Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    mtsol
    @mtsol
    thanks @ancasarb
    shmyer
    @shmyer
    Thanks! @ancasarb
    tellarajesh
    @tellarajesh
    hi folks, just wondering is any road map available, like what features are coming , any plans to support spark3.0 . Also in docs it's mentioned custom transformers adding in Python and C support , is it fully in python or actual code in Scala then make bindings for PySpark?
    himanshusolanki
    @himanshusolanki

    Hi,
    can someone please help me with where exactly(in which file of this project) should I be putting this dependency to use xgboost.

    ml.combust.mleap mleap-xgboost-spark_${scala.binary.version} ${ml.combust.mleap.version}
    a configured example would also be helpful.

    I created the jars using master repo and tried using that but faced this error -
    error: object XGBoostRegressor is not a member of package ml.dmlc.xgboost4j.scala.spark
    import ml.dmlc.xgboost4j.scala.spark.XGBoostRegressor

    I had installed these jars -

    bundle-ml
    mleap-base
    mleap-core
    mleap-executor
    mleap-spark-base
    mleap-spark-extension
    mleap-spark
    mleap-xgboost-runtime
    mleap-xgboost-spark

    Alexis BRENON
    @AlexisBRENON

    Hi. I use MLeap in my project as well as GRPC, and I would like to upgrade ScalaPB to latest version (v0.10.11). However, upgrading it make mleap fail to serialize models that were serializable earlier with the following error:

    A needed class was not found. This could be due to an error in your runpath. Missing class: scalapb/Message
    java.lang.NoClassDefFoundError: scalapb/Message

    So I suppose that this is just a version mismatch between my version of scalaPB and the one expected by MLeap (I see that it still use the v0.7.1.
    Do you see any reason against upgrading scalaPB ? If no, would you appreciate a PR for this ?

    Anca Sarb
    @ancasarb
    hi all, just a small note to say the latest documentation is available at https://combust.github.io/mleap-docs/, thanks!
    anu-srivastava
    @anu-srivastava
    Hi All, Just want to bubble up this issues combust/mleap#475 and wanted to understand by when JDK11/Spark3 support will be added?
    Akarsh Gupta
    @akarsh3007
    How can we change the log level on mleap-serving, I am using the docker-image:0.11.0 ?
    antonkw
    @antonkw
    @seme0021 @hollinwilkins
    hi guys!
    Really quick question. I work with the team that use mleap 0.13, they fixed particular issue there. Is it possible to raise PR against 0.13 to see new minor version (0.13.1)?
    As far as I see all versions are chronologically ordered but still decided to double check if old versions have some non-obvious maintenance.
    Thanks!
    Ryan Vogan
    @voganrc

    Hi @ancasarb

    We think we've found an issue with MLeap's DenseTensor indexing code.

    It doesn't seem to follow row-major or column-major order, and has a different behavior from the SparseTensor indexing found in the same file below.

    Could you take a look at our combust/mleap#760 PR when you get a chance?

    Anca Sarb
    @ancasarb
    Hey sure thing, I’ll take a look tomorrow
    3 replies
    indranilr
    @indranilr

    Any

    Hi All, Just want to bubble up this issues combust/mleap#475 and wanted to understand by when JDK11/Spark3 support will be added?

    Hi @ancasarb Looking for support of Spark 3.1.1, XGboost 1.3.x ,Scala 2.12 and JDK 8 as indicated in this issue : combust/mleap#751 , could you please comment on expected availability of the fix ?

    Anca Sarb
    @ancasarb
    Hey, I’m reviewing that #760 PR should be done in the next day or so. And then I will finish off #475
    austinzh
    @austinzh
    Thanks Anca.
    indranilr
    @indranilr
    Anca, would combust/mleap#751 be solved as well with #475 ?
    mcarb123
    @mcarb123

    Hello! Im trying to serve a simple Pyspark LR model, since Mleap runtime is not available (AFAIK) in python im sending Rest calls to the mleap-spring-boot docker server.

    My model takes a single sparse vector as input, and with some trial/error work I managed to send requests in json format successfully.

    Does anyone know if it's possible to send sparse input in protobuf format? My ideal scenario since latency is a great constraint would be to be able to send gRPC/protobuf inference requests.

    My current body (json) that I could successfully send looks like

    {
       "rows":[
          [
             {
                "indices":[
                   [
                      544
                   ],
                   [
                      3818
                   ]
                ],
                "values":[
                   1,
                   1
                ],
                "dimensions":[
                   615504
                ]
             }
          ]
       ],
       "schema":{
          "fields":[
             {
                "type":{
                   "base":"double",
                   "type":"tensor",
                   "dimensions":[
                      615504
                   ]
                },
                "name":"features"
             }
          ]
       }
    }
    Marian Diaconu
    @neboduus

    Hi Everyone! Do you think is possible to transform a Map[String, FeatureClass] to a DefaultLeapFrame using some kind of automatic mechanism?

    Im using Scala
    The FeatureClass would be :
    FeatureClass(value: String, _type: MyType)

    What Id like to do is to have a dynamic Map (therefore contains a dynamic no of features) that can be transformed into a DefaultLeapFrame that is then fed to the model itself.

    Marian Diaconu
    @neboduus
    anyone here?
    Calin Cocan
    @calinrc
    Hi Everyone. I am trying to import MLeap source project in my IDE but I have a newer JDK in place (Open JDK - 15) and SBT fails resolving 'sun.misc.Cleaner' dependency. I have looked up in the source code and this is used in a single method from Platform class (allocateDirectBuffer) and the method is not used in any place. I might be reading the code incorrectly but in case is not used can be removed and allow MLeap compile (at least) with newer java version than 8?
    Thanks
    austinzh
    @austinzh
    Hi @ancasarb Would you please check if combust/mleap#760 is good to merge? Or if you have further comments, Please let me know.
    Ryan Vogan
    @voganrc
    Also combust/mleap#664 and combust/mleap#763 are small when you get a chance!
    Ryan Vogan
    @voganrc
    Hi @ancasarb, could you take a look at our Spark 3 PR when you get a chance?
    combust/mleap#765
    Arun
    @icarusin
    Hello all. May I know when is mleap v0.19.0 release planned?
    Noah Pritikin
    @cappaberra
    There's been some chatter recently on the log4j dependency that MLeap is pulling in via the springboot framework. Just FYI to all in chat here: combust/mleap#792
    Alex Holmes
    @alexholmes
    Hi folks - it looks like the Python 3.7 CI tests are failing for a no-op PR (combust/mleap#807). I'm not a Python expert so I was wondering whether anyone has seen the error before and would have any tips on how to fix. When I try running locally (with Python 3.7.9) I'm not able to reproduce the error on Travis:
    [7395] /home/travis/build/combust/mleap/python$ /home/travis/virtualenv/python3.7.1/bin/python setup.py sdist --formats=zip --dist-dir /home/travis/build/combust/mleap/python/.tox/dist >.tox/log/GLOB-0.log
    2923ERROR: invocation failed (exit code 1), logfile: /home/travis/build/combust/mleap/python/.tox/log/GLOB-0.log
    Jon Hoffman
    @hoffrocket
    hey all, i noticed that the xgboost-runtime can be configured to use the h2o XGBoost-Predictor. that's supposed to be more performant, but isn't in practice on my hardware (and it looks like it was last benchmarked several years ago against xgboost 1.1). so i'd like to continue to run the native xgboost4j library, but still take advantage of the optimization to only compute the probability here: https://github.com/combust/mleap/blob/master/mleap-xgboost-runtime/src/main/scala/ml/combust/mleap/xgboost/runtime/XGBoostPredictorClassification.scala
    anyone have ideas for how i can limit the native xgboost Classification op to only compute probability? would i have to define my own ml.combust.mleap.xgboost.runtime.bundle.ops.XGBoostClassificationOp class?