We are working on an ML project (Computer Vision specifically) and we are in the process of productionizing our models. What kind of tools do you use and what's your workflow for integrating, testing and deploying your models? Do you have any suggestions or tips about what to do and what to avoid?
Having put many models into production in an almost real time environment (ad servers that need predictions in < 10 ms) I would say a large portion of what you need depends on your production requirements. The system I work on is very high volume 100k request / sec and we have massive amounts of data so it complicates everything.
First, I would highly recommend wrapping your ML models in some kind of microservice. Depending on your production requirements and if the ML is in Python a fairly simple Flask/Sanic web server should be sufficient. This is great because you can leave all your feature transformation code as is in Python.
If your production environment has very low latency requirements you are going to have some work cut out for you. You'll most likely have to rewrite all your transformation code in a faster language like Go or Java. You might also need to implement the inference code as well to get the speed you need. This adds considerable time and adds a ton of surface error for potential insidious bugs. The ML will still make predictions, but they will be wrong or very slightly wrong.
Because I'm working with larges of amounts of data and my source of truth is Parquet logs in S3, the pipelines start with Spark. We do as much data wrangling as possible in Spark to get things into a manageable size to create our train/dev/test sets. This data gets uploaded to S3.
The datasets are then trained on EC2 instances using Pandas & sklearn. When everything is fully automated the Spark job will push a message onto an SQS queue with the S3 path of the fresh dataset. An EC2 instance will be polling that queue and pull down the data and train a new model.
The final result of training my case is a text or binary model file that goes back up to S3. Our prediction microservice polls an S3 bucket and pulls down any updated model files and swaps out the running models.
Tips:
1. Instrument everything! Hopefully you have something like graphite/datadog/prometheus in place already, but you'll want metrics on your predictions.
2. Exception tracking on everything especially anything in your model creation pipeline. Sentry or something like that.
3. Try and keep everything as simple as possible.
Thanks for this; I really appreciate the detail here. There seems to be a lack of these kinds of explanations around.
One piece did give me a bit of surprise:
You might also need to implement the inference code as well to get the speed you need
I've never had the super-low-latency requirements you have, but as you point out this seems amazingly error-prone. I'd love to hear anything else you can share about the cost-benefit analysis you do before deciding to go this route, and if there's any tools or languages you've had a better time with than others for this.
I have similar experience with GP, which is how I ended up writing Gorgonia (https://github.com/gorgonia/gorgonia). Serialization of models is easy: .npy files are very good formats though I know a number of people who will disagree - I find they typically prefer pb as a serialization format, which I think is better for over-the-wire not for storage
Also like GP, I used to work in advertising. The low latency bits are because RTB servers want you to respond within a certain amount of time. When I was in advertising, our solution was to precalculate a whole bunch of things, and throw them into redis. It was then a simple lookup of the hash of a vector (for bidding related stuff)
Yeah, the requirements are pretty different than most Data Science teams, especially the very low latency requirements.
The constraints force us to use simple models like linear regression and logistic regression some of the time or at least as a version 1. The inference here is straightforward, multiply and add then take the sigmoid if doing logistic regression.
What we tried to do initially was integrate with C/C++ APIs where possible. We ran into some issues with speed and bugs doing this though, which is why we wrote the inference ourselves. Where we had issues was calling the XGBoost C API from Go. It was extra overhead and too slow. In our benchmarks our implementation in pure Go was many times faster than calling the C API. We also found the multithreaded version to be slower than the single threaded. We found this to be true when calling XGBoost from Java and from Go. We also found this to be true in our own inference implementation it was always faster to walk the trees in a single go routine rather than create some number of worker go routines to walk the trees in parallel.
We were very careful implementing the inference ourselves to make sure the predictions matched. What we did to verify this was create a few toy datasets of about 100 rows with sklearn's make_classification function. We then trained a model using the reference implementation, saved the predictions and the model. We then loaded this model into our implementation and made predictions on the same dataset. We wrote unit tests to compare the predictions and make sure they are the same within some delta. We were able to get our implementation to be within 1e-7 of the reference implementation, in this specific case XGBoost. It was actually more time consuming to deal with parsing the inconsistent JSON model output of XGBoost than it was to implement the GBDT inference algorithm. We also had to make a slight change to the XGBoost code to write out floats to 18 decimal places when writing out the JSON model in order to get the two implementations to match.
Yeah, I had a very hard time believing that the multithreaded approach would be slower. Its so counterintuitive since at first blush it seems that walking N trees is an embarrassingly parallel problem. I tested up to 1000 trees and single threaded was still faster. I'm sure at some point the multithreaded approach will win out, but its beyond the number of trees and max depth we are using.
I've half-convinced myself it's because we're talking about GBM's and not Random Forests (where my mind goes first). One of the smart things about XGBoost is parallelizing training by multithreading the variable selection at each node, but that doesn't apply to inference; I imagine you gotta predict trees sequentially since each takes the previous output as an input? Now I wonder what those extra threads were even doing...
Each tree can be traversed independently. Each tree is traversed until a leaf node is reached. Those leaf values are summed for all the trees. The sum of all the leafs plus a base score is then returned. In the case of binary classification that sum is passed through a sigmoid function. For linear regression the sum is returned.
I wrote the inference step of XGBoost in Go. It will make predictions after loading in an XGBoost JSON model. Writing the training portion of XGBoost would be much harder.
Shameless plug: I cover this topic in my book, Reactive Machine Learning Systems: https://www.manning.com/books/reactive-machine-learning-syst... In particular, chapters 7 and 8 cover the design of model servers and model microservices, from publishing through to serving. And chapter 9 focuses on how build CI/CD pipelines for ML systems.
Definitely lots more options than I even had space to cover in a single book, though. This is an area where the tooling continues to rapidly improve.
I have deployed ML algorithms into production including computer vision and data science models. From my experience, it's based on the application and cost we are comfortable with.
1. Keras/Tensorflow based algorithm(Applicable for any compute intensive or GPU-capable algorithm): Deployed the method (as a flask service) inside a Docker container along with a queueing system(for reliability w/ redis). We can now decide the server type and the kind of orchestration tool we can use to manage these containers. Following are some options for it,
a. ECS on AWS
b. Kubernetes
c. Docker Swarm
d. custom orchestration tool
2. If your ML model is a simpler with a small (enough) model size, then using Lambdas on AWS would also work. This can provide high throughput and low cost per request if your computation time isn't very high.
Tips:
i. Have the memory flush in the code after the service is used so that there is no memory leak.
ii. You can use tools such as htop to understand the memory usage.
iii. Regarding system performance, you can use prometheus to gather stats along with grafana dashboard to view them.
I consider CI/CD essential for achieving a seamless workflow for a data scientist, and after having faced the same problems ourselves, our team and I have been working on Datmo, a tool to help companies more easily and cost-effectively deploy + manage their models in production.
3. Another important component for CI/CD is the integration of tools like Airflow in order to schedule and monitor workflows. This helps us in deploying newly trained models.
We faced the issue of building CV models a lot as grad students but at that time reliability wasn't really something we had to solve for. Once we had to implement them for industrial applications we found we had to ensure there was reproducibility and versioning throughout the process. Now that we have put a few computer vision algorithms into production we decided to architect our code in the following way.
1) training code is written with normal packages (we tend to prefer keras with a tensorflow backend), these are trained on our own GPUs since this is often the cheapest and are done in bulk --> side note is that TPUs/GPUs may one day be better but certainly too expensive currently.
2) prediction / inference code will also be written in the similar packages but will be tied with the final weights files that we get from the model
3) deployment code -- in order to enable a reliable system we have used celery for distributed queueing and converting our prediction functions into celery tasks which can then me passed to workers that can process and return the result to an API endpoint. This allows us to scale our workloads as needed with throughput (depending on request requirements)
This architecture allows us to test during training time using our training code and validation sets, while also enabling testing of different models versions through our prediction APIs. We often would just write scripts for testing that we then run via our CI/CD workflow.
Tip: keep your code as simple as possible and don't rewrite your code unless your throughput requirements mandate it. good error handling will go a long ways here and likely will be easier if written in a language you're most familiar with.
I don't work on computer vision specifically, and can generally suffer a few seconds of latency, but we have a two-stage process that I think applies fairly generally
1. all lowish-level production code (importing, transforming, modeling) is written in packages, which are thoroughly unit-tested with a CI system
2. that code is wrapped up into docker container(s) in separate repositories, which are built and integration-tested with CI. In addition to the Dockerfile and any testing scripts, there's usually a single code file here which handles I/O specifics, API endpoints, and primarily calls code from the package
This works well with R or Python and should work with others; we use Gitlab for the free private repos and awesome built-in CI.
This doesn't cover where the data or models are stored, but that varies more per-project for us. Lately we've been using Pachyderm and loving it, but you can get pretty far with a postgres instance for data and storing trained model objects in S3/GCS.
Hi i'm founder of https://bitbank.nz a crypto currency live prediction dashboard/API/bulk data service.
Our system streams in market data from exchanges, creates forecasts with python/sk-learn and displays the data, we also have background processes that updates our accuracy over time once real data is available.
We test our code with the normal python unit/integration/end to end testing methods locally with local copies of all of our components (except firebase for live UI updates, we don't have a dev version of that yet just use live forecast data when testing the UI charts/display), would probably get expensive/cumbersome to setup dev environments with local firebases in them.
with deployment we simply ssh into machines, git pull latest code and supervisorctl restart so its fairly low tech, the forecaster has a roughly minute outage when we deploy new models because there is a decent process that computes and caches a data structure of historical features for use in the forecaster.
In terms of maintaining a reliable online stream of data input, feature computation/prediction pipeline we run the code under supervisor aswell as running a manager process under supervisor, that manager process checks if conditions are turning bad (OOM/no progress updates by the forecaster) and restarts things if anything goes wrong.
For testing we also use the standard training/test data split when running backtesting/machine learning optimisation algorithms to train parameters of the algorithm.
If things perform better on training and test data over a long enough time period to build confidence then we will deploy a new model.
Using graphite/graphana to monitor prediction accuracy over time is a good idea as mentioned already :) and some kind of alerting/monitoring if things go down.
I'm a statistician by trade so I mostly do prototype work. As far as building models my key workflow is using a R and Rstudio. The biggest issue is data management. I suggest a good API or wrapper for a data source that has all of the ETL already done for the most part. R connects very well to most database systems. RStudio makes development easier with connectivity to GitHub or other popular version control systems.
As far as putting into production I'm not as familiar. Yet I hear that a good Python workflow would probably work best.
For private repos, I'd say gitlab is an order of magnitude (or two) better than bitbucket. Or, it clearly was 2 years ago, and while I haven't kept up with bitbucket, gitlab has improved by leaps and bounds in those two years.
The killer features for me are nested subgroups (which bitbucket may have, but github does not) and a really awesome CI system with a generous free tier (2000 minutes/month). For R packages, we have it setup very similar to github + travis (devtools::check() every push), and for deployable bits we have it build containers and run integration tests on them. Super impressed with all we get for free there.
Also per the production environment. The key is having the same machine learning libraries available in both the development and production so you can plug the model in with little problems with dependencies. For this reason most folks that will go into production, particularly web applications, will tend to both develop and implement in Python, Java, etc.
Disclaimer: I compete in this space and may compete with your internal team, your cloud vendor, or something you are interested in.
FWIW: Production is an overloaded term. CI may not even be applicable here. Say you're doing batch inference where you need to run jobs every 24 hours on a large amount of data: That might be tied to some cron job.
That being said you could use a CI system for that in theory.
There are also other factors here: What other kind of things do you want to track? Experiments results? Wrong results by your machine learning algorithm?
Concisely: What kind of deployment requirements do you have and what are your goals?
If you are edoing real time, what does "deployment" even mean? Are you serving in real time via a rest api? Are you doing streaming? What are your throughput requirements? What about latency? Is that even hooked up to a CI system?
Something that is vaguely related: How do you test the accuracy of different models across your cluster? Say you want to do a self deployment, what if you want to tie that to say: a workspace where you produced the results?
Is that hooked up to a CI system? If so, what's your use case?
Then there's that common hand off from data scientist to production, what does that look like? A sibling thread mentioned some of these things.
If anyone else is curious about this stuff, we deploy deep learning models in locked down environments both on kubernetes as well as touching hadoop clusters.
Happy to answer questions.
Shameless plug (I'm head of product at Algorithmia): we've written a lot of algorithms ourselves including many computer vision algos [1] that are published publicly for use. The underlying platform at Algorithmia is able to run production models at scale and puts a simple REST API in front of all of the algorithms so developers can easily call them from your production apps. The algorithms you see in the marketplace are publicly published but the platform runs private algorithms for software teams as well. Check it out: [2]
We use DeepDetect everywhere. I started coding it up in 2015 out of immediate need for my customers in production at the time and waited for something else to come out. But until now, we've sticked to it and customers ask for it when they see us using the pre-configured pipelines. Some run their production with the DD server. So three years later we're still putting down some time on improving it, and support for caffe2 is coming up among other things. It already has built in semantic search due to customer demand. Welcome any feedback an comments btw. Also we're hiring, contact us, we're at http://jolibrain.com/
I try to do as much of the transformations as possible in tensorflow using the Datasets API so that I don't have to write them in another language/system in production.
I had tried to do the same. However given the CPU/GPU imbalance on AWS GPU instances I have resorted to building a fully "rendered" training set and doing all the transforms in spark.
If you use Algorithmia.com you can add your model in the language of your choice (on GPUs if you want) and it will do all of the Devops and give you an API end point. You get free credits at sign up and quite a few each month for testing.
First, I would highly recommend wrapping your ML models in some kind of microservice. Depending on your production requirements and if the ML is in Python a fairly simple Flask/Sanic web server should be sufficient. This is great because you can leave all your feature transformation code as is in Python.
If your production environment has very low latency requirements you are going to have some work cut out for you. You'll most likely have to rewrite all your transformation code in a faster language like Go or Java. You might also need to implement the inference code as well to get the speed you need. This adds considerable time and adds a ton of surface error for potential insidious bugs. The ML will still make predictions, but they will be wrong or very slightly wrong.
Because I'm working with larges of amounts of data and my source of truth is Parquet logs in S3, the pipelines start with Spark. We do as much data wrangling as possible in Spark to get things into a manageable size to create our train/dev/test sets. This data gets uploaded to S3.
The datasets are then trained on EC2 instances using Pandas & sklearn. When everything is fully automated the Spark job will push a message onto an SQS queue with the S3 path of the fresh dataset. An EC2 instance will be polling that queue and pull down the data and train a new model.
The final result of training my case is a text or binary model file that goes back up to S3. Our prediction microservice polls an S3 bucket and pulls down any updated model files and swaps out the running models.
Tips: