# MCmp project notes (sktime-neuro)
###### tags: `MSc projects 2021`

Students: Emilia Rose, Aiden Ruskbrooke and Freyja Harvey

possibilities
1. EEG classification
With an intern we launched a companion package
https://github.com/sktime/sktime-neuro
this is in a very early stage of development, and contributing to this would be really helpful. There is a lot still to do, and its less formal than sktime proper
2. Deep learning with time series
https://github.com/sktime/sktime-dl
if this interests you
3. Time series clustering
This is a fairly new module in sktime, there is lots of interesting stuff to do.
4. Time series classification
This is our main area, plenty to do here too
5. Application focus: in addition to EEG classification, you could look at any area of time series classification that appeals to you, such as human activity recognition

#### 14/10/21. First meeting

agreed to do the sktime-neuro project. Need to set up objectives, specify tasks and assign
Action: All to install sktime-neuro and run through Svea's workbook

Tony tasks
set up project on github
to start gathering papers
arrange meeting with Saber Sami
arrange meeting with psy guys

Possible first objectives
1. Collect more EEG classification data sets
2. Make it so we can use sktime-dl
3. Improve the continuous integration
4. Run benchmarking experiments with hive cote 2 on existing problems.
5.

### 19/10/21

Aim: Compare modern TSC algorithms against traditional methods for EEG classification

Objectives:
1. Background research
What are EEG/MEG?
Who uses them for classification and how do they do it
What is state of the art in TSC

2. Datasets.
find and collate here
https://github.com/uea-machine-learning/tsml_repo/issues

3. Code.
all in stime-neuro
sktime-dl for algorithms
MNE (the EEG processor)
BIDS for data representation

4. Experiments.
Evaluate HIVE-COTE2.0, ROCKET and InceptionTime
Evaluate others
Build evaluation framework
run on the cluster

Assessment:
organise a seminar/practicals 15%
final report 60%
demo/presentation 25%

MOSCOW:
Must have:
background section
Some new datasets
sktime-neuro experiments with tsc algos and deep learning
a comparative study

Could have:
new data and new problems (find the bike!)
new algorithms/pipelines used in the literature
a workshop paper

Wont have:
interface
everything else

## Meet 26/10/2021

Quick catch up between Jason, Emilia, Aiden and Freyja (Tony away for half-term and Aaron teaching!). Progress being made on the background; Freyja focusing on getting started with sktime while Emilia and Aiden are focusing more on EEG side of the background as they used sktime previously in their final year projects.

## Meeting 5/11/21

1. Freyja looking at open source and attempting the type hint issue
2. Aiden working through EEG and sktime-neuro
3.

## meeting 7/12/21

1. Freyja not had much time to work on this
2. Aiden has preprocessed two datasets, will post on repo github

## Peer Meeting 2021-12-16

1. Freyja has done nothing. Will start work again in next couple days, as deadlines are done.
2. Same for Emilia. Work on pip integration now.
3. Aiden, running experiments, to collect data.

* Collect and test further EEG data on the library
* Adapt and clean up existing code to make it more readable, maintainable, and usable.
* Review and resolve existing issues (integrate SKTime DL, Pip)
* Additions that might be necessary. Tune algorithms so they run better on EEG data.
* Comparison of sktime-neuro against other systems that are already existant.
* EEG Data transformations and processing. Impact of different filtering methods.
* STRECH: Labeling tools that take in existing formats, allow for tagging, and produce whatever we standardise on.

## Peer Meeting 2022-02-02
Emilia typing for this meeting, sorry for spelling
Begin with discussion on current difficulties, Emilia and Aiden notes issues with the required versions of pandas and numpy for sktime, possibly should be sorted out by the sktime team in the future?
Emilia -
* Work on conversion of data formats, e.g. conversion of .edf files to .ts :)

Freyja -
* Autodownloaders for datasets
* added pipenv tooling.
* Type hints to a bunch of stuff

Aiden -
* Running experiments

What we're working on now
* Starting the report!
* Aiden going to run experients with and without preprocessing, to justify sktime-neuro
* Overall report plan will be drafted up, we're planning at the moment to try have the majority of it experimental analysis
* Noticed some possible small bugs in sktime, Aiden raising this as an issue

Report structure
* Intro
* BG
* SKtime
* FLOSS Software
* Classifiers that we used and why
* EEG Processing (Data conversion, how best to optimise EEG data for sktime)
* Results
* Compare our results to existing research papers

## Supervisor meeting 3/2/22

Present: Tony

Chair/note taker: Tony

Agenda:
1. Review notes and actions from last meeting
2. Summary of current progress
3. Overall project plan
4. Plan for coming week(s)
5. Date next meeting

Notes:
1. Review notes and actions deferred until next meeting.
2. Aiden running experiments on the cluster
3. Frejya working on auto downloaders, on BCI datasets
4. Emilia working on converters to ts files.

Actions:

1. Aiden to raise issue on tde if it persists.
2. Aiden to raise new datasets issues on the repo
3. Frejya to talk with Tony about putting raw data for auto download on tsc.com
4. All: start report on overleaf, give link for it next week, review structure then
5. Set provisional weekly time weds 14:00
6. Frejya to make type hint PR on sktime proper.

## Supervisor meeting 09/02/2022

Notetaker: Freyja

Agenda:
1. Review notes of last meeting
2. Report made at https://www.overleaf.com/7411859938kdntxnkzvfgr

Actions:
1. Background into EEG classification
2. Evaluation of MNE and it's utility
3. Pull together bits of early report writing
4. Collate new papers if existant

## Peer meeting 03-02-2022

Notetaker: Aiden

Agenda:
1. Review current work done on report
2. Evaluate state of sktime-neuro development
3. Discuss further work on report and current goals

Freyja:
* Looking at fixing code, such as the examples
* Working on auto-downloaders
* Type hints

Aiden:
* Write up for Riemannian classification
* Will look into implementation into sktime-neuro

Emilia:
* Working on data transformers

Actions:
1. Continue working on report
2. Look into implementing additonal preprocessing methods in sktime-neuro - PREP pipeline
3. Look into adapting classifiers for EEG data
4. Continue experiments on EEG data

## Peer meeting 2022-04-27 (Freyja Chairing)

Freyja:
* Maintainance
* Multivariate detrending
* With aiden, look at column ensemble

Aiden:
* Collected data for classifiers that do exist

Emilia:
* more transformers for MNE data

Agenda:
* Look into making datasets smaller via onset analysis
* Full integration of MNE -> PrEP -> Sktime -> results
* More literature analysis, look for comonalities in papers.
* Make spreadsheet of literature analysis https://dblp.org/search?q=EEG+classification
* Impact of heterogenous column ensembles vs non-heterogenous

## Peer meeting 2022-05-17 (Aiden Chairing)

Report notes:
* Finish writing reports
* Suprising results, processing reduced accuracy
* Create document for authorship
* Write up conclusion and future work
* Check earlier sections of report