Skip to content

Latest commit

 

History

History
79 lines (56 loc) · 4.9 KB

TODO.md

File metadata and controls

79 lines (56 loc) · 4.9 KB

TODO

  • Port equation menu from czi-prosemirror?
  • Build mosaic plot
  • Support LaTeX Equation in exported PDFs
  • Ability to select multiple smoothing values / change them in real-time (similarly, for histogram ability to choose binwidth)
  • Ability to set font family for reports
  • "Undo" button for plots (investigate Plotly's capabilities in this regard)
  • Include "Text to TeX" equation parser
  • Add React component for toggling between Text source / intermediate TeX representation / rendered equation.
  • Add option to display a line grid instead of a white page as sketchpad background
  • Add save reminders in data explorer
  • Log actions in instructor bar
  • Consider adding vertical gridlines to plots
  • Automatic email reminders for uncompleted labs
  • Allow expressions in number question input fields
  • Mobile responsiveness
  • Remove any R dependencies for the mosaic plot
  • Display additional information when hovering over bars of barcharts (percentage, description, other variables?)

Completion for components in lessons

  • Completion Status
    • Lesson ID
    • Component ID
    • Viewed / Interacted / Completed

cmetric: [name, all|exclusions|inclusions, rule, time-filter, type]

type weigthing vector

type=#|missing type=boolean

rules: binary proportion (0/1), binary trimmed proportion, thresholded proportion, average, weighted average, average with exclusions, median, ...

Add commas when formatting numbers in model outputs (e.g., 1230091131 -> 1,230,091,131)

add weight property to answer choices (non-negative, at least one correct answer weight is positive)

TODOs:

  • Reshape the session data (add id, component to session data) id, component type, value separate (as object, no JSON.stringify in components necessary)

  • Logging of actions

    • Completeness

  • focus hook to compute time in component higher order component model for injecting component type etc.

  • migration: keep old endpoints for backwards compatibility, but reshape database session data and add new endpoints

Data Explorer:

  • Under "Models," there should be a option for "Split Data" that opens a new window that allows you to specify the proportion of data to be put in the training set, with maybe a second data entry for setting the random number seed (so we can have reproducible results). A button at the bottom would say "Split" and when this is clicked, a new column is generated saying "Train" and "Test." If someone does this multiple times, this one column gets overwritten...no data_split1, data_split2, etc.

  • All models should have a click box that says "Apply to Training Data Only" (or something like that); all models should generate a column of predicted output. (This would be true even if we don't click on the button...there should be a generated column that includes predictions for all training (i.e., all) data.)

  • Regression models should have test-set MSE output if the training data button is clicked, otherwise the all-data MSE should be output. Classification models should have an additional field for mapping predictions (probabilities for class 1) to actual classifications. So a classification model should lead to the generation of a predicted probability column and a predicted class column.

  • Allow for generation of confusion matrices (which should be a separate thing and which should provide multiple outputs...the matrix, the MCR, etc.); and ROC curves (which need the probabilities for the test set, and which should appear as a curve with associated metrics: the AUC, and which thresholds minimize the MCR and Youden's J).

  • Variable Selection:

    • First, there should be a button that allows one to simply include all variables that are not the response variable in the predictor field, so that we don't need to go through and click on every single predictor variable we want.

    • If the number of selected predictor variables is < 25 for a multiple linear regression model, or < 10 for a multiple logistic regression model, we should be able to perform best-subset selection. The input to BSS should include a click-box that asks for the desired metric to minimize: AIC or BIC. The output from BSS should be a plot that shows the AIC/BIC score (possibly in log-space), and a list of the variables included in the best model.

    • forward- and backward-stepwise selection.

    • For linear and logistic regression, perhaps we can include a click-button (default off) for generating variance-inflation factors for the model.

  • Interface:

    • simple linear regression refers to "explanatory variable" and all other models refer to "predictor(s)." We should change this to "predictor".

    • logistic regression should be grouped with the inferential models: simple linear regression / multiple linear regression / logistic regression / lasso

    • As we add predictors to an RF model, the default value for # of predictors to check at each split should change to floor(sqrt(# pred)) rather than staying at 1