CODEWITHSUNDEEP
rploth2orocauc
pfohl
pfohl

Reputation: 21

How to plot the AUCPR curve from h2o model in R?

I did some kind of data analitycs with the h2o.ai platform in R and I want to receive the AUCPR curve for the model / the prediction.

I already tried to use "PRROC" package, but it sees either not working or to slow for my dataset (1.4 Million instances). For the other available packages I don't really know how i can extract the data from h2o model.

pr <- h2o.predict(V_PUF_AGG1_NPI_ALLEX_BINAR.drf.tt.standard, data.test)

gives me a prediction matrix (which i can use for further proceeding?):

  predict        p1           p2
1       1 0.9999427 5.731940e-05
2       1 0.9999606 3.939748e-05
3       1 0.9999744 2.556443e-05
4       1 0.9999659 3.413081e-05
5       1 0.9999606 3.939748e-05
6       1 0.9999545 4.554749e-05

[987141 rows x 3 columns]

So I'm searching for a quick solution to plot AUCPR curve.

It's easy to get the ROC curve, but there is no way to get an AUCPR curve directly from h2o:

plot(h2o.performance(V_PUF_AGG1_NPI_ALLEX_BINAR.drf.tt.standard, valid=T), type='roc')

Thanks - John

Upvotes: 1

Views: 1113

Answers (1)

Peer Christensen
Peer Christensen

Reputation: 104

You can plot the PRROC of e.g. the test set in the following way, where modis your H2O model object and test_hfis your test set.

library(tidyverse)

perf <- h2o.performance(mod,test_hf)

metrics <- as.data.frame(h2o.metric(perf))

metrics %>%
  ggplot(aes(recall,precision)) + 
  geom_line() +
  theme_minimal()

Upvotes: 4

Related Questions