Fabian Höring

Software Engineer @ Criteo

[Fabian Höring @f_hoering]

Check out our article written by
@MaximeVono, Julien Nigri and myself about Trusted Execution Environments (TEEs) for Privacy Sandbox proposals and some experiments we did with our workloads.

[Criteo Tech @CriteoEng]

Discover the power of Trusted Execution Environments (TEEs) in #PrivacySandbox! 🚀Our latest article explores how to address advertising use cases with privacy. Learn about Attribution Reporting API (ARA) and TEE mechanisms like remote attestation 👇 https://t.co/0eyAlOcMLd

[Criteo Tech @CriteoEng]

👀#mustread article by @f_hoering 👏👇

[Criteo @criteo]

We tested Google’s Topics API and have compiled preliminary results. Learn about our findings and recommendations for the future: https://t.co/RKjgjMQ8Ik

#AdTech

[Fabian Höring @f_hoering]

Some of my work of documenting those different tools for the @ApacheSpark community. Glad to see that this starts getting traction.
Also a link here to my recent @europython talk which covers the PySpark example with PEX and the motivations behind that.
https://t.co/ffH1CSXTao

[Matei Zaharia @matei_zaharia]

Some nice ongoing work to make it easier to ship various kinds of #Python environments with @ApacheSpark applications, as part of Project Zen: https://t.co/QpReMKUh1k

[Fabian Höring @f_hoering]

Our contributions at Criteo to the @MLflow project.

[Jean-Denis Lesage @jdlesage]

Yesterday, at the end of my talk at #DataAISummit , I presented 2 @mlflow plugins we develop. 1/ https://t.co/k10NVghya0. Run your MLProject in your hadoop cluster by adding "-b yarn" to your command line.

[Fabian Höring @f_hoering]

New model rolled out in Prod. It actually beats all Wide&Deep papers on our Criteo 1TB dataset.

[Criteo AI Lab @CriteoAILab]

What did it take to beat our production Engine ? A story of model composition written by our engineering manager @loic_le_bel available on @medium! → https://t.co/iB2cNMoodO #MachineLearning #AI

[Fabian Höring @f_hoering]

@chrisgugi And also 'export CLASSPATH=$(${HADOOP_HOME}/bin/hadoop classpath --glob)' as this seems not to be set

[Fabian Höring @f_hoering]

@chrisgugi Check_hadoop_env won't work in this case but it is just an env checker trying to read/write files with https://t.co/OFggoPm9m4.GFile api.

[Fabian Höring @f_hoering]

@chrisgugi With globally I mean ask the system admin to set those en vars for all users. But if you just want to try it out and the so files are available I think you can just use the env param in run_on_yarn or alternativly execute export LD_LIBRARY_PATH=.. in pre_script_hook.

[Fabian Höring @f_hoering]

@chrisgugi No. You need to do that on your own. Possible that you cam just use the env param or prehook param in run_on_yarn (at least for testing) to do it. We have it set globally on our cluster.

[Fabian Höring @f_hoering]

@chrisgugi You need to set up libhdfs on each hadoop node as tensorflow will try to connect. The pip tf-yarn package only needs to be installed on the gateway from where you launch the job.

[Fabian Höring @f_hoering]

@chrisgugi Can you create a github issue and post the stacktrace ?

[Criteo Tech @CriteoEng]

Learn how we build complex pipelines using Hadoop, Yarn, and MlFlow.

#Criteo #Engineering #DeepR #MachineLearning #coding #optimization #dataconfiguration

https://t.co/RT8BxHgyIX

[Fabian Höring @f_hoering]

@shirleman We use Cassandra as a backend for graphite for time series data. I didn't find a blogpost. Some info here:
https://t.co/Q7AcMh8VJ3

[Fabian Höring @f_hoering]

@shirleman We don't use Cassandra as an alternative to hdfs storage. We store 100 PB in our Hadoop cluster and use hdfs as a storage. Here is a blogpost with some insights about our Hadoop cluster https://t.co/24F3gdJVs4

[EuroPython @europython]

We just surpassed 1000 registrations.

What an amazing Python community we have.

See you tomorrow :-)

#EuroPython #Python

[Kowalski Nicolas @kowalski_nico]

Happy to share that we released tf-yarn 0.5.1, now compatible with #TensorFlow 2.2!

Tf-yarn allows you to run #TensorFlow code on #hadoop clusters in just a few lines of code.

https://t.co/wlZhCucevq

[Fabian Höring @f_hoering]

Direct link for my proposal https://t.co/1JV3VcC3SC

[Fabian Höring @f_hoering]

I'm happy to be a speaker at online EuroPython 2020 😀. My submission 'Building reproducible distributed applications at scale' was accepted. https://t.co/UGqXLoEtrb #python #EuroPython