h2oai / steam

DEPRECATED Build, manage and deploy H2O's high-speed machine learning models.

DEPRECATED

This project is no longer maintained, please consider using Enteprise Steam instead.

Steam

Steam is an “instant on” platform that streamlines the entire process of building and deploying predictive applications. It is the industry’s first data science hub that lets data scientists and developers collaboratively build, deploy, and refine predictive applications across large scale datasets. Data scientists can publish Python and R code as REST APIs and easily integrate with production applications.

Steam in an open source product that uses AGPL code. (See the License file.) Contact H2O for information about receiving a licensed version of Steam.

Useful Links

Refer to the following for more information:

• Contributing to Steam
• Development in Steam
• Steam User Guide

Contacting Support

If you're an Open Source community member, you can contact H2O using one of the following methods:

• Click the Support link in the Steam UI to send an e-mail message
• Send an e-mail message directly to [email protected]
• Ask your question on Stack Overflow using the "h2o" tag

steam's Issues

No way of changing convertUnknownCategoricalLevelsToNa using prediction-service-builder

When manually loading a POJO model using Java, I can simply set convertUnknownCategoricalLevelsToNa to True when loading it, like:

EasyPredictModelWrapper model = new EasyPredictModelWrapper(
    new EasyPredictModelWrapper.Config()
        .setModel(rawModel)
        .setConvertUnknownCategoricalLevelsToNa(true));

However, as I understand it, this is not possible when using Steam's prediction service builder. There are no options or parameters that would allow this while building the .war file. So it's impossible to make predictions using unknown categorical labels when using the service builder.

Issue with Steam prediction service builder

I am not able to convert a POJO to war file. It gives the below mentioned error. This is first time i am facing this particular issue. What might be happening over here?

Steam with Sparkling Water cluster?

Can't find documentation on this topic..

Would we be able to use Steam with Sparkling Water cluster (as opposed to pure Flow clusters)?

Our data scientists are using Zeppelin/Jupyter notebooks (with PySpark on YARN) for programmatic definition of their workflows and use Flow on Sparkling Water so they could pass dataframes back and forth between the two environments. Now we're looking how Steam can fit into this mix.

Would it be possible to start a sparkling water Cluster from Steam, or otherwise connect to an existing sparkling water Cluster from Steam?

Would be great to have this documented. Thank you.

Unable see probability ratio for DRF model deployment in prediction builder

Hi Team, We built a DRF model from H2O and deployed to steam. While predicting the result-set we are not able to see probability ratio in prediction results in GUI. For same data I have built GLM model and deployed in steam and while predicting the results I am able to see the probability ration in UI. For DRF model I am able see Response column in prediction builder where for other model types I am not able to see response column in prediction builder.

Below is the screenshot for GLM model prediction results.

Below is the screenshot for DRF model prediction results.

I would like to know is it an issue in Steam or DRF model creation.

Steam-1.1.6 - Error: unknown flag: --admin-name

Hello,

I have following this user guide to install and run steam locally -
http://docs.h2o.ai/steam/latest-stable/Installation.html#installing-and-starting-steam-on-a-local-machine

However, I get error when running

./steam serve master --admin-name=admin --admin-password=admin012

$ ./steam serve master --admin-name=admin --admin-password=admin012

Error: unknown flag: --admin-name
Usage:
  steam serve master [flags]

Flags:
      --authentication-config="ldap.toml": Configuration file for authentication (used in "basic-ldap")
      --authentication-provider="basic": Authentication mechanism for client logins (one of "basic", "digest"), or "basic-ldap"
      --cluster-proxy-address=":9001": Cluster proxy address ("<ip>:<port>" or ":<port>")
      --compilation-service-address=":8080": Model compilation service address ("<ip>:<port>")
      --prediction-service-host="": Hostname to start prediction services on ("<ip>")
      --prediction-service-port-range="1025:65535": Specified port range to create prediction services on. ("<from>:<to>")
      --profile[=false]: Enable Go profiler
      --superuser-name="": Set superuser username (required for first-time-use only)
      --superuser-password="": Set superuser password (required for first-time-use only)
      --web-address=":9000": Web server address ("<ip>:<port>" or ":<port>").
      --web-tls-cert-path="": Web server TLS certificate file path (optional).
      --web-tls-key-path="": Web server TLS key file path (optional).
      --working-directory="var/master": Working directory for application files.
      --yarn-enable-kerberos[=false]: Enable Kerberos authentication. Requires username and keytab.

Global Flags:
  -v, --verbose[=false]: verbose output

unknown flag: --admin-name

Could this be a problem with the latest version?

Thanks,
Abhishek

Prediction service not recognizing DRF model as binomial

Have have a DRF model imported from h2o flow that when deployed to the prediction service, seems to not re recognized as a binary estimator in predict.js used by the service (the global variable isBinaryPrediction in predict.js always returns false). A more detailed explanation of the problem and my current workaround is posted as a stackoverflow question (and self-answer).

Passwords are logged in plain text by web server

Creating a new Steam user results in the new username and password logged in the server output. Even if the credentials were encrypted, they should never be logged anywhere, as this presents a significant security vulnerability. See screenshot below:

Steam isn't compatible with RHEL6

$ ./steam-launcher
2017/10/24 23:21:37 Launching Scoring Service...
2017/10/24 23:21:37 Launching Steam...
STEAM ./steam: /lib64/libc.so.6: version `GLIBC_2.14' not found (required by ./steam)
2017/10/24 23:21:37 failed launching steam: unexpectedly quit: exit status 1

There seems to be a hard dependency on glibc 2.14.
RHEL 6.9 boxes we have come with glibc 2.12:

$ sudo yum list installed | grep glibc
glibc.x86_64 2.12-1.209.el6_9.2 @rhel6-x86_64-2017q3
glibc-common.x86_64 2.12-1.209.el6_9.2 @rhel6-x86_64-2017q3
glibc-devel.x86_64 2.12-1.209.el6_9.2 @rhel6-x86_64-2017q3
glibc-headers.x86_64 2.12-1.209.el6_9.2 @rhel6-x86_64-2017q3

RHEL7 on the other hand comes with glibc 2.17 (at least on RHEL7.3 servers we have).

Any ideas how to workaround this?
Thank you.

could not deploy deepwater model

steam can't deploy deepwater models,is right?