berkeley-stat159 / project-eta Goto Github PK

With PR #40, I realized that we need to standardize where we test our code from. What probably makes the most sense is to do it from tests/. However, the problem is that, for load_data.py specifically, we will usually call those modules from code/. So, the relative paths will not point to the same place.

Thoughts?

Update slides/progress.md with your team name and team members

bash script: https://github.com/berkeley-stat159/project-eta/blob/master/data/getdata.sh

Note that the buttons in your README have references to project-delta in them...

Add a description of what we "plan to do with the data [we've] chosen."

Include a citation to the paper you are using as the basis for your project in paper/project.bib

I'm not seeing the checksums file in the repository or a way to access it. See data.py.

Add project information to README.md.

@yuanyiwu do you want to take a shot at this? Maybe add some info on the paper itself and, if possible, what we plan to do (this may change).

• Be careful about your claims on the benefits of smoothing: I would disagree
  with them, but I think it's mostly a phrasing issue
• Include images that display the results of your preprocessing steps. For
  example, you could show smoothed vs. the original slices and images from the
  PCA
• In section 3.2.1: I'm a little unclear, did you combine all of the runs
  across ALL the patients, or just all of the runs for a single patient for
  the logistic regression? Make sure to make your language more precise for the
  final draft
  • Also, you say it is difficult to discern the whether neural activity is
    due to negative loss response or positive gain respones, but from your
    regression model, it appears you have two different regressors for these
    2 cases - shouldn't you be able to determine something about this by
    looking at the results for the two regressors? Also, couldn't you try
    different regressors that perhaps deal with differences and ratios of
    gain and loss to see if they are more enlightening?
• You very much need to push harder to fill out the results section: at this
  point there is no analysis of either the regression or MVPA approaches.

Create convolved data using a gamma function.
Build a linear regression on the convolved data.

Hey Matt. We updated our design matrix so that we are convolving the gain and loss regressors as well like you suggested.

Can you help us interpeting these images? We notice that there are some small clusters of activiation, but what else can we do to analyze and interpet these images?

The initial description of the project, paper, and goals was not particularly
clear.

The main goal appears to be to reproduce results from the paper - identified
specific results to attempt to replicate (image from paper), but you didn't
present any preliminary analysis towards that aim.

Identified struggles with applying preprocessing and linear modelling concepts
from class to their data. This is a good first step and effort should be
redoubled to accomplish this ASAP.

Seemed to have a good grasp of their overall goal, but no specific, short-term,
managable goals toward that end. You need to complete the regression analysis
to better understand the data and the challenges of reaching their goal to
reproduce the paper's results.

You appear to be using statsmodels, but haven't mentioned it as a dependency.
In order for me to be able to run your code, you need to explicitly mention
dependencies as discussed during lecture 18.

You still have a lot of work to do, but there is still time.

In the graphs tab in this repo, it shows that I have only 3 commits. I know that I have a lot more than that so I am just wondering if I am doing something wrong with my commits. I also do not want to get marked down when I am actually contributing a good portion to our repo. Thanks!

There are unnecessary lines in data/tests/test_data.py. Review and remove them

We need this.