My takeaway from hosting my second workshop

This second workshop I decided to take was related to machine learning using AWS resources and for those unaware of what AWS is, AWS is better known as Amazon Web Services offered by the company we all love shopping from, Amazon.
Continuing from my previous article where I described how my first workshop on Angular JS unfolded, This article shall articulate my experience with my second workshop.

Background: I used to attend classes in LA on working with EC2 instances and setting up virtual machines using AWS. On one such similar instance of attending workshops and meetups, I attended one other on AWS Machine learning hosted by Venturesity in Bengaluru. A good chunk of my friends working in Information Technology sector and some were trying to make a breakthrough in Machine Learning. The number of algorithms and complexity of math involved in many of the MOOCs that I tried learning from were quite overwhelming. After attending this session on AWS machine learning hosted by venturesity with my father, I understood how to upload datasets, train and run batch predictions using it.

This was literally all that I needed. I can now run close to a 1000 predictions for a little over 10 cents or less than 7 Indian Rupees. It no longer made sense for me to indulge in a process that is going to involve me working with tons of algorithms and all that ML mumbo jumbo. I had a simple black Box where I could just chuck in data and it would model and predict batch results almost immediately after deciding upon which algorithm to use and all that hullabaloo. As a person that needed predictions with limited time and money, Amazon Machine Learning was the most appropriate option I found available.
After hours of learning, I decided to take a class for they’d be more people this could have use for.

Day of the session:
A little over Fifty people had RSVP’d on a meetup group for Lean Startup enthusiasts run by the founder of a coworking space I was supposed to conduct my workshop at.

Five people turned up !

This is certainly a lot more than how many turned up for my previous session, but this time there were genuine people that were ready to learn what I had to teach. I had hoped it to be more of a hands on workshop, but the code for promotional credit that was handed out at the venturesity event refused to work. A hurdle. We decided to switch from it being a hands on session to a classroom demo setting to continue with the session.

Within our session, we covered how to train, model and run batch predictions on our sample dataset. Furthermore, I was hoping to let people download and work on datasets from Kaggle to generate their own predictions. We were unable to execute this since people were unable to use their free credit coupon from the previous session.

There were a couple of questions that were asked during this session and I shall be listing them and answers to those questions below.

Questions:

If I have a dataset, could I generate a model using a portion of the dataset and predict based on the remaining data ?
Certainly, As you generate your model, you have options to split it into 70%-30% or use another dataset with same schema to predict against where it can be facilitated.

How do I secure my Datasource and my predicted results ?
Amazon Machine Learning has options to use data stored on S3 and RedShift.
We may allocate permissions to different people based upon their role in the organization and hence control who may or may not view sensitive data.

Could I manually modify the schema and change target data to be predicted ?
Amazon Machine Learning analyses the data that is used to train and generates a schema on its own. We are at liberty to change and modify it prior to running our predictions. We are allowed to set our own targets to be predicted.

How do I improve the accuracy of the prediction?
One may improve the accuracy of the prediction provided by increasing training examples. Ignoring or defining relationships (Underfitting-Overfitting) between Target variable and input values would help differentiate and return accurate results.

References:
Documentation:What is Amazon Machine Learning ?
Improving Accuracy
UnderFitting vs Overfitting
Managing Amazon ML objects

Post this session, I shall be sending everyone a gist of what happened in the session, Questions and Answers and the Promotional code to help them get Kickstarted.

My takeaway from hosting this session was to double check to see if references and material to be provided are worthwhile and working to carry out the functions beforehand.

Next session, we could have a Q&A session before and after since it certainly helped cover the nitty gritty aspects. References and Study material to be offered ought to be tested by a third person to ensure that they are working.
I shall reach out to the company providing the service to help the next time.

This is what I learnt after hosting this session.

I hope this was helpful.

Prahalad Belavadi