Tutorial fixing

[DEKHTIARJonathan]

This PR aims to insure that all the tutorial are working fine and nicely.

This PR solve the following bugs:

• Issue something wrong in tutorial_word2vec_basic.py #476 with model saving in tutorial_word2vec
• tf.flags changed for tf.app.flags
• various bugs in all files fixed

Tested Tutorials:

• tutorial_atari_pong.py
• tutorial_binarynet_cifar10_tfrecord.py
• tutorial_binarynet_mnist_cnn.py
• tutorial_bipedalwalker_a3c_continuous_action.py => FAILING
• tutorial_cartpole_ac.py
• tutorial_cifar10.py
• tutorial_cifar10_tfrecord.py
• tutorial_dorefanet_cifar10_tfrecord.py
• tutorial_dorefanet_mnist_cnn.py
• tutorial_frozenlake_dqn.py
• tutorial_frozenlake_q_table.py
• tutorial_generate_text.py
• tutorial_imagenet_inceptionV3_distributed.py => Not tested yet
• tutorial_image_preprocess.py
• tutorial_imdb_fasttext.py
• tutorial_inceptionV3_tfslim.py
• tutorial_keras.py
• tutorial_matrix.py
• tutorial_mlp_dropout1.py
• tutorial_mlp_dropout2.py
• tutorial_mnist.py
• tutorial_mnist_distributed.py => Not tested yet
• tutorial_mnist_float16.py
• tutorial_mnist_simple.py
• tutorial_mobilenet.py
• tutorial_models_mobilenetv1.py
• tutorial_models_squeezenetv1.py
• tutorial_models_vgg16.py
• tutorial_ptb_lstm.py
• tutorial_ptb_lstm_state_is_tuple.py
• tutorial_squeezenet.py
• tutorial_ternaryweight_cifar10_tfrecord.py
• tutorial_ternaryweight_mnist_cnn.py
• tutorial_tfrecord.py
• tutorial_tfrecord2.py
• tutorial_tfrecord3.py
• tutorial_tfslim.py
• tutorial_tf_dataset_voc.py
• tutorial_vgg16.py
• tutorial_vgg19.py
• tutorial_word2vec_basic.py

[zsdonghao]

Hi, tl.files.save_npz with sess=sess works fine now?

[DEKHTIARJonathan]

@zsdonghao Nope, I have just applied your workaround.
And I also modified a few errors I have noticed in the tutorials. I will check that all of them run fine.

[DEKHTIARJonathan]

@zsdonghao Tutorial VGG16 is not working, can't load the checkpoint.
I have cleaned it a little but still not working, you can checkout the branch of this PR if you want ;)

[DEKHTIARJonathan]

@zsdonghao @lgarithm I don't have time to investigate why some models are not working.
I didn't try the distributed ones, I don't have the setup for it.

Could you please checkout the branch tutorial_fix and try to fix them ;)

Please tell me when some of them are fixed. That way I can flag them as good. Thanks

[zsdonghao]

for tutorial_tf_dataset_voc.py is seem there is a problem on the VOC website...

http://host.robots.ox.ac.uk/pascal/VOC/voc2007/

http://host.robots.ox.ac.uk/pascal/VOC/voc2012/

We can try later?

[zsdonghao]

for tutorial_generate_text.py, I try in on my mac, it works well:

Start learning a model to generate text
Epoch: 1/100 Learning rate: 1.00000000
0.002 perplexity: 8721.404 speed: 2486 wps
0.100 perplexity: 1586.901 speed: 3974 wps
0.199 perplexity: 1021.575 speed: 4101 wps
0.297 perplexity: 766.037 speed: 4146 wps
0.395 perplexity: 625.603 speed: 4162 wps
0.493 perplexity: 533.806 speed: 4164 wps
0.592 perplexity: 481.060 speed: 4146 wps
0.690 perplexity: 439.476 speed: 4109 wps
0.788 perplexity: 404.770 speed: 4090 wps
0.886 perplexity: 375.800 speed: 4086 wps
0.985 perplexity: 352.464 speed: 4080 wps
Epoch: 1/100 Train Perplexity: 350.236
1 : it is a long time , and i would like that . we can not do that . we can not do that . we can not do that . we can not
3 : it is a much , but the fact is , we will make a new new jobs . we need a deal to be a long time , and i want to the
5 : it is a much for the new york . the way , the fact , i want a great time , and i had it to do , the way , and i
10 : it is a long , a lot of a lot to the oil , but a good , the world would not be the “ . ” would like the obama , and

...

[DEKHTIARJonathan]

@zsdonghao there was an error with Unicode and Python3, I just fixed it and now it's working.

Only 4 tutorials to fix and 2 to more to test.

[DEKHTIARJonathan]

tutorial_tf_dataset_voc.py fixed thanks to #657

[DEKHTIARJonathan]

@luomai and myself agreed to merge and fix later the 3 remaining tests