5teffen / advice Goto Github PK

REMOVE:

• Old Global Explanation
• Intro Pages, Instructions, etc.
• Key Features(?)
• Remove tSNE, PCA, Directionality
• Style class selectors
• Simple toolbar

Accomodate custom input.

Retain the same layout but reduce opacity and remove interactive features for the "hidden" filter set. Done from the JavaScript side.

Due to the backend logic of feature selector the removal of a single filter requires additional information. Fix later.

• Default option prompting to choose feature not displaying
• dont allow creation of multiple filters for the same feature

Ensure distance function works for any data set. Include PCA.

Input: Data Set
Output: CSV projection file

It has a bug on the education data (will ask if it can be shared)

Currently the arrow sizes are fixed in D3 and thus do not accurately reflect the results.

Check why the histogram is shifted. Values shown to be around 8-9 but bins are below

Currently floats are converted to integers in the visualisation process. Information is lost. Possible solution might be to move the decimal point in the backend logic.

Clicking on a single sample in the projection view should either prompt a hover local explanation or actually redirect to local explanation URL.

Bug with the left filter set and some categoricals. When clicking on a bin sometimes points outside this bin are selected as well.

test on fico

Current range system does not work optimally for categorical features. Example: gender variable for heart disease dataset.

When selecting a single sample, mean median and arrow starting point should match

It would be useful to add a completion percentage bar for preprocessing part.
Something like this: | ######### | 60%
For large data sets it help see how long does it have left.

When clicking on density the graph is not drawn (need to catch the case when densityData = "no")

Mimic logic from confusion matrix.

Attempt to fix density such that it appears as a smooth gradient. Might require consulting with someone with more experience.

Make histogram align with the single value as opposed to a range.

Median doesnt make sense for categorical. Possibly show the actual string value for the categorical.

Organize all components for better viewing.

Ensure the Bokeh plot can handle any data.

Update the median values when applying filters. Currently appearing static.

The histogram should update based on filters

How should the comparative density be normalized.

Possible solution: The two densities will be generated simultaneously with knowledge of each other. This may cause problems to the workflow.

Lets add the AG-Grid. See this: https://www.ag-grid.com/javascript-getting-started/

It is really simple and will make things look really clean imo.

Allow for more than just 3 sets of filters.

Arrows do not have values. Do not match results.

Implement logic to remove the filters.

A fourth button to remove them. Allows for better customisation.

Have the histograms partially filled based on the selected points.

Create Dropdown. Call the D3 function etc.

when a new sorting order is done, dont reorder but notify of the option to change it

• sort by divergence
• sort by number of counter factuals

I think the show_projection function hasn't been properly updated and its resulting in an error.

Have continuous and discrete points separately. Maybe also juxtapose them with continuous density and histograms.

This required to accommodate multiple filter sets in parallel.

Front End (Oscar)

• Drop down option to select dataset/ reload page.

Back End (Steffen)

• Functions that allow initialisation of all datasets.
• Ensure correct data preprocessing is done.

currently uses an outdated function

Projection Graph fix

The goal of the whole tool is to identify the boundary.

BUG FIX!

Replace the continuous distribution with a histogram

Potentially a new library to create the CF files.

Accommodate the feature selection in the filter summary.