Wondering if there a runWithValidation feature for Gradient Boosted Trees (GBT) in Spark ml to prevent overfitting. It's there in mllib which works with RDDs. I am looking the same for dataframes.
Found a K-Fold Cross Validation support in Spark. It can be done using CrossValidation() with Estimators, Evaluators, ParamMap and number of folds. This helps in finding the best parameters for the model i.e model tuning.
Refer http://spark.apache.org/docs/latest/ml-tuning.html for more details.
Related
I want to use scala and spark to implement Graph algorithm GraphSAGE, then how to do it? Is there any source code?
I want to get the code for my question
I haven´t implemented yet this graph algorithms on top of Spark, the only available implementation, as far as I know, for using deep learning for graph analysis is this. It is a spectral graph convolution for semi-supervised learning, and it is a transductive algorithm. It can be used for node classification. I have plans to include more algorithms in the future like GraphSAGE.
I want to ask is this possible to write a custom loss function for Multi class Classification in Spark using Scala. I want to code multi-class logarithmic loss in Scala. I searched Spark documentation but could not get any hint.
From the Spark 2.2.0 MLlib guide:
Currently, only binary classification is supported.. This will likely change when multiclass classification is supported.
If you are not restricted to a particular classification technique I would suggest using XGBoost. It has a Spark-compatible implementation, and it makes it possible to use any loss function provided you can compute is derivative twice.
You can find a tutorial here.
Also the explanation about why it is possible to use a custom loss function can be found here.
I'm a bit new to Spark ML API. I'm trying to do multi-label classification for 160 labels by training 160 classifiers(logistic or random forest etc). Once I train on Dataset[LabeledPoint], I'm finding it hard to get an API where I get the probability for each class for a single example. I've read on SO that you can use the pipeline API and get the probabilities, but for my use case this is going to be hard because I'll have to repicate 160 RDDs for my evaluation features, get probability for each class and then do a join to rank the classes by their probabilities. Instead, I want to just have one copy of evaluation features, broadcast the 160 models and then do the predictions inside the map function. I find myself having to implement this but wonder if there's another convenience API in Spark to do the same for different classifiers like Logistic/RF which converts a Vector representing features to the probability for it belonging to a class. Please let me know if there's a better way to approach multi-label classification in Spark.
EDIT: I tried to create a function to transform a vector to a label for random forest, but it's super annoying because I now have to clone large pieces of tree traversal in Spark, and almost everywhere I encountered dead ends because some function or variable was private or protected. Correct me if wrong, but if this use case is not already implemented, I think it atleast is well-justified because Scikit-learn already has such APIs in place to do this.
Thanks
Found the culprit line in Spark MLLib code: https://github.com/apache/spark/blob/5ad644a4cefc20e4f198d614c59b8b0f75a228ba/mllib/src/main/scala/org/apache/spark/ml/Predictor.scala#L224
The predict method is marked as protected but it should actually be public for such use cases to be supported.
This has been fixed in version 2.4 as seen here:
https://github.com/apache/spark/blob/branch-2.4/mllib/src/main/scala/org/apache/spark/ml/Predictor.scala
So upgrading to version 2.4 should do the trick ... although I don't think 2.4 is out yet, so it's a matter of waiting.
EDIT: for people that are interested, apparently not only is this beneficial for multi-label prediction, it's been observed that there's 3-4x improvement in latency as well for regular classification/regression for single instance/small batch predictions (see https://issues.apache.org/jira/browse/SPARK-16198 for details).
I am currently looking for an Algorithm in Apache Spark (Scala/Java) that is able to cluster data that has numeric and categorical features.
As far as I have seen, there is an implementation for k-medoids and k-prototypes for pyspark (https://github.com/ThinkBigAnalytics/pyspark-distributed-kmodes), but I could not identify something similar for the Scala/Java version I am currently working with.
Is there another recommend algorithm to achieve similar things for Spark running Scala? Or am I overlooking something and could actually make use of the pyspark library in my Scala project?
If you need further information or clarification feel free to ask.
I think you need first to convert your categorical variables to numbers using OneHotEncoder then, you can apply your clustering algorithm using mllib (e.g. kmeans). Also, I recommend doing scaling or normalization before applying your cluster algorithm as it is distance sensitive.
I do not think gaussian mixture model is available in mllib yet. I am wondering if any good Scala/Java implementation of GMM (suitable for large data) is available elsewhere. Please let me know.
Thanks and regards,
It is available in Spark MLlib now:
http://spark.apache.org/docs/latest/mllib-clustering.html#gaussian-mixture
Have a look at https://issues.apache.org/jira/browse/SPARK-4156
It is still under progress. We can expect it soon in MLLib.