8 ms·
Cloud AI Platform Pipelines
- jamesblonde 7y agoI have a couple of quick takes on this: * Remember kubeflow - the open-source ML platform? Well, on GCP, it doesn't look so open-source anymore. * It looks like TensorFlow Extended has been subsumed (killed off?) by this new managed platform. No Beam support - to be replaced by tf.data service? https://github.com/aaudiber/community/blob/rfc-data-service/rfcs/20200113-tf-data-service.md https://github.com/aaudiber/community/blob/rfc-data-service/...
- doomslice 7y agoThis looks like it's basically just an installer for Kubeflow -- this is not "managed" at all. And where did you get that TFX has been replaced by this? You can run Kubeflow pipelines (including those created by TFX) on this.
- jamesblonde 7y ago"Kubeflow pipelines SDX and TFX SDK ... Over time, these two SDK experiences will merge" from the article.
- doomslice 7y agoThat's just the SDK that they're saying is merging (with TFX being the one that "wins"). If you've used them you'll know they have huge overlap but TFX abstracts a ton of what it's actually doing.
- jamesblonde 7y agoTFX is just a SDK (with a few libraries - like data validation, model analysis, statistics). So, i don't know about saying it wins? The battle of ML pipeline ecosystems is the engine - not so much the API. It had been Beam vs Spark. Now, Google are changing tack and saying it is TensorFlow on Kubernetes with distributed processing vs Spark-based ML pipelines (that may use PyTorch/Tf for training).
- wjarek 7y agoI work on TFX. TFX started as an internal Google project [1] but over the years we've been steadily open sourcing more and more of it. Portability and interoperability are key focus areas for our project, more information in [2]. TFX OSS currently consists of libraries, components, and some binaries. It is up to the developer to pick the right level of abstraction for the task at hand. More information about this in [3]. Many of the ideas in TFX were inspired by an earlier production ML system at Google called Sibyl, more information about Sibyl in [4]. The two systems combined represent over a decade of production ML experience of many engineers and we hope other engineers will find it useful, too. Also of interest may be this overview blog post [5] by yours truly :) [1] ACM KDD '17 "TFX: A TensorFlow-Based Production-Scale Machine Learning Platform" https://dl.acm.org/doi/10.1145/3097983.3098021 https://dl.acm.org/doi/10.1145/3097983.3098021 [2] https://www.tensorflow.org/tfx/guide#portability_and_interoperability https://www.tensorflow.org/tfx/guide#portability_and_interop... [3] https://www.youtube.com/watch?v=zxd3Q2gdArY https://www.youtube.com/watch?v=zxd3Q2gdArY [4] https://www.youtube.com/watch?v=3SaZ5UAQrQM https://www.youtube.com/watch?v=3SaZ5UAQrQM [5] https://blog.tensorflow.org/2019/05/research-to-production-with-tfx-ml.html https://blog.tensorflow.org/2019/05/research-to-production-w...
- jamesblonde 7y agoI am getting downvoted here without anybody actually addressing my arguments - TFX/Beam being replaced by TFX/KubeflowPipelines and the new platform being plugged in tightly to Google's managed services (AI platform, BigQuery, GCS, etc). I didn't think my tone was negative - they are observations that nobody in this subthread is addressing. As a systems person, TFX to me was a layer on top of Beam to tie ML pipelines together. Now, Beam is gone - dataflow appears as a managed service you can use in your pipelines (see the figure on the blog post). TFX is being repackaged as the API for Kubeflow pipelines. That was my take and what i wanted to discuss.
- doomslice 7y agoFor what it's worth, I can't even downvote :) I don't know the _eventual_ direction of Kubeflow/TFX -- but in our TFX pipelines you still get to choose where it runs. From the docs: > Apache Beam is an open source, unified model for defining both batch and streaming data-parallel processing pipelines. TFX uses Apache Beam to implement data-parallel pipelines. The pipeline is then executed by one of Beam's supported distributed processing back-ends, which include Apache Flink, Apache Spark, Google Cloud Dataflow, and others. You can also choose to run it on Kubeflow itself.
- talolard 7y agoMy first thought was how long till it’s discontinued?
- mav3rick 7y agoLonger than the time it will take for the anti Google HN bubble to burst.
- dna_polymerase 7y agoI'm sorry but GP is exactly right. Google has a horrible track record for their products. They can't even profit from vendor-lockin, as they kill off all the products sooner or later leaving a lot of angry customers.
- snug 7y agoName the GCP products that have been killed off?
- rllin 7y agogoogle deployment manager. it has incredibly subpar support for other GCP services. all support interactions resort to them suggesting migration to Terraform. we do use it in production, but not without great headache. this isn't an explicit kill-off, but certainly purposefully offering bad support
- minimaxir 7y agoThis product appears to have a lot of overlap with Cloud Composer (https://cloud.google.com/composer https://cloud.google.com/composer) except with more of an AI focus. Does that mean Cloud Composer might get depreciated?
- fooker 7y agoOf course, this is Google. Everything overlaps with something can get shut down after a promo cycle.
- danols 7y agoThis argument is getting old for B2B products. Would you mind sharing a list of Google Cloud Products or even just Business to Business related products that have been shut down?
- tedivm 7y agoApp Maker, Google Cloud Print, Google Hangouts, Hire by Google, Fabric, Google Station, Google Correlate, Google Translator Toolkit, Google Fusion Tables, G Suite Training, Hangouts on Air, Google Cloud Messaging, Inbox by Google, Google URL Shortener, Google Realtime API, Google Site Search, Google Spaces, Picasa, Google Checkout. I got bored and stopped before finishing 2012, but you can go back and find more.
- danols 7y agoApp Maker is probably the worse on that killedbygoogle.com copy & paste list. Rest are B2B really?, the product evolved and got re-branded/consolidated or meh, was it really widely used anyway? Not a single one is a Google Cloud product. I am no Google fanboy and get frustrated by a lot of things they do. But I think the Google kills everything argument for B2B products is getting tiresome. Especially in a Google Cloud Platform context. Their actions on the Google Maps API absurd price hike and the recent GKE pricing structure change debacle is a whole other story and worth a lot of criticism.
- plexicle 7y agoThe pre-built pipeline components look interesting. I wish Cloud AI offered an API for the kind of results that I get when I use Google Lens. None of the ML offerings seem to come even close with labels.
- sologuardsman2 7y agoTheir Cloud Vision API [1] might be what you're looking for. [1] https://cloud.google.com/vision https://cloud.google.com/vision
- notlukesky 7y agoAlso take a look at: https://chooch.ai/ https://chooch.ai/
- plexicle 7y agoYeah that's the one I was talking about. It gives some labels about stuff in the image and it works well. But Google Lens is on a different level -- you can send it a picture of a breed or a snake and it will return species and subspecies. Can't find anything like that in the GCP offering.
- tixocloud 7y agoWhat sort of results do you get back from Cloud Vision API? The Google Lens example you mentioned is what I assumed it would do.
- plexicle 7y agoNothing near that specific unfortunately. Not much more than "snake" (and occasionally wildly inaccurate species). At least with the ~20 or so photos that I just tested with.
- tixocloud 7y agoLooks like you're not the only one who's interested: https://support.google.com/photos/thread/17424160?hl=en https://support.google.com/photos/thread/17424160?hl=en
- nancycut10 7y agohttps://www.hackerslist.co/ https://www.hackerslist.co/
- thefounder 7y agoJust make sure you build some abstractions and have a backup plan. Don't get locked-in their API. It may/will ruin your business.
- siganakis 7y agoCloud AI Platform Pipelines appear to use Kubeflow Pipelines on the backend, which is open source [1] and runs on Kubernetes. The Kubeflow team has invested a lot of time on making it simple to deploy across a variety of public clouds [2], [3]. If Google were to kill it, you could easily run it on any other hosted Kubernetes service. I haven't used Cloud AI Platform Pipelines, but have spent a lot of time working with Kubeflow Pipelines and its pretty great! [1] https://github.com/kubeflow/pipelines https://github.com/kubeflow/pipelines [2] https://www.kubeflow.org/docs/aws/ https://www.kubeflow.org/docs/aws/ (Deploy to AWS) [3] https://www.kubeflow.org/docs/azure/ https://www.kubeflow.org/docs/azure/ (Deploy to Azure)
- mandelken 7y agoI wonder how this might work with NLP workflows that use a custom embedding? I have not found a documentation or blog describing such a memory intensive step in the pipeline without running a custom instance/docker. In that case I can just run the whole model on that instance..
- mycpuorg 7y agoTrying to understand how is any this is new or different from AWS SageMaker or even Azure ML? IIUC AWS has way better tools/services to integrate with. Most Google's Cloud Services have been very flaky and inconsistent.