h2o changes data types?

[Gregor Lauth]

Hello everyone,

I am trying to run a simple prediction model and save the result in the original dataframe. When first running the regression, it works, yet after accessing the h2o and predicting the class probabilities, I instead receive a "no observations r(2000)". A quick look tells me that the variables still exist and that a probability column was added. The code itself is only a slight modification of the example code.

----
h2o clear
use "$temp\Pre_reg", clear
gen d_time = (time_tuli > 90)
label define dtime 0 "Short Stay" 1 "Long Stay"
label values d_time dtime

reg age pre_20

_h2oframe put, into(auto)
_h2oframe change auto

_h2oframe split auto, into(train valid) split(0.8 0.2) rseed(19)
_h2oframe change train

global predictors sexe age
h2oml rfbinclass d_time $predictors, ///
cv(2) h2orseed(123) ///
ntrees(100) maxdepth(6) ///
balanceclasses minobsleaf(5)

_h2oframe change auto
h2omlpredict foreignhat_tpr, pr
clear
_h2oframe get auto

reg age pre_20
----

Thanks in advance for your help!

Best,
Gregor

[Aram Dallakyan]

Hi Gregor Lauth,
Is either of the variables age or pre_20 of the H2O data type enum? You can check the data types by using the command
. _h2oframe describe
For definitions of data types in H2O, please see https://www.stata.com/h2o/h2o19/h2of...ame-intro-what
If this is the case, the error after regress is likely due to Stata having difficulty interpreting the H2O enum type and converting it to a string. You can use the Stata command encode to convert this variable back to a numeric.

I hope this helps.