Add repr for SymbolBlock

[vandanavk]

Description

SymbolBlock uses Block.__repr__ when print(SymbolBlock) is executed. In Block.__repr__, data is printed only if it is of type Block. Adding a __repr__ to SymbolBlock class that prints the output symbol. This behavior is similar to printing pure Symbol API.

Fixes #13616

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

• The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant JIRA issue created (except PRs with tiny changes)
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage:
• Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
• Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
• Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
• Code is well-documented:
• For user-facing API changes, API doc string has been updated.
• For new C++ functions in header files, their functionalities and arguments are documented.
• For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
• Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html
• To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

Changes

• Add a repr for SymbolBlock
  Executing

import mxnet as mx
from mxnet.gluon.model_zoo import vision

alexnet = vision.alexnet(pretrained=True)
alexnet.initialize()
alexnet.hybridize()
alexnet(mx.random.uniform(shape=(1,3,224,224)))
alexnet.export("./alexnet")

block = mx.gluon.nn.SymbolBlock.imports('alexnet-symbol.json', ['data'], 'alexnet-0000.params')
print(block)

gives the output
Before

SymbolBlock(

)

After change

SymbolBlock(
<Symbol alexnet0_dense2_fwd> : 1 -> 1
)

Example with a model with more than one output:

import mxnet as mx
from mxnet import gluon

data = mx.sym.Variable('data')
topk = mx.sym.topk(data, k=3, ret_typ='both')
ctx = mx.cpu()

pre_trained = gluon.nn.SymbolBlock(outputs=topk, inputs=data)
print (pre_trained)

Output:

SymbolBlock(
<Symbol topk2> : 1 -> 2
)

Comments

@eric-haibin-lin @Ishitori

[vandanavk]

@mxnet-label-bot add [Gluon, pr-awaiting-review]

[szha]

Could you add a test? Also, what does the current string representation look like, for example for alexnet?

[eric-haibin-lin]

+1 on szha's comment

[vandanavk]

Thanks for inputs @szha @eric-haibin-lin. Added an example in the PR description. Will add a test and submit shortly

[vandanavk]

@szha @eric-haibin-lin added a test in test_import() of test_gluon.py

[szha]

why use __dict__? or -1 index?

[szha]

would it be more useful to show also the input count and output count?

[vandanavk]

corrected dict to self._cached_graph.
-1 index as the last element of cached_graph is the output. I could have used cached_graph[1], but did not want to hit index out of bounds (in case it ever happens).

[vandanavk]

I think showing all layers like how HybridSequential prints, would be most useful, but not sure how to do that for Symbol API and SymbolBlock models.

[szha]

Printing the inner structure might be too complicated and also not readable, thus I suggested printing the count.

Why would cached_graph[1] ever hit index out of bounds?

[vandanavk]

something like this?

SymbolBlock(
<Symbol dense1_fwd>
)
Number of inputs: 1
Number of outputs: 1

[szha]

<Symbol dense1_fwd>: 1->1 ?

[vandanavk]

@szha thanks for the inputs. submitted after addressing the comments. Added an example of multiple outputs in the PR description

[wkcn]

Thanks for your contribution! LGTM
I think the test case Is too strict. We can use 'in' to test it.

[vandanavk]

@wkcn Do you mean lines[0] check? I did something similar to https://github.com/apache/incubator-mxnet/blob/a6a4fe188d4847a608952fb69ff68e7ab91c5050/tests/python/unittest/test_gluon.py#L237