Cloud ML Engine: Qwik Start进入实验练习
There is an issue in the script. The directory is name 'census' instead of 'servo'
My output didn't match the guide. I'll double check first.
It was a good experience. There were some typos in the instructions, but I still managed to complete the lab. One annoying thing I found out was that I am not able to copy the commands from the lab to the Google Cloud Shell.
problem finding output/export/Servo (ticket submitted)
Pretty good lab with challenge at the end to get everything working. Cool that you can deploy a tensorflow application to the web.
Model export path is wrong (Servo instead of census). Colut ML job failed.
there are errors: Servo directlroy does not create and hence we can't launch the prediction
There was an error during cloud training, it seems to be an issue with tensorflow version. But the lab was very well written.
Had a couple of problems and needed to re-run the learning as first time tensorboard didn't find any output files (although they appeared to be in the output folder). Couldn't locate the Servo directory despite looking, so ended the lab.
steps to clone asking for github credentials that are not in the Setup instructions
Doesn't work. Tried three times. Following are my notes on errors as I went. After local training run, there is no output/export/Servo Used dir under output/export/census instead, in gcloud ml-engine local predict \ --model-dir output/export/census/<timestamp> \ --json-instances ../test.json Running that gave: WARNING: 2017-12-15 14:54:42.047495: I tensorflow/core/platform/cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA CLASS_IDS CLASSES LOGISTIC LOGITS PROBABILITIES  [u'0'] [0.06173502653837204] [-2.7211809158325195] [0.9382650256156921, 0.06173503026366234] Running single instance trainer in cloud failed with Python exception. Key part seems to be: File "/root/.local/lib/python2.7/site-packages/trainer/task.py", line 28, in run_experiment train_spec = tf.estimator.TrainSpec(train_input, AttributeError: 'module' object has no attribute 'TrainSpec' Command '['python', '-m', u'trainer.task', u'--train-files', u'gs://qwiklabs-gcp-37213d53c77191cd-mlengine/data/adult.data.csv', u'- -eval-files', u'gs://qwiklabs-gcp-37213d53c77191cd-mlengine/data/adult.test.csv', u'--train-steps', u'5000', u'--verbosity', u'DEBUG', '--job-dir', u'gs://qwiklabs-gcp-37213d53c77191cd-mlengine/census2']' returned non-zero exit status 1