EXTREMUM_web/base/notes.txt
2024-11-15 20:50:36 +02:00

136 lines
4.6 KiB
Plaintext

Tasks:
1) Glacier training
1.1) preprocessing disable
1.2) parameterization
1.2.1) add a radio button (custom, default) for custom the user would be able to evaluate the parameters
while for default he would have access to presets for glacier training following zhen dong's presets
1.3) file management after training
2) Tabular training
3) Uploaded files exploring
----------
14/10/2024
----------
1) glacier downloads the respective dataset -> dataset is given as a parameter (done)
2) IntervalImportance error (when training a 1dCNN model and want to save the importance for future use in plots).
gc_latentcf_search_1dcnn_function.py
Commented out for now...
3) Counterfactual.html looks really bad should be improved (a bit)
4) after 1dCnn train there is an error with classification_report
figure out what to plot after train of 1dCnn to improve
charts.html content
5) add learning rate as parameter to glacier_compute_counterfactuals.py
6) check for more datasets
7) positive negative labels in pipeline json to be part of the dataset information and not of glacier model (done)
8) load experiments text is still json format, should look prettier
9) add ford-a dataset (done)
10) hard coded positive and negative label values during selection of timeseries dataset (done)
11) If no experiments to load case (done)
Counterfactuals.html
1) new experiment, load experiment buttons
1.1) spacing
1.2) aHave buttons active all the time unless another is clicked
2) New experiment run
2.1) enable loaded experiments
3) Load experiments
3.1) when changing timeseries page scrolls up and it is weird
----------
21/10/2024
----------
1) counterfactuals for tabular-> features to vary could have a drop down and a drop up to hide when needed (done)
2) Original Data table does not take up all the card space
3) Tsne should be appended to the DOM imidiately since it is common plot for all the pre trained models
----------
23/10/2024
----------
1) replace dice ml
2) fix counterfactuals.html for time series (prettyfie)
25/10/2024
1) description of extremum
2) functionalitites->explain
3) pages depend on (enabled or disabled) the actions the user took (train a model enables pre-trained.html and counterfactuals.html
otherwise disables)
4) explain workflow
5) home page seperate from dataset selection
6) main buttons
1) tabular
2) timeseries
3) upload -> information
7) Train.html dataframe summary text
- Functionality:
1) after training done: save model as (give a name) with all the cases
Give message: "Model trained succesfully"
"See the results: (click)" -> demonstrate what is displayed at charts.html but in a different page
Goal is to distinguish between pre trained models and newly trained models
save the model or not
1) Save (name, message etc, prompt train another model or counterfactuals)
2) or Retrain (prompt back to the train.html)
2) delete available pre trained models if needed with the use of a button
3) check if coutnerfactual has similar class
- give information about the classifiers
- merge both tables together
feature|original value|counterfactual value
Focus on the web version:
1) no upload
2) no train[]
----------
29/10/2024
----------
1) choose between uploaded files (done)
2) add seperate home page that prompts to dataset selection (done)
----------
3/11/2024
----------
1) add message if cf on tabular takes too much time (done in backend should fetch in the front)
2) then should add a second tsne after the comoutation of the counterfactuals that would be further
down below in the same page where the user can scroll with the use of a button (done)
----------
3/11/2024
----------
1) finished the layout for the tabular data
2) need to fix the layout for the timeseries too (done)
3) need to train glacier wildboar for all the timeseries datasets
----------
6/11/2024
----------
1) fix the layouts (done)
2) should add fade ins to all the pages (only in counterfactuals.html and only for timeseries datasets now)
3) should add proper info modal windows for the respective content (all the pages, only for timeseries datasets now)
----------
6/11/2024
----------
1) add proper fade in to all the pages
2) solve some frontend bugs
3) try docker
4) apply styling options to opensource version
Todo:
1) apply error message if cf takes too long to compute
2) delete uploaded files
web extremum:
accuracy in classification report needs an individual cell
hide error after successful cf
move label table
upload
glacier wildboar paper