo.vision.alexnet(pretrained=True))
+print(net.prefix, net[0].prefix, net[1].prefix)
+```
+
+ sequential1_ sequential1_alexnet0_ sequential1_alexnet1_
+
+
+## Saving and loading
+
+Because model0 and model1 have different prefixes, their Parameters also have
different names:
+
+
+```python
+print(model0.collect_params(), '\n')
+print(model1.collect_params())
+```
+
+ model0_ (
+ Parameter model0_dense0_weight (shape=(20L, 20L), dtype=<type
'numpy.float32'>)
+ Parameter model0_dense0_bias (shape=(20L,), dtype=<type 'numpy.float32'>)
+ Parameter model0_dense1_weight (shape=(20L, 20L), dtype=<type
'numpy.float32'>)
+ Parameter model0_dense1_bias (shape=(20L,), dtype=<type 'numpy.float32'>)
+ Parameter model0_mydense_weight (shape=(20L, 20L), dtype=<type
'numpy.float32'>)
+ Parameter model0_mydense_bias (shape=(20L,), dtype=<type
'numpy.float32'>)
+ )
+
+ model1_ (
+ Parameter model1_dense0_weight (shape=(20, 0), dtype=<type
'numpy.float32'>)
+ Parameter model1_dense0_bias (shape=(20,), dtype=<type 'numpy.float32'>)
+ Parameter model1_dense1_weight (shape=(20, 0), dtype=<type
'numpy.float32'>)
+ Parameter model1_dense1_bias (shape=(20,), dtype=<type 'numpy.float32'>)
+ Parameter model1_mydense_weight (shape=(20, 0), dtype=<type
'numpy.float32'>)
+ Parameter model1_mydense_bias (shape=(20,), dtype=<type 'numpy.float32'>)
+ )
+
+
+As a result if you try to save parameters from model0 and load it with model1,
you'll get an error due to unmatching names:
+
+
+```python
+model0.collect_params().save('model.params')
+try:
+ model1.collect_params().load('model.params', mx.cpu())
+except Exception, e:
+ print(e)
+```
+
+ Parameter 'model1_dense0_weight' is missing in file 'model.params', which
contains parameters: 'model0_mydense_weight', 'model0_dense1_bias',
'model0_dense1_weight', 'model0_dense0_weight', 'model0_dense0_bias',
'model0_mydense_bias'. Please make sure source and target networks have the
same prefix.
+
+
+To solve this problem, we use `save_params`/`load_params` instead of
`collect_params` and `save`/`load`. This way, the unmatching part of parameter
names (`model0_` and `model1_`) are stripped and only the matching part is
saved.
+
+
+```python
+model0.save_params('model.params')
+model1.load_params('model.params', mx.cpu())
+print(mx.nd.load('model.params').keys())
+```
+
+ ['dense0_bias', 'mydense_weight', 'mydense_bias', 'dense1_bias',
'dense1_weight', 'dense0_weight']
+
+
+**Nevertheless, you are still recommended to manually specify the prefix of
the top level Block (i.e. `model = Model(prefix='mymodel_')`) to avoid any
potential problem during saving and loading.**
+
+## Replacing Blocks from networks and fine-tuning
+
+Sometimes you may want to load a pretrained model, and replace certain Blocks
in it for fine-tuning.
+
+For example, the alexnet in model zoo has 1000 output dimensions, but maybe
you only have 100 classes in your application.
+
+To see how to do this, we first load an pretrained alexnet.
+
+- Note that the output layer is a dense block with 1000 dimension outputs.
+
+
+```python
+alexnet = gluon.model_zoo.vision.alexnet(pretrained=True)
+print(alexnet.output, alexnet.output.prefix)
+```
+
+ Dense(4096 -> 1000, linear) alexnet0_dense2_
+
+
+To change the output to 100 dimension, we replace it with a new block.
+
+- Note that it's important to do this in alexnet's name_scope, otherwise you
will have unmatching names when you try to save and load your model.
+
+
+```python
+with alexnet.name_scope():
+ alexnet.output = gluon.nn.Dense(100)
Review comment:
Maybe it is worth putting here that this works only because the models in
model zoo are inheriting `HybridBlock` and have been built in a way so that
their final output layer is associated with the `.output` property.
I can see people building their own network using`HybridSequential` for
example and trying to update the output layer by calling `output` on their
network.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
