I am currently working with spark mllib.
I have created a text classifier using the Gradient Boosting algorithm with the class GradientBoostedTrees:
Gradient Boosted Trees
Currently I obtain the predictions to know the class of new elements but I would like to obtain the class probabilities (the value of the output before the hard decision).
In other mllib algorithms like logistic regression you can remove the threshold from the classifier to obtain the class probabilities but I can not find a way to do the same procedure with GradientBosstedTrees.
As far as I know, it's not currently possible but it is possible with random forest.
You can see this link...I have explained a procedure here
Predicting probabilities of classes in case of Gradient Boosting Trees in Spark using the tree output
In order to implement the predicted probabilities and thresholds one need to write program using the trees from
print(model.toDebugString)
output. I tried to understand how the tree works to predict which is fairly simple outside Spark.
It seems that in Spark MLLIB it is not possible to obtain the class probabilities.
You can only obtain the final classification decision.
That's a pity because that information would be very useful (If you classify a sample as positive with 99.99% of posibilities is not the same than 51%) and it is not difficult to obtain that information once the model has been trained.
An alternative is using a different software like xgboost: https://github.com/dmlc/xgboost
Related
I am trying for setting the initial weights or parameters for a machine learning (Classification) algorithm in Spark 2.x. Unfortunately, except for MultiLayerPerceptron algorithm, no other algorithm is providing a way to set the initial weights/parameter values.
I am trying to solve Incremental learning using spark. Here, I need to load old model re-train the old model with new data in the system. How can I do this?
How can I do this for other algorithms like:
Decision Trees
Random Forest
SVM
Logistic Regression
I need to experiment multiple algorithms and then need to choose the best performing one.
How can I do this for other algorithms like:
Decision Trees
Random Forest
You cannot. Tree based algorithms are not well suited for incremental learning, as they look at the global properties of the data and have no "initial weights or values" that can be used to bootstrap the process.
Logistic Regression
You can use StreamingLogisticRegressionWithSGD which exactly implements required process, including setting initial weights with setInitialWeights.
SVM
In theory it could be implemented similarly to streaming regression StreamingLogisticRegressionWithSGD or StreamingLinearRegressionWithSGD, by extending StreamingLinearAlgorithm, but there is no such implementation built-in, ans since org.apache.spark.mllib is in a maintanance mode, there won't be.
It's not based on spark, but there is a C++ incremental decision tree.
see gaenari.
Continuous chunking data can be inserted and updated, and rebuilds can be run if concept drift reduces accuracy.
Is there a way to get the significance level of each coefficient we receive after we fit a logistic regression model on training data?
I was trying to find out a way and could not figure out myself.
I think I may get the significance level of each feature if I run chi sq test but first of all not sure if I can run the test on all features together and secondly I have numeric data value so if it will give me right result or not that remains a question as well.
Right now I am running the modeling part using statsmodel and scikit learn but certainly, want to know, how can I get these results from PySpark ML or MLLib itself
If anyone can shed some light, it will be helpful
I use only mllib, I think that when you train a model you can use toPMML method to export your model un PMML format (xml file), then you can parse the xml file to get features weights, here an example
https://spark.apache.org/docs/2.0.2/mllib-pmml-model-export.html
Hope that will help
Taking cue from Is it possible to obtain class probabilities using GradientBoostedTrees with spark mllib?
The answer to this question is given last December. I wanted to know if there is anything in Spark 1.6.1.
I tried to edit the above question but somehow it did not happen.
My query is like most other algorithms in Spark mllib library we can find predicted probabilities. Can we get the same in Gradient Boosting Trees-Classification(http://spark.apache.org/docs/latest/mllib-ensembles.html#gradient-boosted-trees-vs-random-forests) and change the model threshold?
That way I can change the threshold to improve the model performance.
I'm using GradientBoostingClassifier for my unbalanced labeled datasets. It seems like class weight doesn't exist as a parameter for this classifier in Sklearn. I see I can use sample_weight when fit but I cannot use it when I deal with VotingClassifier or GridSearch. Could someone help?
Currently there isn't a way to use class_weights for GB in sklearn.
Don't confuse this with sample_weight
Sample Weights change the loss function and your score that you're trying to optimize. This is often used in case of survey data where sampling approaches have gaps.
Class Weights are used to correct class imbalances as a proxy for over \ undersampling. There is no direct way to do that for GB in sklearn (you can do that in Random Forests though)
Very late, but I hope it can be useful for other members.
In the article of Zichen Wang in towardsdatascience.com, the point 5 Gradient Boosting it is told:
For instance, Gradient Boosting Machines (GBM) deals with class imbalance by constructing successive training sets based on incorrectly classified examples. It usually outperforms Random Forest on imbalanced dataset For instance, Gradient Boosting Machines (GBM) deals with class imbalance by constructing successive training sets based on incorrectly classified examples. It usually outperforms Random Forest on imbalanced dataset.
And a chart shows that the half of the grandient boosting model have an AUROC over 80%. So considering GB models performances and the way they are done, it seems not to be necessary to introduce a kind of class_weight parameter as it is the case for RandomForestClassifier in sklearn package.
In the book Introduction To Machine Learning with Pyhton written by Andreas C. Müller and Sarah Guido, edition 2017, page 89, Chapter 2 *Supervised Learning, section Ensembles of Decision Trees, sub-section Gradient boosted regression trees (gradient boosting machines):
They are generally a bit more sensitive to
parameter settings than random forests, but can provide better accuracy if the parameters are set correctly.
Now if you still have scoring problems due to imbalance proportions of categories in the target variable, it is possible you should see if your data should be splited to apply different models on it, because they are not as homogeneous as it seems to be. I mean it may have a variable you have not in your dataset train (an hidden variable clearly) that influences a lot the model results, then it is difficult even for the greater GB to give correct scoring because it misses a huge information that you cannot make appear in the matrix to compute sometimes for many reasons.
Some updates:
I found, by random, there are libraries that implement it as parameters of their gradient boosting instance objects. It is the case of H2O where for the parameter balance_classes it is told:
Balance training data class counts via over/under-sampling (for
imbalanced data).
Type: bool (default: False).
If you want to keep with sklearn you should do as HakunaMaData told: over/under-sampling because that's what other libraries finally do when the parameter exist.
I am working with sklearn's implementation of KNN. While my input data has about 20 features, I believe some of the features are more important than others. Is there a way to:
set the feature weights for each feature when "training" the KNN learner.
learn what the optimal weight values are with or without pre-processing the data.
On a related note, I understand generally KNN does not require training but since sklearn implements it using KDTrees, the tree must be generated from the training data. However, this sounds like its turning KNN into a binary tree problem. Is that the case?
Thanks.
kNN is simply based on a distance function. When you say "feature two is more important than others" it usually means difference in feature two is worth, say, 10x difference in other coords. Simple way to achive this is by multiplying coord #2 by its weight. So you put into the tree not the original coords but coords multiplied by their respective weights.
In case your features are combinations of the coords, you might need to apply appropriate matrix transform on your coords before applying weights, see PCA (principal component analysis). PCA is likely to help you with question 2.
The answer to question to is called "metric learning" and currently not implemented in Scikit-learn. Using the popular Mahalanobis distance amounts to rescaling the data using StandardScaler. Ideally you would want your metric to take into account the labels.