Microsoft Azure Machine Learning Studio Reviews

A project was handed to us before we came to this new client, which involved running a machine learning experiment within ML Studio. The good thing about the solution is the entire workflow can be easily managed in ML Studio because you can track and tag datasets, different pipelines, and multiple transformations. You can add custom code to any of the transformation bits, so it's very flexible in how you design your experiments. You can either design a pipeline or run notebooks. You can do many things, and it's very flexible for many use cases.

ML Studio is very easy to maintain. It's also very portable because it has ARM templates to export to replicate your experiments in separate environments. That's useful if you move an experiment to a different resource group because you want to run a new experiment. It has a strong role-based access control that helps you keep track of who's accessing what, and it has a very good data lineage tool that allows you to version and understand each of the experiments and their results. You have a very good track of everything, and you can easily distinguish between experiments and execution times and which parts where the pipelines are failing. ML Studio gives you a lot of identifiability for each one of the components of your entire experiment.

While ML Studio does give you the ability to run a lot of transformations, it struggles when the transformations are a bit more complex, when your entire process is transformation-heavy, and when your datasets need to be distributed or parallel processed. While it offers you the capability of running distributed computing, it relies on the user to configure it. It does not do it automatically as Databricks would. It is up to the user to maximize ML Studio's use. Still, suppose you do not preemptively configure it to run everything in distributed compute or parallel jobs. In that case, it will just provision a single compute cluster and take longer than other solutions that do that automatically. ML Studio relies on user configuration to run parallel or distributed jobs. When you are new and trying to experiment with it, it could make your workflows much more costly and longer than they should be.

One or two engineers can easily maintain ML Studio without much hassle.

ML Studio's pricing becomes a numbers game. When you're trying to run isolated experiments with simple datasets that are easily tracked, ML Studio does a very good job with its on-demand pricing. At the same time, provisioning the solution and some other internal tools might not be cost-optimized. It might just be directly provisioned from infrastructure direct cost. As your data scales and grows and your transformations become more complex, your cost will probably skyrocket because it will do nothing natively to help you save on that end. Other platforms help you run jobs and allow you to run them distributed with a simple configuration from the UI rather than having the optimized code to do so.

Microsoft Azure Machine Learning Studio is very robust for tracking simple experiments. But it falls short when you run when you want to build an entire machine learning framework on top of it. I rate it a seven out of ten.

We primarily use this solution for data analytics and model building.

The interface is very intuitive.

It is very well organized and the components can be utilized through drag-and-drop.

The data preparation capabilities need to be improved. Using this product, I can not prepare the data very much and this is a bottleneck in machine learning.

There are some features that are not supported, so I have to use either Python or R to accomplish these tasks.

I have been working with the Azure Machine Learning Studio for between six and seven years.

Up to this point, we have not faced much in terms of issues with stability.

Scalability-wise, we have not had to deal with any limitations. The only problem is that when certain options are not there, we have to use Python or R to handle those tasks.

We have not faced any problems so I have not spoken with technical support.

The initial setup is very straightforward. It is not difficult to do.

I feel that this is a great solution. Even for people from the business side, this is a very good product. It is so intuitive that all of the information is there. The interface takes care of the most complex part, which has to do with the modeling. 

I would rate this solution a nine out of ten.

To create quick data analytic experiments, without incurring the time and cost of spinning up servers, setting up Hadoop, etc. 

Although MLS makes it very easy to deploy the resulting machine-learning models via REST API, I primarily use MLS as a means to quickly spin up experiments and create proof of concept models.

Not widely adopted at my old workplace, I only used this to create quick proofs of concept to try to convince management of the viability of a project.

MLS allows me to set up data experiments by running through various regression and other machine learning algorithms, with different data cleaning and treatment tools. All of this can be achieved via drag and drop, and a few clicks of the mouse.

The easy drag and drop can create simple data science experiments. Low barrier to entry allows large number of candidates get started.

The graphical nature of the output makes it very easy to create PowerPoint reports as well.

Enable creating ensemble models easier, adding more machine learning algorithms.

Out of about 150-plus MLS experiments I have done, maybe two or three bugged out. Interestingly enough, those are the ones I can’t delete out of the account.

Scalability, in terms of running experiments concurrently: Good. At max, I was able to run three different experiments concurrently.

Scalability in terms of deploying models: Unknown, I never deployed on Azure.  But I would guess REST API could probably easily handle a few K worth of hits per second, since that is how Microsoft is going to get paid.

Never used it.

The only other solution beyond this would be standard tools used by data scientists, like R, Python, etc. All of these would have a fairly high barrier to entry, requiring programming experience. The main selling point of MLS is the low barrier to entry, where even tech-savvy business people can use it.

Simple. Create MLS live account (preferably paid ones), open MLS, done.

Caveat: Different organizations have different attitudes towards cloud use, especially with sensitive data. At Bridgestone, the hardest part was getting corporate approval to allow me to upload heavily treated, sensitive data to a cloud platform.

To use MLS is fairly cheap. Even the paid account is something like $20/month,  unless you are provisioning large numbers of VMs for a Hadoop cluster.

The main MS makes money with this solution is forcing the user to deploy their model on REST API, and being charged each time the API is accessed. There are several pricing tiers for the API.

If you do not use the API, then value of MLS is to create rapid experiments ($20/month). The resulting model is not exportable to use, thus you’ll have to recreate the algorithms in either R or Python, which is what I did. MLS results gave me a direction to work with, the actual work is mostly done in R and Python outside of MLS.

R and Python.

Python + Pandas + scikit-learn: 

Pros: 

• scikit-learn offers better performance for extremely large data sets
• Large-data manipulation tools
• Fairly good set of ML algorithms

Cons:

• High barrier to entry, in terms of skill and knowledge
• Fairly labor intensive to create large number of experiments

R + caret:

Pros:

• Very good amount of ML algorithms (so many it may cause paralysis from too much choice, 200-plus algorithms)
• Good performance, unless the data set is extremely large

Cons:

• High barrier to entry
• Data manipulation is a pain, you probably want to use another tool to pre-treat the data before loading it into R dataframes

For data science professionals or programmers I would rate this solution a four out of 10. A major feature is missing: creating ensemble models. This can be achieved with the tool, but it's clumsy and slow.

For marketing or business professionals I would rate it an eight out of 10. It has a low barrier to entry, and can quickly create models that can be used for proof of concept and justify further investment in a full data science or Big Data project.

R and Python, in my mind, are still the way to go for a true data science/predictive analysis project. MLS's value is the ease of use and low barrier to entry. If one is not a programmer or statistician, MLS is a good way to get a project started, create a proof of concept.

My primary use case is for supervised and unsupervised learning models.

The most valuable feature is its compatibility with Tensorflow.

In the future, I would like to see more AI consultation like image and video classification, and improvement in the presentation of data.

I've been using this solution for a year.

The stability is questionable, given that Microsoft will be retiring the classic version of this product in 2024, and it's unclear how this will affect projects created on the classic version.

This solution is scalable.

The initial setup was straightforward, though you do need some experience with Azure administration in order to install it.

I evaluated Amazon SageMaker, which is a bit more advanced than Azure Machine Learning, with more functionalities and only a slightly higher price.

I would rate this solution as seven out of ten.

We're using the solution in order to give the customer a 360 degree view. Also, we use it if clients want to do machine learning with AI at a more reasonable cost.

Right now, we are just testing the customer insights from Microsoft.

The UI is very user-friendly and that AI is easy to use.

Usually, we also use the machine learning studio to build up the data logistics in machine learning.

On the customer side, the solution should do more to push companion marketing.

When you use different Microsoft tools, there are different pricing metrics. It doesn't make sense. The pricing metrics are quire difficult to understand and should be either clarified or simplified. It would help us sell the solution to customers.

The solution should simplify switching between platforms in the studio.

I've been dealing with the solution for two years.

I've only used the solution a couple of times. I haven't noticed any bugs and when I used it, it worked quite smoothly.

I don't have enough knowledge about the solution's scalability to be able to comment on it. Right now, we have about 5,000-6,000 users on the solution. Most are data scientists, and IT admins.

I've personally been in touch with technical support and I found them quite helpful.

I've only ever worked with Microsoft Azure. We didn't previously use a different solution.

The initial setup is very straightforward.

Our clients do the implementation with the help fo consultants like us.

The pricing and licensing are difficult to explain to clients. Their rationale for what things cost and why are not easy to explain.

I'm a consultant. Our company is partners with Microsoft.

Users will find it easy to get into Azure. Even if they aren't always in touch with Azure, they'll find themselves in touch with the dynamic field. Users have to get into Azure because once they get into the cloud, they should have some basic understanding of Azure itself.

I'd rate the solution eight out of ten. However, I don't know their competitors, so I can't really compare them to others on the market.

I wouldn't say it's necessarily about liking everything about the platform entirely. It's more about what do we want? In terms of machine learning, there are times that we have to get into it and customize it, etc. We can use the ready-made models that are available without really having to code encrypt them with our bitcoin code — our model doesn't need to be too complex. Deployments and everything, in general, can be automated from a CI/CD perspective as well.

I really can't see where it needs much improvement. My experience is only half-matured and is still maturing.

I don't think we have reached the stage where the customer has enough cohesion to really complain about anything. Also, a Microsoft team is personally involved which really simplifies the process.

In the machine learning world, when you are defining the model, typically people go for an interesting library of algorithms that are available. It's an imperfect scenario. The world is not as ideal as we think: how we draw a mathematical or theoretical formula is not exactly as it seems. With encryption, this uncertainty is actually much higher — that's why you need to tweak your mathematical formula or completely customize it. For this reason, my team has a development platform where they can customize code when it fails.

I have been using this solution since June.

Regarding the stability and scalability — so far so good; however, we're still exploring quite a bit. It's too early to really comment because the customer has already paid. They've just started their journey. We are yet to explore exactly what and how they want to use it. 

So far, we haven't had a situation where we have needed to raise a ticket for support on a technical front.

Currently, we're handling any issues internally because we're still in the initiation stage. It's going to take some time for us to really get our hands into it, but so far it's been a really good experience. Based on various conversations that I was part of, I think our customer really appreciates the support coming from our people.

 Compared to similar solutions, Microsoft Azure Machine Learning Studio is quite new so the initial setup wasn't much of a challenge. The data processor can pose a bit of a challenge, but the real complexity is determined by the skill of the implementation team.

I would Definitely recommend Azure Machine Learning Studio — no doubt about it, it's a full-contact solution. Having said that, it really depends on the customer's appetite and what they're comfortable with. For example, I have interacted with people who prefer a basic Google cloud platform — from an AML perspective, they just feel like it's primarily Google. Not because of AML per se, it's more from a data storage perspective, which in this case, works better.

Personally, I come from a VFA site in the financial sector. Over there, the customers are really conscious about hosting their station or their data, especially on the cloud. Typically, they are very restricted because they are not comfortable hosting customer data on the cloud. This is where I think Azure or Google or even AWS fall short — they don't play any role there. Because of this, people actually customize their solutions or model them to fit their custom sites and customer-based solutions. 

Overall, I would give this solution a rating of seven. It's a great option if you are fairly new and don't want to write too much code. As long as the model is not too complex, it's a pretty easy solution to roll out.

We're setting up the environment for our data science and IT project. It is a protected environment for protected data. So, there's a lot of architecturing in this solution.

Their support is helpful.

It is not easy. It is a complex solution. It takes some time to get exposed to all the concepts. We're trying to have a CI/CD pipeline to deploy a machine learning model using negative actions. It was not easy. The components that we're using might have something to do with this.

I don't know about its stability yet. We're facing some issues, and we are approaching the product team for help. It might also have something to do with our environment setup. Our environment is inside V-Net, and we have a lot of security requirements.

We're working with their team to resolve the issues. Having someone to assist you makes it easier. We have someone at Microsoft to help us with it. They're very helpful.

I have worked a little bit with Open Source.

We are only testing, and we have to be very careful of the restrictions. I'm a little bit aware of the issues about ML Ops, and I am trying to see if Azure Machine Learning Studio can address those issues. For now, I would rate it a seven out of 10. I have to explore it more.

We initially moved to this solution because our company needed to complete a system upgrade. We had to move the Db2 data to a AS400 system. 

Their web interface is good but the on-prem site interface is outdated. This solution could be improved if they could integrate the data pipeline scheduling part for their interface. When we are scheduling, they provide only one exclusion per day in the initial scheduling. We then have to configure it through the Linux front jobs if we want a high value job. It would help us and our customers if this was possible from the initial interface itself.

I have been using this solution for a few months. 

This is a stable solution. 

We have had limited engagement with the customer support team but when we have needed their help, they were helpful. 

Positive

The infrastructure and the software configuration part was done by one of my teammates. It was completed in two working days. We did experience some issues with the board communications which extended the time to complete the setup. This was only for the DataStage installation which is one of many components of this solution.

I would advise others to identify the communication between servers and the client tools correctly as well as the user allocation for those. If you are working from a client environment and connecting to the server, it is important that the configuration is done correctly.

I would rate this solution an eight out of ten.