Predicting Baby Weight with TensorFlow on Cloud ML EngineGo to Lab
Really good hands-on Google Datalab example of using Google Dataflow (= Apache Beam) + ML Engine + Deep Learning with TensorFlow to predict numerical values
The training in Cloud ML did not work (it ended with an error), so I could not even try out the model.
Failed to complete lab due to errors. First there were errors in notebook that permission denied for BigQuery. When trying to do everything again the datalab web preview returned "Error: Could not connect to Cloud Shell on port 8081."
There is a error in the lab. It looks like the csv file has change from 10 fields down to 6 fields
Deploying the ML model tooks endless.
important content but a really buggy lab: output directory writes fail and dir not used in reading (preproc_df vs preproc), inconsistent use of filenames (00000* and 00001*), example train and eval data incompatible with code (6 columns vs 10 columns), testing rest is limited to a few regions which were not mentioned in the beginning (slow)
Pre-processed files could not be run without significant changes to the notebook, columns and their order are not the same.
Notebook hasn't been updated to match changes in the source data! Yet another waste of an hour of my life.