Machine Learning Exploratory

For the last three weeks I have been exploring Google Cloud and the GCloud ML (Machine Learning Engine) via a video labeling competition on Kaggle.  I thought I had no expectations other than that it would require me to apply the techniques learned in our Consumer Decision-Making class in a more depth project, and given this was a video classification endeavor there'd be some significant learning.  In fact, one of my expectations, which I took for granted, was that there would be some type of graphic user interface, a la SAS Enterprise Miner.  Initially I expected it to be Google's Tensorboard software, a modeling tool, which so far as I have experienced is not manipulatable, though it appears to be marketed as such. Maybe that comes later in the game. So far I have simply completed the training, but am stuck at building a model.  So here are my learnings for marketers wishing to try this out on their own.

1. Talk about a steep learning curve.  A solid knowledge of Python is a MUST.  Last year I took an intro C++ class.  The syntax is different but the principles are the same so far.  Sadly, rudimentary knowledge of C++ or in my case a crash course in Python via my kid's "Python for Kids" book isn't enough to follow more than the conceptual framework of what is going on, even if you look up every function name on the screen.  In this case much of the code is given to you, and it appears that all you have to do is plug and run, but mapping to files is a stinker, and understanding the basics, like why when you want to import a library from a source it says it doesn't have access to that source, and knowing how to troubleshoot that, can be a real roadblock.

2.  Graphic Interface Tools are you friend, and as obtuse as they can seem when you're first introduced to them, when you jump from that back to the building blocks using code alone, the GUI option begins to feel like child's play by comparison.  So, my overall takeaway,  learn the GUI tools cold, and it pray that they're the next step to all AI software development.

3.  Be patient, and thorough.  The more I read, re-read, and and look up each miniscule related term and feature, the more it makes sense.  3 textbooks and several website tutorials later it begins to feeling somewhat within grasp.

4.  Video-tutorials would be awesome.  I can't quite figure out why, but there are no video tutorials linked to Google Cloud or GitHub instructions, not even with a Quick Start.

If you are interested in learning more, here are a few videos that are not related to the service providers, but may be helpful for you.  Sentdex suggests learning Python up to level 3 right off the bat in this video.

https://youtu.be/OGxgnH8y2NM?list=PLQVvvaa0QuDfKTOs3Keq_kaG2P55YRn5v

And on that note, I will let you go, because point #5 is that learning this is time-consuming.  If at session's end I can give step by step recommendations for pulling it all together without jumping all over the place to do so, I will.

Comments

Popular posts from this blog

Data Visualization Tool of the Day-Plotly

Are Digital Marketers Endangered?

Business Analytics in Politics