Time_Majority_Voting

Time Majority Voting is a new algorithm that can better predict EEG datasets.

Time Majority Voting Algorithm for EEGEyeNet

Note: the first four bullet points will guide you to download the dataset and reformat it as a CSV file with the format we planned our code on. If you face difficulties with these steps or want to skip them, contact the authors, and they will send you the CSV file directly to start from the fifth bullet point.

1. To get the data into the expected format for our code to run properly, first, you should go to this link https://github.com/ardkastrati/EEGEyeNet, download this GitHub repository on your device, and add the file (process_eegeye.py), which you can find in our code folder, to the folder you just downloaded from the GitHub. Add an empty folder named data to the same folder.

2. Next, go to this link: https://osf.io/ktv7m/. There, on the left, you will find two folders, the first of which is named "Dropbox: EEGEyeNet." If this first folder has a (+) sign next to it, click on the (+) sign. Otherwise, go to the next step directly.

3. Now, you should see four folders pop up. Open the folder named Prepared (Click the + sign next to the folder named "prepared"). When you do that, you will see many files showing up. Scroll down till you see four files that start with "LR_task_with", download the last one of them on your device, and then move it to the data folder you just created above.

4. Run the (process_eegeye.py). After the code finishes running, you should see a new CSV file created in the same directory named EEGEyeNet-data.csv.

5. Now, put the CSV file, which you just generated, in our coding directory, where the [README.md] and all other TMV files exist. Then, Run TMV_EEGEyeNet.ipynb to execute the TMV algorithm on the EEGEyeNet dataset. You can also decide to run which subjects by adjusting the variables "start_subject" and "end_subject." You can uncomment and comment out the corresponding names and classifiers in the second block to determine what classifiers you want to run.

6. Once the code finishes running, you should go to the (EEGEyeNet_analysis) folder and check the (output) folder inside. In the (output) folder, you will see a JSON file that includes all accuracies and standard deviations for all algorithms applied to all subjects. (You don't need to do anything with the file, take a look if you are curious)

7. Go back to the (EEGEyeNet_analysis) folder, and run [EEGEyeNet_analysis.py] to generate the two plots and the 3 CSV files that document the results in our paper. All plots and files will be generated in the same directory. So, once generated, you should access them easily.

Side Note: There are 369 subjects in total. Run time varies based on the number of subjects you are going to run.

Time Majority Voting Algorithm for BCI Competition

1. First, go to the BCI Competition Dataset IV 2a Website. On this page, scroll down to the Download of data sets section and add your name and email. You will then get an email with a link, user, and password. If you don't see it, check the spam folder. The title of the email should be BCI Competition IV - data set request.

2. Open the link in this email and use the user and password to open the site. Go to the section named Data Sets 2 and click on the download link next to the first bullet point (Data sets 2a). After you click the link, a zip file including matlab data files should be downloaded on your device. It's worth mentioning that Google may block you from downloading the file; so, you may have to change your google premissions or use another browser.

3. Once you get the zip file downloaed, unzip it, copy all the files and go to the folder called "BCI-extraction" which you wil found in our code folder. In that folder, open the folder "dataset" and paste all the files you had there.

4. Then, run feature_extraction.py. This will generate 18 .npy files (two for each subject) in the "BCI-extraction" directory.

5. After you see all 18 .npy files in your "BCI-extraction" directory, run preprocess_bci.py. This will generate another 18 .csv files (one train.csv and one test.csv for each subject) in the "BCI-extraction" directory.

6. Then, create a folder called "BCI-Data" in the same directory as your TMV_BCI_Competition.py file and move the 18 train and test .csv files mentioned above into this folder. You should put this folder in the same directory as your TMV_BCI_Competition.py file. Then, you should create a new folder called "output" in the same directory, and inside this "output" folder, create another empty folder named "BCI-Data".

7. Run TMV_BCI_Competition.py to execute the TMV algorithm on the BCI competition dataset. You can decide to run which subjects by adjusting the variable "total_subjects" on line 85 and the starting position of the for loop on line 94. Currently, the code runs subjects 1 to 9. You can uncomment and comment out the corresponding names and classifiers in the two lists in lines 26 to 51 to determine what classifiers you want to run. Currently, the code runs LDA, Nearest Neighbors, Random Forest, Linear SVM, and Decision Tree, and then performs TMV on the two best algorithms.

8. If you want to change the name of your dataset folder, you should change the variable "data_source" (line 23) to the new name. What's more, note that you also need to change the name of the folder you created inside the output folder, where the results will be stored, to the new name.

9. The output/BCI-Data contains example results for all nine subjects' training datasets.

Time Majority Voting Algorithm for TCR and RWT dataset

1. First, upzip the data folder and put the data folder in the same directory as the TMV_TCR_RWT.ipynb file.

2. You should create a folder called Time_Majority_results and put the Time_Majority_results in the same directory as the TMV_TCR_RWT.ipynb file. “Time_Majority_results” will store all the outputs from the TMV_TCR_RWT.ipynb. Then, you should create two new folders called "tcr" and “rwt” in the folder Time_Majority_results. Then, create a new folder called “TMV” inside each of the “tcr” and the “rwt” folders.

3. In the second block, change the “data_source” to the name of the datasets you want to experiment on (the default is “tcr”; you can change it to “rwt” to run the RWT dataset).

4. You can also modify the variables “subject_id_start” and the “subject_id_end” to determine which subject to start and end running the TMV. The minimum value for “subject_id_start” is 1, and the maximum value for “subject_id_end” is 17 for the TCR dataset and 14 for the RWT dataset. The results will be in the Time_Majority_results/tcr or Time_Majority_results/rwt depending on the dataset you run on.

5. Next, you can adjust “subject_id_search,” representing the subject id you want to generate the TMV results. This will generate the visualization of the subject’s task 1’s graph for all six sessions and the heatmap of the TMV prediction. These graphs will be generated inside the Time_Majority_results/tcr/TMV or Time_Majority_results/rwt/TMV depending on the dataset you run on.

Side node: It took me around 10 - 20 minutes to finish running Time_Majority_results/rwt/TMV. Running time varies depending on different computers.