ihgumilar / hyperscanning2-redesign Goto Github PK

Experiment of hyperscanning2-redesign is adding questionnaire inside VR and also improved the UNITY that is used for the experiment.

Python 99.64% Makefile 0.16% Batchfile 0.20%

hyperscanning2-redesign's People

Contributors

Stargazers

Watchers

Forkers

sheenawang12

hyperscanning2-redesign's Issues

Notes on saved list (*pkl file) that contains inter-brain synchrony scores & indices of deleted epochs

Interbrain synchrony scores

Save the into pkl file
with open('list_circular_correlation_scores_all_pairs_direct_pre_no_filter.pkl', 'wb') as handle:
pickle.dump(list_circular_correlation_direct_pre_no_filter_all, handle,
protocol=pickle.HIGHEST_PROTOCOL)

NOTE : The structure of list that has been saved (pkl file) is each pair will have 4 lists, which has the following order
* Theta, Alpha, Beta, and Gamma. So for example, the first 4 lists are inter-brain synchrony scores that are calculated from subject 1 and 2.
* then move the 2nd four lists which are calculated from subject 3 and 4.

So, if there are 15 pairs, then there will be (15 x 4) 60 lists in each pkl file

Delete data that is in the same part with EEG data, which is deleted during pre-processing. It is available in the previous script by getting the index of deleted epoch of EEG

Error

It seems that it is related to indices of deleted epochs EEG

Count average of actual score per significant pair of electrode (out of all subjects)

Please refer to #32 to get started or develop from there

Start working here 0375c4d
There is a cell that has already some ideas

ToDo

Count average significant actual score of specific connections out of all pairs (from dictionary), which have key (out of all participants). Use this code to start #32

Create branch of EEG analysis

EEG-analysis
Done

Look at script to analyze EEG data using hypyp

05016c7 Has been adjusted to the current data. But there is still an error

Mixed results

NOTE :
Before we run analysis,

we take average score of questionnaire for pre and post
then substract post and pre

6ef7b5b

Correlation between SPGQ total & Direct

Significant correlations in :

coherence - theta : (0.6189193814333278, p-val : 0.03189176226829951
coherence - beta : (0.564254914717613, p-val : 0.05599109700636175)
coherence - gamma : (0.6105006072150732, p-val : 0.034995765473106906)
PLV - gamma : (0.6538593554639102, p-val : 0.021091591502908406)

Good reason why we may want to choose PLV, does not require stationary of EEG = the stability of core characteristics of the time series

https://academic.oup.com/scan/article/16/1-2/72/5919711?login=false

Use 'ccorr': circular correlation coefficient (which is the same with the previous analysis)

Correlational analysis EEG and Questionnaire

Analysis :

Correlational analysis with EEG
- The number of connections and questionnaire score of SPGQ
- The number of connections and questionnaire score of subscales of SPGQ
- The number of connections and questionnaire score of Co-Presence

NOTE : Remember that the index starts from 0

total_sig_ccorr_theta_connections,
total_sig_ccorr_alpha_connections,
total_sig_ccorr_beta_connections,
total_sig_ccorr_gamma_connections,
total_sig_coh_theta_connections,
total_sig_coh_alpha_connections,
total_sig_coh_beta_connections,
total_sig_coh_gamma_connections,
total_sig_plv_theta_connections,
total_sig_plv_alpha_connections,
total_sig_plv_beta_connections,
total_sig_plv_gamma_connections

Count significant connections and give labels

60bff86 start from this commit. It is new file for counting significant connections and give labels
Count significant connections

Label which electrodes that are significantly correlated
List from the connections that have most connections to the least one

ToDo :

Get the actual score of PLV, ccor, or coh, if it is significant
Get the index of that actual score, if it is significant
Average actual score for PLV, ccor, or coh, out of all connections from all pairs
Adjust the variable names that are under Get labels of electrode pairs. Make it more meaningful and organized. So that it will be easy to read in the future (IMPORTANT git add EEG/analysis/statistical_analysis_inter-brain_connections.py )

Function to calculate how many percentage of looking and not looking

Do the following steps

- [ ] Call this function for every pair instead of all pairs that have been combined
The result is a list that contains whether look at or not throughout the experiment
Calculate the percentage of looking and not looking ,i.e. (number of looking / len(total_looking_not_looking)) * 100 for each pair (return value of the function)

Put FP1 & FP2 as reference channel both detecting eye blink (double check if new ICA label is able to detech eye blinking first)

Has been integrated into HyPyP

Change end for loop in combine_fif_files.py

Change end for loop in combine_fif_files.py here 9aa1e7a
This is the example

Do for

Experimental data

Even subjects (10 and onwards, eg. 10, 12, 14, etc..)
Odd subjects (10 and onwards, eg. 10, 12, 14, etc..)

Baseline data

Even subjects (10 and onwards, eg. 10, 12, 14, etc..)
Odd subjects (10 and onwards, eg. 10, 12, 14, etc..)

Analyze pre-processed data of eye tracker

1. Populate odd subjects and even subjects into one separate dataFrame for all conditions , i.e. averted_pre* , averted_post* , direct_pre*, direct_post*, natural_pre*, and natural_post*. It means for each condition, eg. averted_pre* , there will be two dataFrames, one for odd subjects and one for even subjects.
```
  - [x]   This has been turned into a function
```
2. Convert GazeDirections, which are in cartesian (x, y, z) to degree, see Functions to convert cartesian to polar degree, for both Right and Left eyes. Put the converted value (degree) into a new column (eg. eye right-left), by using this code. Inside it, a function gaze_direction_in_x_axis_degree(x, y) is resulting a degree that indicates a movement of the eye in xaxis (right or left). A function gaze_direction_in_y_axis_degree(y, z) is resulting a degree that indicates a movement of eye in yaxis (up or down).
- It has an issue when we run it with **averted_pre_even** subject. See the issue below here.
3. Check whether the values (degree) that are under the column of (eg. eye right-left) are within the range of fovea, see Function to check whether a degree within fovea (30 degrees in total) and use the function check_degree_within_fovea(gaze_direction) inside this script. If it is within the fovea, it will result it 1 otherwise 0. Save those new values (1 or 0) into a new colum, eg in_fovea.
4. Compare values of in_fovea for both x-axis and y-axis whether both of them are 1. For example, if both of them are 1 then give a label or value 1 in a new column, e.g. look_each_other, otherwise 0. Do this for both right and left eyes as well. Of course, for both odd and even subject dataframes.
5. Since the sampling rate is 125 / per second (125 rows for each second) , we need to count how many "1" as well as "0" in column look_each_other within a second (125 rows). We need a threshold Threshold of within a second to determine "looking" (conscious) or "not looking" (unconscious) #10. For example, if the threshold is 13, when we count "1" in column look_each_other within a second (125 rows), there are 13 of "1", then we need to create a new label 1 which indicates "really" or "scientifically" looking, otherwise 0. Put those new values (0 and 1) into a new column, eg. sig_looking
6. Do the step 2-4 for both odd and even subjects' dataFrames for each eye condition.
7. Check whether each pair looked at each other or not. Take a value from column in_fovea of odd and even subject, then check if both are giving 1 (must be 1 for both of them), which indicates look at each other in the same range of fovea. If so, give label 1 and put it into a new column again, eg. look_each_other, otherwise give label 0, that indicates that their eyes were not matched within a second.
8. Count a proportion of value 1 and 0 in column sig_looking. For example, if there are 60 "1" in the column sig_looking , then we can calculate the proportion : (60 / 1800) x 100 = 33 %. It means that out of 120 seconds (2 minutes), their eyes met only in 33%~40 seconds). Note 1800 = 120 (seconds) x 15 subjects (odd or even)

Process demographic data

56802ef Use this commit

Init code for demographic data

Put into dataframe
Plot values of each column
Only age that is numerical (float) data. Other than data is object type
Data has been preprocessed from original raw data (*.csv) so that it is easy to processed

Remove x-axis and y-axis visualization

339f7fe start from here. It is able to visualize

Module of Questionnaire

Under Questionnaire

Questionnaire class

def scoring_questionnnaire

Keep going down below until

def diff_score_questionnaire_pre_post
For this, we need to provide parameter for the name of column. Tweek a little bit the code from
Combine SPGQ Total score of subject1 & 2, etc..
NOTE : With subtraction of post and pre

def corr_eeg_questionnaire

print("Averted")
for i in range(len(diff_averted)):
    print(F"{i}, {pearsonr(diff_averted[i], substracted_averted_non_SPGQ_total)}")

Save data after being pre-processed and code for pre-processing

428592f In progress. Now it is still in Direct pre (experimental data)

Put new ICA label function and include only the brain data

Create branch of EEG pre-processing

EEG-pre-processing
Done

Separate between baseline and experimental data

Module of EEG

Pre-processing is here d95cf3e

Create new columns that calculate the total score of each sub-questionnaire for each questionnaire (there are two different questionnaires)

Edit the package of hypyp by adding new ICA label (new function in MNE-Python)

Notes : Code to count average significant actual score of specific connections out of all pairs (from dictionary), which have key

# Python3 code to demonstrate working of
# Convert list of dictionaries to Dictionary Value list
# Using loop
from collections import defaultdict
import numpy as np

# initializing lists
# list_temp = [{"Gfg" : 6},
# 			{"Gfg" : 8},
# 			{"Gfg" : 2},
# 			{"Gfg" : 12},
# 			{"Gfg" : 22}]

list_temp =[]
a = {"a" : 1}
aa = {"a" : 2}
b = {"b" : 1}
bb = {"b" : 3}

for i in range(4):
    list_temp.append(a)
    list_temp.append(aa)
    list_temp.append(b)
    list_temp.append(bb)

# printing original list
print("The original list : " + str(list_temp))

# using loop to get dictionaries
# defaultdict used to make default empty list
# for each key
res = defaultdict(list)
for sub in list_temp:
	for key in sub:
		res[key].append(sub[key])
	
# printing result
print("The extracted dictionary : " + str(dict(res)))

average_a = np.mean(res["a"])

print(f"Average score of a {average_a}")

Create branch of eye tracker pre-processing

eye-tracker-pre-processing
Done

Viz-Add title of figure of plot

fbb72dd it is done here

Fix EyeTracker data of Subject 16 (S16)

Done

Count average score of significant connections for each pair (combine all electrodes)

total_sig_ccorr_theta_connections, total_sig_ccorr_alpha_connections, total_sig_ccorr_beta_connections, total_sig_ccorr_gamma_connections,

total_sig_coh_theta_connections, total_sig_coh_alpha_connections, total_sig_coh_beta_connections, total_sig_coh_gamma_connections,

total_sig_plv_theta_connections, total_sig_plv_alpha_connections, total_sig_plv_beta_connections, total_sig_plv_gamma_connections

Get a previous script from repository of Hyperscanning2 and upload it into the repository

Uncomment the ICA in the hypyp package that is currently commented

Order of using code for questionnaire data

ANCOVA SPGQ & Co-Presence questionnaire

Calculation total score of each sub-scale of SPGQ
Calculation total score of SPGQ
Calculation total score of Co-Presence
Do ANCOVA for total score of SPGQ
Do ANCOVA for total score of Co-Presence

All the above stuff can be done via this code b0c996d

Next step is #33 which is not implemented yet.

Do that step #33 , when all actual data of EEG have been calculated completed by this, point no. 1

- Create new columns that calculate the total score of each sub-questionnaire for each questionnaire (there are two different questionnaires).

See the script.
Create a new template

Notes : To see deleted indices of EEG data

import pandas as pd
del_list = pd.read_pickle(r"/hpc/igum002/codes/Hyperscanning2-redesign/data/EEG/pre-processed_eeg_data/list_deleted_epoch_indices_direct_pre.pkl")

for i in range(len(del_list)):
    print(len(del_list[i]))

Change the path where the deleted epoch indices of EEG data are located

Notes: Label sequence for extract eeg file

This is needed as an input for separating baseline and experimental data from raw csv file (26 files of csv)
[1,2,1,2,1,2,3,4,3,4,3,4,5,6,5,6,5,6,7,8,7,8,7,8,9,10]

Count total significant connections for each eye condition (out of all subjects)

Combine all number of significant connections into one list

Loop

Get the length of significant connections for each frequency (theta, alpha, beta, and gamma).
Put into different list for each Pre / Post condition.
Put into a combination table like below so that we can use it for ANCOVA analysis

Run ANCOVA analysis

Notes : Running permutation in several screen of linux

** Activate environment
source /hpc/igum002/environments/hyperscanning2_redesign_new/bin/activate
** Go to folder that contains file to run
cd /hpc/igum002/codes/Hyperscanning2-redesign/EEG/analysis/permutations/

time python permut_file_name.py output:
for example, time python permut_averted_pre.py output:

Screen :

natural post
natural pre
direct post
direct pre
averted pre
averted post

Create new columns that calculate total score of each questionnaire (combine all sub-questionnaires)

Analysis :
1.1. Correlational analysis with EEG
b. Plot for total scores
c. Plot for each questionnaire or subsection

Calculate Repeated measure ANOVA or significant difference for Pre Vs Post (t-test) for each eye condition, i.e., averted, natural, and direct ( total score )

Actually we need Repeated measure ANOVA NOT T-TEST

856b6c6
Compare total score of SPGQ pre vs post

It also compares total score of CoPresence
ToDo :

Need to add standard deviation score for each t-test
T-test between subscale of SPGQ (pre vs post)
Correlate CoPresense and SPGQ
Correlate CoPresence and subscales of SPGQ

Pre-processing

Separate EEG between baseline & experimetal data using this code 2e2b86e
Combine pre and post data (for each baseline and experimental), for all eye conditions, using this code 432da5e

ToDo: Change loop from 16 to whatever length of files that are available)

Clean EEG data for both baseline and experimental data using this code a619b0b

ToDo : Update bad channels, in case the data has increased / updated

# TODO :Define bad channels for each participant

Assign bad channels for each eye condition