brainhack-school2020 / abide-fmri Goto Github PK

@mschoettner will work on the "leave site out" cross validation classifier script.

@emilyemchen will work on a CV classifier script that does group 10-fold cross validation.

e.g. Educational background, learning goals, etc.

Download the files as .png and then insert them into jupyter notebooks in Markdown cells.
Format: ![SlideX](slides/X.png)

For the preparation function it would be nice if it would check if the features have already been extracted and then use the saved file instead.

In order for everyone to be able to run the scripts, we should include a Requirements.txt file that lists the required packages. We could also think about alternative ways to make sure everyone can run the code, like virtual environments or containers.

Create the notebook we will use for our final presentation, implemented in Rise.

See scripts for leave-site-out or group-k classifier for issue when loading dataset with the following code:

#Import modules for this step
from nilearn import datasets
import pandas as pd
import os
#Fetch data using nilearn.datasets.fetch
abide = datasets.fetch_abide_pcp(data_dir=/path/to/data),
                                 pipeline="cpac",
                                 quality_checked=True)

#Load phenotypic data into pandas dataframe
abide_pheno = pd.DataFrame(abide.phenotypic)

#Create array to hold unique site names
groups = abide_pheno["SITE_ID"].unique()

Receiving this warning message:

VisibleDeprecationWarning: Reading unicode strings without specifying the encoding argument is deprecated. Set the encoding, use None for the system default.
  output = genfromtxt(fname, **kwargs)

Any idea what's going on here? Does not necessarily prevent script from running.

@pbellec @surchs

Section that goes over the project, its goal, and the results we obtained

Try to create an environment that contains just the code necessary to run the scripts from this repo, then include instruction to do so in the README.

I tried running the script after your commit @anproulx , but I am getting the following error:

Traceback (most recent call last):
  File "./script-scaffolding.py", line 100, in <module>
    run_analysis()
  File "./script-scaffolding.py", line 91, in run_analysis
    X_features_pca, y_target = prepare_data(args.data_dir, args.output_dir)
  File "./script-scaffolding.py", line 56, in prepare_data
    X_features[i]=list(X_features[i][np.triu_indices(64)])
IndexError: too many indices for array

Also the function prepare_data does not return anything, but it should return X_features and y_target.

Please create a new branch, fix the errors on the branch by testing the code and then do a pull request.

Also make sure to add appropriate sections @emilyemchen.

It would be nice if there was some information on each of the notebooks, summarizing what it does and why in the README file.

How do we want to do this? Should we have the file be stored in the same directory in this GitHub repo or should we just make a note that the user should download the files and put them all in the same directory? I think it would be better to have all of the files downloaded in the correct nesting personally.

@mschoettner

In order to make code more modular, the script that is now the scaffolding will be converted to just do the preparation of the data.

Hi guys ,

Should we reorganize our abide-fmri repo? Maybe put all our .ipynb in a separate folder... prep data+requirements together? What do you think?

Since it was a big part of the learning, explain the tools we used to work as a team.

@anproulx will work on the 10-fold cross validation script.

Hi @glatard @agahkarakuzu,
Our group has added the section to our README.md file with links to our GitHub repositories for our Week 3 data visualizations!

It would be nice if the scripts could be called from the command line using arguments. This can be done with an argument parser such as argparse.

Before going into the ML, to get our feature (X) ready, we need to:

1. Remove half the matrix since it is redundant
2. Apply PCA to keep only relevant features (less computational cost also)

I will add this code to the script-scaffolding.py