AWS Lambda is a serverless computing platform that lets users run code without provisioning or managing servers. With Lambda, users can run code for virtually any type of app or backend service—all with zero administration. It takes of requirements to run and scale code with high availability.
$NaN
Per 1 ms
Azure Machine Learning
Score 8.2 out of 10
N/A
Microsoft's Azure Machine Learning is and end-to-end data science and analytics solution that helps professional data scientists to prepare data, develop experiments, and deploy models in the cloud. It replaces the Azure Machine Learning Workbench.
$0
per month
TensorFlow
Score 7.7 out of 10
N/A
TensorFlow is an open-source machine learning software library for numerical computation using data flow graphs. It was originally developed by Google.
Most of the machine learning platforms these days support integration with R and Python libraries. So, the use of reusable libraries is not an issue. TensorFlow performs well in cloud hosting and support for GPU/TPU. However, where it lacks compared to Azure is a graphical …
Lambda excels at event-driven, short-lived tasks, such as processing files or building simple APIs. However, it's less ideal for long-running, computationally intensive, or applications that rely on carrying the state between jobs. Cold starts and constant load can easily balloon the costs.
TensorFlow is great for most deep learning purposes. This is especially true in two domains: 1. Computer vision: image classification, object detection and image generation via generative adversarial networks 2. Natural language processing: text classification and generation. The good community support often means that a lot of off-the-shelf models can be used to prove a concept or test an idea quickly. That, and Google's promotion of Colab means that ideas can be shared quite freely. Training, visualizing and debugging models is very easy in TensorFlow, compared to other platforms (especially the good old Caffe days). In terms of productionizing, it's a bit of a mixed bag. In our case, most of our feature building is performed via Apache Spark. This means having to convert Parquet (columnar optimized) files to a TensorFlow friendly format i.e., protobufs. The lack of good JVM bindings mean that our projects end up being a mix of Python and Scala. This makes it hard to reuse some of the tooling and support we wrote in Scala. This is where MXNet shines better (though its Scala API could do with more work).
User friendliness: This is by far the most user friendly tool I've seen in analytics. You don't need to know how to code at all! Just create a few blocks, connect a few lines and you are capable of running a boosted decision tree with a very high R squared!
Speed: Azure ML is a cloud based tool, so processing is not made with your computer, making the reliability and speed top notch!
Cost: If you don't know how to code, this is by far the cheapest machine learning tool out there. I believe it costs less than $15/month. If you know how to code, then R is free.
Connectivity: It is super easy to embed R or Python codes on Azure ML. So if you want to do more advanced stuff, or use a model that is not yet available on Azure ML, you can simply paste the code on R or Python there!
Microsoft environment: Many many companies rely on the Microsoft suite. And Azure ML connects perfectly with Excel, CSV and Access files.
Developing test cases for Lambda functions can be difficult. For functions that require some sort of input it can be tough to develop the proper payload and event for a test.
For the uninitiated, deploying functions with Infrastructure as Code tools can be a challenging undertaking.
Logging the output of a function feels disjointed from running the function in the console. A tighter integration with operational logging would be appreciated, perhaps being able to view function logs from the Lambda console instead of having to navigate over to CloudWatch.
Sometimes its difficult to determine the correct permissions needed for Lambda execution from other AWS services.
Theano is perhaps a bit faster and eats up less memory than TensorFlow on a given GPU, perhaps due to element-wise ops. Tensorflow wins for multi-GPU and “compilation” time.
I give it a seven is usability because it's AWS. Their UI's are always clunkier than the competition and their documentation is rather cumbersome. There's SO MUCH to dig through and it's a gamble if you actually end up finding the corresponding info if it will actually help. Like I said before, going to google with a specific problem is likely a better route because AWS is quite ubiquitous and chances are you're not the first to encounter the problem. That being said, using SAM (Serverless application model) and it's SAM Local environment makes running local instances of your Lambdas in dev environments painless and quite fun. Using Nodejs + Lambda + SAM Local + VS Code debugger = AWESOME.
Amazon consistently provides comprehensive and easy-to-parse documentation of all AWS features and services. Most development team members find what they need with a quick internet search of the AWS documentation available online. If you need advanced support, though, you might need to engage an AWS engineer, and that could be an unexpected (or unwelcome) expense.
Community support for TensorFlow is great. There's a huge community that truly loves the platform and there are many examples of development in TensorFlow. Often, when a new good technique is published, there will be a TensorFlow implementation not long after. This makes it quick to ally the latest techniques from academia straight to production-grade systems. Tooling around TensorFlow is also good. TensorBoard has been such a useful tool, I can't imagine how hard it would be to debug a deep neural network gone wrong without TensorBoard.
AWS Lambda is good for short running functions, and ideally in response to events within AWS. Google App Engine is a more robust environment which can have complex code running for long periods of time, and across more than one instance of hardware. Google App Engine allows for both front-end and back-end infrastructure, while AWS Lambda is only for small back-end functions
It is easier to learn, it has a very cost effective license for use, it has native build and created for Azure cloud services, and that makes it perfect when compared against the alternatives. As a Microsoft tool, it has been built to contain many visual features and improved usability even for non-specialist users.
Keras is built on top of TensorFlow, but it is much simpler to use and more Python style friendly, so if you don't want to focus on too many details or control and not focus on some advanced features, Keras is one of the best options, but as far as if you want to dig into more, for sure TensorFlow is the right choice
Positive - Only paying for when code is run, unlike virtual machines where you pay always regardless of processing power usage.
Positive - Scalability and accommodating larger amounts of demand is much cheaper. Instead of scaling up virtual machines and increasing the prices you pay for that, you are just increasing the number of times your lambda function is run.
Negative - Debugging/troubleshooting, and developing for lambda functions take a bit more time to get used to, and migrating code from virtual machines and normal processes to Lambda functions can take a bit of time.
Productivity: Instead of coding and recoding, Azure ML helped my organization to get to meaningful results faster;
Cost: Azure ML can save hundreds (or even thousands) of dollars for an organization, since the license costs around $15/month per seat.
Focus on insights and not on statistics: Since running a model is so easy, analysts can focus more on recommendations and insights, rather than statistical details
