CoxPH: NPE instead of a params validation error

[exalate-issue-sync]

getting NPE when running Cox PH model

use data file:

[^dallas_shelter_survdata_FY18-20.csv.gz]

{code:r}library(data.table)
survdata = fread("dallas_shelter_survdata_FY18-20.csv.gz")

coxph_features = c("chip_status", "surv_hours", "event")
survdata.hex = as.h2o(survdata[, ..coxph_features])
survdata.hex$event = as.factor(survdata.hex$event)
survdata.hex$chip_status = as.factor(survdata.hex$chip_status)

survdata.coxph = h2o.coxph(model_id = "Dallas-animal-shelter-coxph-model",
training_frame = survdata.hex,
stop_column = "surv_days",
event_column = "event"){code}

getting error:

[^h2o_127.0.0.1_54321-5-error.log]
[^h2o_127.0.0.1_54321-3-info.log]

[exalate-issue-sync]

Jan Sterba commented: this is caused by {{"surv_days"}} missing from the training dataset, the fact that this causes a NPE is a bug in deed that we need to fix

[exalate-issue-sync]

Gregory Kanevsky commented: NPE also happens when {{start_column }} doesn’t exist.

[exalate-issue-sync]

Gregory Kanevsky commented: When covariates are not numeric or factors the following error reported:

{quote}java.lang.ArrayIndexOutOfBoundsException: Index 0 out of bounds for length 0

java.lang.ArrayIndexOutOfBoundsException: Index 0 out of bounds for length 0
at Jama.Matrix.(Matrix.java:113)
at hex.coxph.CoxPH$CoxPHDriver.calcModelStats(CoxPH.java:392)
at hex.coxph.CoxPH$CoxPHDriver.computeImpl(CoxPH.java:563)
at hex.ModelBuilder$Driver.compute2(ModelBuilder.java:246)
at hex.coxph.CoxPH$CoxPHDriver.compute2(CoxPH.java:231)
at water.H2O$H2OCountedCompleter.compute(H2O.java:1551)
at jsr166y.CountedCompleter.exec(CountedCompleter.java:468)
at jsr166y.ForkJoinTask.doExec(ForkJoinTask.java:263)
at jsr166y.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:974)
at jsr166y.ForkJoinPool.runWorker(ForkJoinPool.java:1477)
at jsr166y.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:104)

Error: java.lang.ArrayIndexOutOfBoundsException: Index 0 out of bounds for length 0
In addition: Warning message:
In .h2o.processResponseWarnings(res) :
Dropping bad and constant columns: [intake_type, animal_breed, intake_condition, chip_status].{quote}

we would benefit from more explicit message indicating that features (covariates) must be numeric or factors.

[exalate-issue-sync]

Jan Sterba commented: added missing validation and improved the error messages

[h2o-ops]

JIRA Issue Migration Info

Jira Issue: PUBDEV-7492
Assignee: Jan Sterba
Reporter: Gregory Kanevsky
State: Resolved
Fix Version: 3.30.0.3
Attachments: Available (Count: 3)
Development PRs: Available

Linked PRs from JIRA

#4578

Attachments From Jira

Attachment Name: dallas_shelter_survdata_FY18-20.csv.gz
Attached By: Gregory Kanevsky
File Link:https://h2o-3-jira-github-migration.s3.amazonaws.com/PUBDEV-7492/dallas_shelter_survdata_FY18-20.csv.gz

Attachment Name: h2o_127.0.0.1_54321-3-info.log
Attached By: Gregory Kanevsky
File Link:https://h2o-3-jira-github-migration.s3.amazonaws.com/PUBDEV-7492/h2o_127.0.0.1_54321-3-info.log

Attachment Name: h2o_127.0.0.1_54321-5-error.log
Attached By: Gregory Kanevsky
File Link:https://h2o-3-jira-github-migration.s3.amazonaws.com/PUBDEV-7492/h2o_127.0.0.1_54321-5-error.log