Fix imagenet training script with data loader (#184)

* in initial structure

* fix

* improve verification

* slight modification

* update tutorial

* prepare recio

* fix training script

* fix tutorial

* update scheduler to include warmup, cosine and step

* update tutorials

* fix

* fix

* fix lint

* Trigger CI

* improve doc

* change save_params to save_parameters

* temporary download test files

* update download command

* fix

* add imagenet mixup

* fix

* fix

* update training results

* remove test commands in jenkins

* fix tutorial

* fix

* Trigger CI

* Trigger CI

* download files for tutorial

* update

* fix

* remove download scripts, waiting for rec file to be added

* retry download

* Trigger CI

* remove download commands

* including download, update required pip version

* download for doc building

* fix path

* finalize tutorial shape, hold for pr in mxnet repo

* update resnet152 results

* need the nightly built mxnet

* fix yml

* fix doc

* fix doc

* fix doc?

* rename class, fix doc

* fix

* improve

* fix doc

* fix training script

* fix

scripts/classification/imagenet/train_imagenet.py

@@ -197,7 +197,7 @@ def batch_fn(batch, ctx):
         normalize
     ])
     transform_test = transforms.Compose([
-        transforms.Resize(256),
+        transforms.Resize(256, keep_ratio=True),
         transforms.CenterCrop(224),
         transforms.ToTensor(),
         normalize