Inconsistent results for data with Nulls

[cnll0075]

Hi,

when I evaluate the model result in python it got different result from node.js in production. It turns out it is the problem with handling Null values. We used the API you mentioned: matrix(data, row, col, missing), and there is still about ~10% differences in terms of scores.

Could you please tell me what could the problem be, or give an example or how to deal with it? Right now we are doing something like this in production:

featureValues.push(item[col.name] || NaN)
})
})
featureValues = new Float32Array(featureValues)

let inputMatrix = new xgboost.matrix(featureValues,
  row,
  column, NaN)

[qinwf-nuan]

Please provide a simple example with code and data to reproduce the issue and the version of software and platform you are using. Thanks.