Develop Reference

pmml model created from xgboost in R leads to different result than original model in R

I have a ranking task, where my training data looks like this:
session_id item_id item_features target
---------------------------------------------
session1 item1 ... 1
session1 item2 ... 0
...
sessionN item1 ... 0
sessionN itemX ... 10
sessionN itemY ... 0
...
I am using xgboost in R with the objective "rank:pairwise" for training the model. xgboost expects grouped data (same session_id) to be bunched together in the training and test sets. The lines belonging to the same session_id have to be specified using the function setinfo() (e. g. setinfo(model, 'group', group_info).
When I evaluate the model in R, applying new data works perfectly. However, I have used the package pmml to convert the model into a pmml file in order to use it in Java.
In Java the pmml file gets parsed and evaluated via the org.jpmml pmml-evaluator dependency (v. 1.3.15). Feeding the same data as in R to the org.jpmml.evaluator.Evaluator yields different results, though. The results are mostly negative values - which is no valid result in my setup- all predicted targets should be positive.
I have come up with two possible explanations:
There might be a bug in the pmml conversion in my scenario
I have no idea, where I can apply the equivalent of setinfo() in Java. Since I am only applying the model to a single session at a time, I was under the impression that I did not need to specify it. But maybe, I was wrong.
Please contact me for fully working example including training and test data, I will send via mail. But for starters, here is the R code from training the model:
library(xgboost)
example_matrix_train <- xgb.DMatrix(X, label = y)
setinfo(example_matrix_train, 'group', example_train_groupInfo)
example.model <- xgboost(data = example_matrix_train, objective = "rank:pairwise", max.depth = 8, eta = 0.2, nthread = 8, nround = 10, verbose=0)
library(pmml)
library(pmmlTransformations)
xgb.dump(example.model, "example.model.dumped.trees")
logfile <- file(paste0("pmml_example_model",Sys.Date(),".txt"), open="a")
sink(logfile)
pmml(example.model, inputFeatureNames = colnames(example_train), outputLabelName = "prediction1", xgbDumpFile = "example.model.dumped.trees")
sink()
Any help is welcome

I have come up with two possible explanations: There might be a bug in the pmml conversion
This is the true explanation - the pmml package is producing incorrect PMML for XGBoost models. The technical reason is that it is using XGBoost text dump file as input, but the information contained therein is incomplete (eg. rounded threshold values).
If you're looking to export XGBoost models into PMML, then you should be using the r2pmml package, which is using XGBoost binary files as input.

In truth, the 'pmml' package currently does not support the 'rank:pairwise' objective function you need. The upcoming release of the 'pmml' package (version 1.5.3) includes a check for unsupported objective functions.

Java - xgboost DMatrix input

When creating a DMatrix in java with the xgboost4j package, at first i succeed to create the matrix using a "filepath".
DMatrix trainMat = new DMatrix("...\\xgb_training_input.csv");
But when I try to train the model:
Booster booster = XGBoost.train(trainMat, params, round, watches, null, null);
I get the following error:
...regression_obj.cc:108: label must be in [0,1] for logistic regression
now my data is solid. I've checked it out on an xgb model built in python.
I'm guessing the problem is with the data format somehow.
currently the format is as follows:
x1,x2,x3,x4,x5,y
where x1-x5 are "Real" numbers and y is either 0 or 1. file end is .csv
Maybe the separator shouldn't be ',' ?

DMatrix gets an .libsvm file. which can be easily created with python.
libsvm looks like this:
target 0:column1 1:column2 2:column3 ... and so on
so the target is the first column, while every other column (predictor) is being attached to increasing index with ":" in between.

jcrfsuite training file format

From what I understand from the example of POS Tagging given in the examples of jcrfsuite. The training file is tab separated and first token is the label. But I do not get the BigCluster| thing. Can somebody help me with how to specify tokens in training file.
Example below:
O BigCluster|00 BigCluster|0000 BigCluster|000000 BigCluster|00000000 BigCluster|0000000000 BigCluster|000000000000 BigCluster|00000000000000 BigCluster|0000000000000000 NextBigCluster|0100 NextBigCluster|01000101 NextBigCluster|010001011111 POSTagDict|D POSTagDict|N POSTagDict|^ POSTagDict|$ POSTagDict|G NextPOSTag|V 1gramSuff|i 1gramPref|i prevword| prevcurr||i nextword|predict nextword|predict currnext|i|predict Word|I Lower|i Xxdshape|X charclass|1, first-shortcap prevnext||predict t=0
Test file format:
! BigCluster|01 BigCluster|0110 BigCluster|011011 BigCluster|01101100 BigCluster|0110110011 BigCluster|011011001100 BigCluster|01101100110000 BigCluster|0110110011000000 NextBigCluster|1000 NextBigCluster|10001000 NextBigCluster|100010000000 POSTagDict|V NextPOSTag|, metaph_POSDict|N 1gramSuff|n 2gramSuff|nn 3gramSuff|mnn 4gramSuff|mmnn 5gramSuff|mmmnn 6gramSuff|ammmnn 7gramSuff|aammmnn 8gramSuff|aaammmnn 9gramSuff|daaammmnn 1gramPref|d 2gramPref|da 3gramPref|daa 4gramPref|daaa 5gramPref|daaam 6gramPref|daaamm 7gramPref|daaammm 8gramPref|daaammmn 9gramPref|daaammmnn prevword| prevcurr||daaammmnn nextword|. nextword|. currnext|daaammmnn|. Word|Daaammmnn Lower|daaammmnn Xxdshape|Xxxxxxxxx charclass|1,2,2,2,2,2,2,2,2, first-initcap prevnext||. t=0

What is specified after the label is a list of feature-name and feature-value.
It is in a sparse representation instead of tabular representation.
BigCluster is just one of the features and it's relevant to the specific example only. You should create your own features if you are training from scratch.

I have noticed that CRFsuite does not care for the naming convention nor feature design of labels and attributes, because treats them as strings.
CRFsuite learns weights of associations (feature weights) between attributes and labels, without knowing the meaning of labels and attributes. In other words, one can design and use arbitrary features just by writing label and attribute names in data sets, just find the best posible attributes for your example and run some experiments with different sets of attributes and features. And you will good to go.

How To Evaluate Predictive Neural Network in Encog

We're a creating a neural network for predicting typhoon occurrence using several typhoon parameters as input. So far, we have been able to generate data and train the neural network using Encog 3.2. Right now, we need to evaluate the results of the training.
We're using the ForestCover project (in Encog 3.2 examples) as reference, however the said project's Evaluation code is for a classification neural network. Thus, we can't evaluate our neural network following said project's code.
We also checked the PredictMarket project (in Encog 3.2 examples) since it is a predictive neural network. But we're having a difficulty in using the MLData.
MLData output = network.compute(inputData);
We want to extract the contents of output and compare it to the contents of the evaluation.csv for neural-network evaluation.
Is there a way we can extract/convert the output variable into a normalized value which we can then compare to the normalized evaluation.csv?
or
Is there a way we can modify the ForestCover Evaluate.java file to be able to evaluate a predictive neural network?
Thank you.

Here is a C# example (Java should be similar) which writes out a .csv file (TestResultsFile) with both the de-normalized expected and actual results so you can compare them with an Excel graph.
var evaluationSet = (BasicMLDataSet)EncogUtility.LoadCSV2Memory(Config.EvaluationNormalizedFile.ToString(),
network.InputCount, network.OutputCount, true, CSVFormat.English,
false);
var analyst = new EncogAnalyst();
analyst.Load(Config.NormalizationAnalystFile);
// Change this to whatever your output field index is
int outputFieldIndex = 29;
using (var resultsFile = new System.IO.StreamWriter(Config.TestResultsFile.ToString()))
{
foreach (var item in evaluationSet)
{
var normalizedActualOuput = (BasicMLData)network.Compute(item.Input);
var actualOutput = analyst.Script.Normalize.NormalizedFields[outputFieldIndex].DeNormalize(normalizedActualOuput.Data[0]);
var idealOutput = analyst.Script.Normalize.NormalizedFields[outputFieldIndex].DeNormalize(item.Ideal[0]);
var resultLine = String.Format("{0},{1}", idealOutput, actualOutput);
resultsFile.WriteLine(resultLine);
}
}
Much of this came from ideas from Abishek Kumar's Pluralsight Course
If you really want to compare Normalized values, just remove the two calls to "Denormalize".

Convert OpenIE triplet to N-Triplet (NT)

I downloaded and used OpenIE4.1 jar file (downloadable from http://knowitall.github.io/openie/) to process some free text documents and produced triplet-like outputs along with the text and confidence score, for instance,
The rail launchers are conceptually similar to the underslung SM-1
0.93 (The rail launchers; are; conceptually similar to the underslung SM-1)
I wrote a java parser to extract OpenIE triplets which confidence score is >= 0.85 and
need to know the way to convert it to N-triplet (NT), format look like.
Not sure if I need to be familiar with the ontology that I'm trying to map to.

After discussion with my colleagues. This is what I should do to create N-Triplet(NT) and Detailed Java codes can be found in another Question: Use RDF API (Jena, OpenRDF or Protege) to convert OpenIE outputs
Create a blank node identifier for each distinct :subject in the file (call it node_s)
Create a blank node identifier for each distinct :object in the file (call it node_o)
Define a URI for each distinct predicate
Create these triples:
1. node_s rdf:type <http://mypage.org/vocab#Corpus>
2. node_s dc:title “The rail launchers”
3. node_s dc:source “Sample File”
4. node_s rdf:predicate <http://mypage.org/vocab#are>
5. node_o rdf:type <http://mypage.org/vocab#Corpus>
6. node_o dc:title “conceptually similar to the underslung SM-1”