# Timeline
###### tags: `organisation`

gantt
    dateFormat  YYYY-MM-DD
    title       Timeline
    excludes    weekends
    axisFormat  %a-%d
    %% (`excludes` accepts specific dates in YYYY-MM-DD format, days of the week ("sunday") or "weekends", but not the word "weekdays".)

    
    section Data Visualisation
    Clustering(Sa, Ch):active, a1, 2021-12-13, 24h
    PCA on MorphoCut features(Sa): done, a2, 2021-12-10, 24h
    PCA on CNN features(Ch):active, a3, 2021-12-13, 24h
    
    section Data/Feature Exploration
    Data augmentation for CNN (Ro, Je):active, b1, 2021-12-13, 24h
    Feature extraction for Baseline Random Forest (Mi, Da, Ro): done, b2, 2021-12-08, 24h
    
    section Model Experimentation
    Set up baseline Random Forest (Mi, Da, Ro):done, c1, 2021-12-07, 48h
    Set up baseline CNN(Me, Ro, Je, At):done, c2, 2021-12-07, 48h
    Running CNN experiments (At): active, c3, 2021-12-13, 24h
    CNN VGG16 implementation(Me): active, c4, 2021-12-13, 24h
    Comparison of feature importance for each of the models (RF vs CNN): active, c5, 2021-12-13, 48h
    
    section Documentation/Write-up
    PCA on MorphoCut features(Sa): active, d1, 2021-12-13, 24h
    MorphoCut feature extraction(Da): active, d2, 2021-12-13, 24h
    Random Forests(Mi): active, d3, 2021-12-13, 24h
    
    section Code Implementation
    incorporate pre-trained CNN in Scivision(At, Al): active, e1, 2021-12-13, 24h
    incorporate pre-trained RF in Scivision(Ro, Mi, Da, Al): active, e2, 2021-12-13, 24h
    
    section Deadlines
    DSG catch-up session: crit, f1, 2021-12-10, 12h
    Final presentation session: crit, f2, 2021-12-16, 12h
    Report submission: crit, f3, 2021-12-17, 12h
    
    

To do (week 2):

[x] GitHub refactor to one branch, AC
[x] Copepod / non copepod without detirtus Random forest with new test / train
[x] Species without detritus Random forest new test / train
[] Report writing for Random Forest and results
[] Report writing feature extraction using morphocut
[] Report writing background on plankton classification
[] PCA on the morphocut data RB, DS
[] Explainable CNN
[] Explore https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.StratifiedKFold.html MC
[] Explore https://vitalflux.com/k-fold-cross-validation-python-example/, MC
[] Copepod / non copepod without detirtus CNN with new test / train
[] Species without detritus CNN new test / train

- Let's all use the new test/train split that Rob made, both for Random Forest and CNN
- CNN will also apply augmentation to the train dataset

*Classification Evaluation Methods:
Accuracy
Precision/Recall
F1 Score
Confusion Matrix
ROC/AUC Curve https://scikit-learn.org/stable/modules/generated/sklearn.metrics.roc_auc_score.html

*Evaluate Model
RF => R2 RMSE
CNNs => Loss vs Accuracy

CNN team update:
[] Attiq: CNN experiments (https://hackmd.io/SQGlE6fQSh-2_5BLfRJUDA) - wihtin the next 24h
[] Rohit, Jennifer: data augmentation - updates within the next 24h
[] Meghna: VGG