Week 3/27

[agallagher23]

3-24 Progress Outline.pdf

[agallagher23]

Tasks completed:

• Worked with group to define what we think success might look like in the next couple of weeks
• Collaborated to compile outline and send to WF

[xybljy0122]

3/27 meeting w.wells Fargo

Issue: unable to run the PiML high code Rosh will look at the code

Bin - more flexible the model

Hyperparameter does not have to match More concerned to compare overfit/undersfit (Only looking at the Complexity)

One regression and one classification at least. One dataset not conclusive. California housing (regression) and another one classification data set.

[xybljy0122]

3-27 Progress Outline.pdf

[everettbu]

Found error in PiML, communicating with Nengfeng on how to fix

[reinabhatkuly]

Tasks completed:

• Worked with group on plan to optimize parameters, overfitting and underfitting
• Worked with group on creating a detailed outline of the final report

Some Meeting Notes:

• Max depth not as important
• Focus on comparison using overfitting and undercutting as making the models similar to each other would bee challenging
• EBM - not necessary to use max depth - could use number of interactions and number of beings (more beings more model flexibility of model).
• Light GBM -use max depth
• Not necessary that they have the same hyper parameters
• Two datasets - try one classification and one rational to see how models compare to one another

[agallagher23]

We all met together this afternoon to try to troubleshoot the PiML errors and work on the final report. With the errors still persisting, we are not able to do any comparative work. In the meantime, we will work on the final report. That said, we are feeling anxious to get the errors figured out so that we are not behind. Everett emailed the WF team asking for additional guidance.

Also, we are hoping to clarify how important it is for us to use another dataset (a classification dataset in addition to the California housing regression dataset). @mikeizbicki - what are your thoughts on an additional dataset?

[mikeizbicki]

Also, we are hoping to clarify how important it is for us to use another dataset (a classification dataset in addition to the California housing regression dataset). @mikeizbicki - what are your thoughts on an additional dataset?

I think it should be pretty easy to do a classification dataset as well once you have the code working again. I think the PiML library had a handful of classification datasets built-in, so just pick one of those and use it. There's no need to find a separate one.