Add Brain Invaders datasets #283
Conversation
|
Great that there are already some plots. If this is merged, I would still just run my scripts to create plots similar to the ones for the other datasets and to make sure that in the process of adding these to MOABB there is no regression or something. OT: Do we want to "officially" host these plots somwhere? Currently my webserver is still working, but at some point I will probably run into traffic limitations 😅 |
|
Thank you for your help @sylvchev :) Do you think we could also integrate the other datasets in MOABB in the future? |
|
Thank you for your work! @jsosulski where are the sanity checks (code and plots)? |
|
|
||
| from moabb.datasets import download as dl | ||
| from moabb.datasets.base import BaseDataset | ||
|
|
||
|
|
||
| BI2013a_URL = "https://zenodo.org/record/1494240/files/" | ||
| BI2012a_URL = "https://zenodo.org/record/2649069/files/" | ||
| BI2013a_URL = "https://zenodo.org/record/2669187/files/" |
There was a problem hiding this comment.
I am currently running into page not found issues on zenodo (I think thats why CI is currently failing as well) so I cant check, but: is there a difference in bi2013a between the old and the new URLs? If so we probably need to bump minor version.
There was a problem hiding this comment.
The files are available on https://zenodo.org/record/1494240/ and https://zenodo.org/record/2649069/
The old link was v2, the new link is v7. The difference is the storage format: it was gdf and it is now csv + mat. The data are the same. I will make a version bump for to include these datasets.
| chtypes = ["eeg"] * 17 + ["stim"] | ||
| X = loadmat(file_path)[condition].T | ||
| S = X[1:18, :] | ||
| stim = (X[18, :] + X[19, :])[None, :] |
There was a problem hiding this comment.
Probably not necessary for this first merge to make it work with current binary P300, however if there is stimulus information available, we could keep it to be able to classify the letter / stimulus x,y location in the Braininvaders case.
There was a problem hiding this comment.
I agree, this is a good thing that the letter information are available in the data. We could use this for P300-speller classification.
There ebrains.eu but it seems complex to host data. OSF might be a better choice: https://help.osf.io/article/386-project-storage, there is up to 50 Gb of data for public project it seems |
|
@toncho11 the code is here. Note that this script will download ALL available P300 datasets in MOABB and create the plots which takes a) very long and b) a lot of disk space. Currently I hosted the plots for all P300 datasets in MOABB here: http://public.jan-sosulski.de/moabb_sanity/master.html |
Yes, if we have the corresponding paradigms, that is why I could not add them right away. |
|
I just tried out to load the dry EEG data from your branch @sylvchev and noticed that the data needs to be scaled by 1-e6, as it is stored in uV and mne expects V. Probably worth checking how the other paradigms store the data. |
|
Hi, |
|
I just tested |
Yes, I have seen that, but in Pedro's repo the data are not formatted like the other. So I'm leaving it for another PR. |
|
thanks for your remark @jsosulski, several datasets were not stored in the correct format. I checked all dataset and scaled down with uncorrected values. |
Adding new ERP datasets from GIPSA-lab.
The datasets were almost supported, as @plcrodrigues wrote repos based on MOABB. Currently, they are not up to date and require an old MNE version. This is a blocking issue in pyriemann-qiskit.
@jsosulski All datasets have been through a sanity check: you could see that @plcrodrigues have plotted evoked potentials and made basic classification tests.
I added all P300 datasets but py.ALPHA.EEG.2017-GIPSA (only alpha waves), py.PHMDML.EEG.2017-GIPSA (music listening), py.VR.EEG.2018-GIPSA (recording during VR session) that are not P300 datasets.