yardstick
yardstick copied to clipboard
Tidy methods for measuring model performance
To improve accessibility, we can add `alt` labels to the output of the `autoplot` methods. Here is a very rough idea of a possibility: ``` r library(tidymodels) #> Registered S3...
Currently in {probably} we have ad-hoc support for a distance metric implemented as: ``` distance = (1 - sens) ^ 2 + (1 - spec) ^ 2 ``` It would...
We need to be able to compute these out-of-sample statistics in `yardstick`. We could have `weibull_log_loss()`, `lognormal_log_loss()`, and so on. Alternatively, we could also use `censored_log_loss(distribution = "family name")` in...
See [this reference by Powers](https://scholar.google.com/scholar?hl=en&as_sdt=0%2C5&q=Powers++Informedness+and+markedness&btnG=).
The original code for the `poisson_log_loss` was taken from {MLmetrics} and it works well overall, but has some edge cases when Inf is returned eg `dpois(ypred, 0) = 0`, `dpois(ypred,...
Similar to #488: Is there an easy way to get metrics like accuracy and precision per threshold (0, 0.01, ..., 1.00)? Otherwise I'll write functions that do compete metrics rowwise...
https://yardstick.tidymodels.org/reference/sens.html For the table, I think you should use TP, FP, FN, TN instead of A, B, C, D to make things clearer like wikipedia does. https://en.wikipedia.org/wiki/Confusion_matrix Also the formulas...
`classification_cost()` takes probabilities and computes cost from them. Since we are going to be doing more with setting thresholds, we should probably have a version that takes the hard class...
It would be nice to have a flag to compute proportions out of total instead of counts. `conf_mat` object is not a matrix, so I can't do something like `conf_mat(data,...