jpmml/r2pmml

XGboost converstion issue

Kawalierus opened this issue · 10 comments

Kawalierus commented

I have tried recreating example as provided in example (https://github.com/jpmml/r2pmml#package-xgboost). Unfortunately I receive error message :

SEVERE: Failed to convert RDS to PMML
java.lang.ClassCastException: org.jpmml.rexp.RStringVector cannot be cast to org.jpmml.rexp.RIntegerVector
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:283)
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:265)
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:230)
at org.jpmml.rexp.XGBoostConverter.ensureFeatureMap(XGBoostConverter.java:209)
at org.jpmml.rexp.XGBoostConverter.encodeSchema(XGBoostConverter.java:66)
at org.jpmml.rexp.ModelConverter.encodePMML(ModelConverter.java:70)
at org.jpmml.rexp.Converter.encodePMML(Converter.java:39)
at org.jpmml.rexp.Main.run(Main.java:149)
at org.jpmml.rexp.Main.main(Main.java:97)

Exception in thread "main" java.lang.ClassCastException: org.jpmml.rexp.RStringVector cannot be cast to org.jpmml.rexp.RIntegerVector
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:283)
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:265)
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:230)
at org.jpmml.rexp.XGBoostConverter.ensureFeatureMap(XGBoostConverter.java:209)
at org.jpmml.rexp.XGBoostConverter.encodeSchema(XGBoostConverter.java:66)
at org.jpmml.rexp.ModelConverter.encodePMML(ModelConverter.java:70)
at org.jpmml.rexp.Converter.encodePMML(Converter.java:39)
at org.jpmml.rexp.Main.run(Main.java:149)
at org.jpmml.rexp.Main.main(Main.java:97)
I have notice that in the similar topic (https://www.gitmemory.com/issue/jpmml/r2pmml/64/718139304) you suggested that label should be factorized. Nevertheless it would cause error when providing label as factor to xgboost function [17:10:50] amalgamation/../src/objective/multiclass_obj.cu:120: SoftmaxMultiClassObj: label must be in [0, num_class).

I have initially tried to run conversion on poisson count model and obtained identical error (and prediction there is frequency so I do not see where strings are to be converted to integers).

R version: 4.0.3
Xgboost version: 1.3.2.1
r2pmml version: 0.25.1

I would greatly appreciate support regarding this issue.

vruusmann commented

java.lang.ClassCastException: org.jpmml.rexp.RStringVector cannot be cast to org.jpmml.rexp.RIntegerVector
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:283)

This is the line 183 of XGBoostConverter source code:
https://github.com/jpmml/jpmml-r/blob/1.4.2/src/main/java/org/jpmml/rexp/XGBoostConverter.java#L283

As you can see, the converter expects that the name attribute of the Feature Map (aka FMap) object is a R factor. You've been giving it a R string instead.

To fix resolve this conversion error, simply change the type of the name attribute to R factor:

iris.fmap = as.fmap(iris.matrix)

# THIS!
iris.fmap$name = as.factor(iris.fmap$name)

#64

These two issues are close relatives, because they both reflect inproper typing of FMap attributes (name here, type there).

All FMap attributes must be R factors.

vruusmann commented

All FMap attributes must be R factors.

Leaving this issue open for now - the r2pmml::as.fmap utility function should always force-cast all FMap attributes to R factors before returning the result to the end user.

Kawalierus commented

Thank you very much for such a quick response.

Unfortunately your advice have not resolved the issue in my case. Please note the code as below:

library("xgboost")
library("r2pmml")

data(iris)

iris_X = iris[, 1:4]
iris_y = as.integer(iris[, 5]) - 1

iris.matrix = model.matrix(~ . - 1, data = iris_X)

iris.DMatrix = xgb.DMatrix(iris.matrix, label = iris_y)
iris.fmap = as.fmap(iris.matrix)
iris.fmap$name = as.factor(iris.fmap$name)
iris.xgb = xgboost(data = iris.DMatrix, missing = NULL, objective = "multi:softmax", num_class = 3, nrounds = 13)

r2pmml(iris.xgb, "iris_xgb.pmml", fmap = iris.fmap, response_name = "Species", response_levels = c("setosa", "versicolor", "virginica"), missing = NULL, ntreelimit = 7, compact = TRUE)

still results in the error:

SEVERE: Failed to convert RDS to PMML
java.lang.ClassCastException: org.jpmml.rexp.RStringVector cannot be cast to org.jpmml.rexp.RIntegerVector
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:284)
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:265)
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:230)
at org.jpmml.rexp.XGBoostConverter.ensureFeatureMap(XGBoostConverter.java:209)
at org.jpmml.rexp.XGBoostConverter.encodeSchema(XGBoostConverter.java:66)
at org.jpmml.rexp.ModelConverter.encodePMML(ModelConverter.java:70)
at org.jpmml.rexp.Converter.encodePMML(Converter.java:39)
at org.jpmml.rexp.Main.run(Main.java:149)
at org.jpmml.rexp.Main.main(Main.java:97)

Exception in thread "main" java.lang.ClassCastException: org.jpmml.rexp.RStringVector cannot be cast to org.jpmml.rexp.RIntegerVector
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:284)
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:265)
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:230)
at org.jpmml.rexp.XGBoostConverter.ensureFeatureMap(XGBoostConverter.java:209)
at org.jpmml.rexp.XGBoostConverter.encodeSchema(XGBoostConverter.java:66)
at org.jpmml.rexp.ModelConverter.encodePMML(ModelConverter.java:70)
at org.jpmml.rexp.Converter.encodePMML(Converter.java:39)
at org.jpmml.rexp.Main.run(Main.java:149)
at org.jpmml.rexp.Main.main(Main.java:97)

vruusmann commented

Exception in thread "main" java.lang.ClassCastException: org.jpmml.rexp.RStringVector cannot be cast to org.jpmml.rexp.RIntegerVector
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:284)

See - you fixed the type of the name attribute, and now the exception is happening one line later (284 instead of 283)

Please see my earlier resolution - "All FMap attributes must be R factors"

iris.fmap = as.fmap(iris.matrix)

iris.fmap$id = as.factor(iris.fmap$id)
iris.fmap$name = as.factor(iris.fmap$name)
iris.fmap$type = as.factor(iris.fmap$type)
vruusmann commented

My XGBoost example works fine with R 3.X.

Looks like R 4.X has botched matrix column types (were factors before, are strings now).

Kawalierus commented

Thank you very much again. Apologies for not noticing this at first glance. We are making steady progress as I have reached another type of error by applying all your corrections.

Code:
library("xgboost")
library("r2pmml")

data(iris)

iris_X = iris[, 1:4]
iris_y = as.integer(iris[, 5]) - 1

iris.matrix = model.matrix(~ . - 1, data = iris_X)

iris.DMatrix = xgb.DMatrix(iris.matrix, label = iris_y)
iris.fmap = as.fmap(iris.matrix)

iris.fmap$id = as.factor(iris.fmap$id)
iris.fmap$name = as.factor(iris.fmap$name)
iris.fmap$type = as.factor(iris.fmap$type)

iris.xgb = xgboost(data = iris.DMatrix, missing = NULL, objective = "multi:softmax", num_class = 3, nrounds = 13)

r2pmml(iris.xgb, "iris_xgb.pmml", fmap = iris.fmap, response_name = "Species", response_levels = c("setosa", "versicolor", "virginica"), missing = NULL, ntreelimit = 7, compact = TRUE)

Error:

SEVERE: Failed to convert RDS to PMML
java.lang.IllegalArgumentException
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:295)
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:265)
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:230)
at org.jpmml.rexp.XGBoostConverter.ensureFeatureMap(XGBoostConverter.java:209)
at org.jpmml.rexp.XGBoostConverter.encodeSchema(XGBoostConverter.java:66)
at org.jpmml.rexp.ModelConverter.encodePMML(ModelConverter.java:70)
at org.jpmml.rexp.Converter.encodePMML(Converter.java:39)
at org.jpmml.rexp.Main.run(Main.java:149)
at org.jpmml.rexp.Main.main(Main.java:97)

Exception in thread "main" java.lang.IllegalArgumentException
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:295)
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:265)
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:230)
at org.jpmml.rexp.XGBoostConverter.ensureFeatureMap(XGBoostConverter.java:209)
at org.jpmml.rexp.XGBoostConverter.encodeSchema(XGBoostConverter.java:66)
at org.jpmml.rexp.ModelConverter.encodePMML(ModelConverter.java:70)
at org.jpmml.rexp.Converter.encodePMML(Converter.java:39)
at org.jpmml.rexp.Main.run(Main.java:149)
at org.jpmml.rexp.Main.main(Main.java:97)

vruusmann commented

java.lang.IllegalArgumentException
at org.jpmml.rexp.XGBoostConverter.loadFeatureMap(XGBoostConverter.java:295)

See for yourself:
https://github.com/jpmml/jpmml-r/blob/1.4.2/src/main/java/org/jpmml/rexp/XGBoostConverter.java#L295

Looks like it's not allowed to convert the id attribute to R factor; leave it to be a R integer.

iris.fmap = as.fmap(iris.matrix)

iris.fmap$name = as.factor(iris.fmap$name)
iris.fmap$type = as.factor(iris.fmap$type)
Kawalierus commented

And we have reached another one:

SEVERE: Failed to convert RDS to PMML
java.lang.IllegalArgumentException: 1730313018.1919250021
at org.jpmml.xgboost.Learner.load(Learner.java:82)
at org.jpmml.xgboost.XGBoostUtil.loadLearner(XGBoostUtil.java:93)
at org.jpmml.xgboost.XGBoostUtil.loadLearner(XGBoostUtil.java:57)
at org.jpmml.xgboost.XGBoostUtil.loadLearner(XGBoostUtil.java:45)
at org.jpmml.rexp.XGBoostConverter.loadLearner(XGBoostConverter.java:309)
at org.jpmml.rexp.XGBoostConverter.loadLearner(XGBoostConverter.java:242)
at org.jpmml.rexp.XGBoostConverter.ensureLearner(XGBoostConverter.java:218)
at org.jpmml.rexp.XGBoostConverter.encodeSchema(XGBoostConverter.java:80)
at org.jpmml.rexp.ModelConverter.encodePMML(ModelConverter.java:70)
at org.jpmml.rexp.Converter.encodePMML(Converter.java:39)
at org.jpmml.rexp.Main.run(Main.java:149)
at org.jpmml.rexp.Main.main(Main.java:97)

Exception in thread "main" java.lang.IllegalArgumentException: 1730313018.1919250021
at org.jpmml.xgboost.Learner.load(Learner.java:82)
at org.jpmml.xgboost.XGBoostUtil.loadLearner(XGBoostUtil.java:93)
at org.jpmml.xgboost.XGBoostUtil.loadLearner(XGBoostUtil.java:57)
at org.jpmml.xgboost.XGBoostUtil.loadLearner(XGBoostUtil.java:45)
at org.jpmml.rexp.XGBoostConverter.loadLearner(XGBoostConverter.java:309)
at org.jpmml.rexp.XGBoostConverter.loadLearner(XGBoostConverter.java:242)
at org.jpmml.rexp.XGBoostConverter.ensureLearner(XGBoostConverter.java:218)
at org.jpmml.rexp.XGBoostConverter.encodeSchema(XGBoostConverter.java:80)
at org.jpmml.rexp.ModelConverter.encodePMML(ModelConverter.java:70)
at org.jpmml.rexp.Converter.encodePMML(Converter.java:39)
at org.jpmml.rexp.Main.run(Main.java:149)
at org.jpmml.rexp.Main.main(Main.java:97)

vruusmann commented

java.lang.IllegalArgumentException: 1730313018.1919250021

Duplicate of jpmml/jpmml-xgboost#54

TLDR: XGBoost 1.3.X switched model saving data format from binary to JSON.

Two solutions:

  • Order XGBoost 1.3.X to save model in binary data format.
  • Downgrade XGBoost to 1.2.X.
vruusmann commented

@Kawalierus I've released R2PMML version 0.25.2 to GitHub, which is able to convert XGBoost 1.3(.3) models now.