Munu sairamesh

icon location Machine learning researcher

Results 16 comments of Munu sairamesh

Request for -invert-hash type feature as in Vowpal Wabbit.

@eisber I was looking at your reply for https://github.com/microsoft/SynapseML/issues/1364. I am finding it difficult to understand your statement "> Additionally it depends on the learning algo how the features are...

Getting typeError: 'JavaPackage' object is not callable. for LightGBMRegressor.

@imatiach-msft, I run on the cluster. I had tried using mmlspark which was not working and felt outdated. Then I tried using synapseML. I have attached the screenshot which confirms...

Getting typeError: 'JavaPackage' object is not callable. for LightGBMRegressor.

Yes, I missed it. I am using now mmlspark:1.0.0-rc4. Thanks @imatiach-msft

XGBoost-Spark training fails sometimes due to "java.lang.NumberFormatException: For input string: "inf""

I am facing the same above issue for objective "count:poisson" but works fine for the objective function. I am using data bricks environment, importing xgboost as import ml.dmlc.xgboost4j.scala.spark.{XGBoostRegressor}. can anybody...

When objective is "count:poisson", getting error "java.lang.NumberFormatException: For input string: "inf"

@trivialfis could you pls give some hint to fix this issue?

When objective is "count:poisson", getting error "java.lang.NumberFormatException: For input string: "inf"

@trivialfis yes it happens in JVM package. As I am using distributed environment, I have to use JVM based package.

Add label weight for cb

Hi @olgavrou. Is it straightforward to extend the weighted option to cb?

Add label weight for cb

@olgavrou It is a blocker. Currently, we have found some workaround but it would be nice to have this feature implemented. I will like to try to add this feature....

Add label weight for cb

Hi @olgavrou , Sorry for the late reply. I was traveling so the delay. I understood the workaround. Thanks for the help.

Avg loss of new model is not less than the avg loss of the old model even when the old model has not seen new data points.

@JohnLangford For evaluation, I am splitting the production data (generated using the old model policy) into the train_ data(80%) and test_data (20%). old_model_loss_on_test_data = avg cost of the test_data. I...