fix nightly CI failure

[roywei]

should fix:
(1) #15374
Nightly test on amp tutorial is still failing after reverting the large tensor PR due to new change on the tutorial in #15118 requires np.float32

------------------



---------------------------------------------------------------------------

NameError                                 Traceback (most recent call last)

<ipython-input-11-7f79a6994fc9> in <module>

      9     # Run dummy inference with the converted gluon model

     10     result = converted_model.forward(mx.nd.random.uniform(shape=(1, 3, 224, 224),

---> 11                                                           dtype=np.float32))

     12 

     13     # Below is an example of converting a symbolic model to a mixed precision model



NameError: name 'np' is not defined

NameError: name 'np' is not defined

(2) #15199
using fixed seed to reduce flakiness in accuracy assertion.

[roywei]

cc @lebeg @apeforest @anirudh2290

[szha]

this shouldn't be done here. there's a decorator with_seed for this purpose

[roywei]

@mxnet-label-bot add [CI, test]

[lebeg]

@roywei could you apply changes that @szha requested? The nightly tests are still broken.

[roywei]

I have verified test_estimator passed. but test tutorials is still failing due to other tutorials are failing. This change is needed nonetheless

----------------------------------------------------------------------
Ran 2 tests in 341.805s

OK
build.py: 2019-07-12 01:33:11,865Z INFO Waiting for status of container 594316b5c329 for 600 s.
build.py: 2019-07-12 01:33:12,139Z INFO Container exit status: {'Error': None, 'StatusCode': 0}
build.py: 2019-07-12 01:33:12,139Z INFO Container exited with success 👍
build.py: 2019-07-12 01:33:12,140Z INFO Stopping container: 594316b5c329
build.py: 2019-07-12 01:33:12,141Z INFO Removing container: 594316b5c329