Amazon SageMaker Reviews

I sit with our stakeholders, analyze the data, and then we come up with the solution, whether it goes to traditional machine learning or AI. We are dealing with one of the financial companies for document pre-processing. Normally they are opening the document and doing manual work, such as identifying the loan numbers, ID numbers, and all of that. We are using AI to scrape the document and pass our system prompt to automatically take the IDs in the UI itself. You just need to upload the document to the system. Amazon SageMaker will be analyzing that document and then coming up with the IDs and what IDs are present in the document.

There will be many features in Amazon SageMaker itself, but we don't know whether the feature is there or not, particularly the documentation part. Whatever the new releases will be, they will not post very fast.

It is very easy to deploy Amazon SageMaker. The documentation is also very good.

It is good because we are able to collaborate with our notebooks. At a time we can develop simultaneously and work on different use cases in the same notebook itself.

It takes some work. We need to refer to the documentation. The documentation is good regarding what other providers we are able to connect with. Out of five, I can say 3.5.

I have been using this for three years now.

Amazon SageMaker is definitely stable.

It depends upon the model, basically. On average, the normal models take around 10 to 15 minutes of time.

It takes some amount of time for them to solve the ticket.

Comparatively, GCP offers very low cost when compared to Amazon SageMaker. People are moving from Amazon SageMaker to GCP because of the cost constraints. The same thing will be offered at a very less price in GCP. In AWS, it is more costly.

It is very easy and, out of five, I would rate it 4.5.

The implementation team definitely helped with this solution.

Amazon SageMaker definitely provides ROI. Previously we benchmarked it at around 9.8% or something like that.

In our data science projects, we have a whole range of uses with Amazon SageMaker. We use it for anything from predictive asset health through to computer vision, along with Bedrock and some of the new AI services.

We don't use Amazon SageMaker for AI; we use Bedrock and other tools.

The tangible benefits we have observed from using Amazon SageMaker include improved time to insight and generally the common stack that is easier to support over time.

The product has been stable and scalable.

Their experience could still improve around scaling for ops and scaling operations. There aren't many people that know how to do that effectively. We're not really looking for support so much on how to solve problems anymore; it was more about how to scale and manage it across our different data science teams.

They could add features such as managing environments, experiment management across those environments, and the integration with training datasets as you go through those experiments. Not all those things sit in Amazon SageMaker; they sit outside, and that's a challenge that AWS has to think about more broadly. It's the actual data landscape, not the ML products, that's the problem.

I have been working with Amazon SageMaker for about six years.

We faced difficulties with implementation. When we first built our first lake, there weren't really any lake services. It was all microservices and containers. None of the lake services that AWS has today existed back then, so we've got a pretty complex and fragmented environment.

Parts of our team have been satisfied with Amazon SageMaker generally.

The product has been stable and scalable.

Their experience could still improve around scaling for ops and scaling operations. There aren't many people that know how to do that effectively. We're not really looking for support so much on how to solve problems anymore; it was more about how to scale and manage it across our different data science teams.

Technical support for Amazon has been pretty good; our account team depends on the challenges we have faced. The issue is not so much their support as the fragmented nature of their services.

I would rate their support for Amazon SageMaker a seven out of ten.

Positive

We've used a hodgepodge of solutions historically with our existing data infrastructure, basically lakes we've built in each cloud. We're actually moving all of that to Databricks at the moment.

We've had third parties help us manage these issues, but we've also had our own teams; it's a combination.

We have seen a return on investment. It depends on the different projects. Some do, some don't, depending on which ones we've been using.

If you manage it effectively, their pricing is reasonable. It's similar to anything in the cloud; if you don't manage it properly, it can be expensive, but if you do, it's fine.

We're okay with our current solutions.

Amazon SageMaker is one of the products we've been dealing with along with Azure ML.

I manage a data science team and our machine learning and AI infrastructure, so I have not been using it myself as a consultant or developer.

For machine learning itself, it's fine. The nuance is what the rest of your data infrastructure looks like and what other platforms you have. Particularly with AI ramping up, you have to go to another service again, so it depends on the types of solutions that you're actually trying to tackle.

For machine learning, I would give Amazon SageMaker a seven or eight out of ten.

I would recommend Amazon SageMaker to others.

Overall review rating: 7/10

Positive

My primary use case for Amazon SageMaker revolves around setting it up for our data scientist users. I am involved in the provisioning part of it, working on tasks like security setup and infrastructure, akin to an SRE role. It is mostly utilized for provisioning SageMaker for data scientists in my organization so they can utilize the tool.

Amazon SageMaker's sharing capability with the Feature Store is something I find valuable and is actively used by users in our environment. The evolution from SageMaker Classic to SageMaker Studio, particularly the UI part of Studio, is commendable. The infrastructure manages multiple notebook instances, providing flexibility. However, initial setup time might be slow due to spinning up of GPU or EC2 instances. 

Compared to classic notebooks that have one instance per user, Studio allows for multiple notebook instances, despite certain limitations like the lack of no-code and low-code capabilities.

Improvement is needed in the no-code and low-code capabilities of Amazon SageMaker. This would empower citizen data scientists to utilize the tool more effectively since many data scientists do not have a core development background. 

Additionally, dealing with frequent UI updates can be challenging, especially for infrastructure architects like myself. It involves effort to migrate to new UIs, making the updates not seamless. User auditing requires enhancements as tracking operations performed by users can be difficult due to dynamic IP validation and role propagation.

I have been working with SageMaker for more than four years.

I rate the stability of Amazon SageMaker between seven and eight. It is fairly stable despite occasional challenges with frequent UI changes.

Amazon SageMaker is scalable and works well from an infrastructure perspective. I would rate its scalability eight out of ten.

The technical support from AWS is excellent. I would rate it nine out of ten.

Positive

The initial setup for our type of infrastructure was complex as we had to restrict instance types and manage user domains. However, once the initial setup was completed, the system became seamless to use.

Pricing for Amazon SageMaker is high. That said, organizational discounts are available. For a single user, prices might be high yet could be cheaper for user-managed services compared to AWS-managed services.

Competing products to Amazon SageMaker include DataIKU and Databricks. These tools offer features that compete with SageMaker.

My overall rating for Amazon SageMaker would be seven out of ten. I recommend it to other users depending on the use case. 

For solution stability, I would rate it between seven to eight out of ten. For scalability, it would be an eight out of ten. Amazon SageMaker is suitable for companies with medium to large user bases, around 400 to 1,000 users.

The primary use cases of this solution include designing, deploying, testing, and evaluating machine learning models. It is also utilized to develop ML models quickly, operationalize them effectively, and deploy those models in production.

The most valuable features include the ML operations that allow for designing, deploying, testing, and evaluating models. 

Additionally, the AutoML capability is beneficial. These features facilitate rapid development and deployment of AI applications. The pricing and performance efficiency of the solution is better compared to some alternatives. SageMaker provides easy deployment, rating an eight on a scale of ease.

One area for improvement is the pricing, which can be quite high. Integration of the latest machine learning models like the new Amazon LLM models could enhance its capabilities.

I used SageMaker two years ago, having three years of experience with it.

Stability was rated as good, with a score of seven to eight out of ten.

The scalability is considered good. However, the availability of GPU instances can be a challenge, requiring proper planning.

The support is very good with well-trained engineers whose training curriculum is rigorous.

Neutral

While previously using SageMaker, I have moved to Google's AI ecosystem for most of my tasks.

The initial setup is easy, scoring an eight, making it straightforward for users.

The implementation involves ML engineers, data scientists, and ML operations engineers who understand the underlying algorithms.

The return on investment varies by use case and offers significant value in revenue increases and cost saving capabilities, especially in real time fraud detection and targeted advertisements.

The pricing can be up to eight or nine out of ten, making it more expensive than some cloud alternatives yet more economical than on-premises setups.

I have used both Amazon SageMaker and Vertex AI in my work.

On a scale of one to ten, I would rate SageMaker at eight out of ten. Depending on the existing ecosystem, it can be recommended, and monitoring supply constraints is advised for GPU instance usage. 

I primarily use Amazon SageMaker for analytics on cold calling and analyzing financial numbers. Additionally, it helps me connect with customers in real-time using chatbots.

I appreciate the telemarketing analytics and integrations with Chime SDK. They offer insights into everyone making calls in my organization. This allows monitoring and performance grading, as I instantly know when someone has a bad call. It also allows me to gain insights from data.

The main challenge with Amazon SageMaker is the integrations.

I have been using Amazon SageMaker for about a year.

I have never had to contact their support team.

Neutral

The initial setup requires some experience and knowledge of AI.

I have seen a return on investment, probably a factor of four or five.

Amazon SageMaker is expensive, costing close to $1,000 a month.

I recommend learning about AI and studying the AWS machine learning specialization. 

I would rate Amazon SageMaker an eight on a scale from one to ten.

In terms of the tool's use case, if it is serverless, and if the compute involved is not too high, or if it is a PoC kind of a thing, and you want the microservices kind of architecture to be going and go for a pay as you go model, you can use the tool. With the tool, you know what is happening, so maybe you can cut costs by going with an on-premises model and having a stable system for computing.

The models that the tool has, the libraries, and the ML libraries are rich. That is good, and that is one of the features of the tool. The other feature is how the tool interacts with other components of AWS, like Lambda and S3. It's seamless if you are using AWS architecture.

I was working with Amazon SageMaker Ground Truth. The serverless feature is important and worth it because you don't have to spin up an EC2 instance every time or a server. For us, the main attraction is that it's serverless. You pay only for the compute that you use, and then there is its ecosystem. Whenever you use something in AWS, the ecosystem is very rich.

For any cloud provider, the cost has to be substantially reduced, especially in the case of Amazon SageMaker, which is extremely expensive for huge workloads. In EC2, you have spot instances that cut costs tremendously, but you don't have that in Amazon SageMaker. You pay for the local usage. I would like to see better integration with GPUs. GPUs are very expensive for AWS or any cloud provider. NVIDIA has introduced options with Databricks for GPUs, so it would be interesting to see how Amazon SageMaker can parallelize GPU usage. I haven't used it to scale multiple GPUs automatically for model training. The key points are the cost and how effectively they integrate GPUs into the workload for training machine learning models. We want to see how seamless it is and how it can work. I haven't used multiple GPUs scaled automatically. For model training, the first concern is cost, and the second is how effectively they want to integrate GPUs into the workload for training machine learning models.

I have six to seven years of experience with Amazon SageMaker.

After production, a different team handles the tool. I would be there till the PoC phase and then after production we move on to other projects. From my standpoint, I rate the solution's stability a nine out of ten. I don't continuously maintain production workloads, as it is managed by a different team.

The tool is extremely scalable. If you go on and use it for the entire life cycle of ML, it's very expensive. I say it's scalable, but it's expensive. For a production project, I would think that deployment and inferencing can be in SageMaker. We will have to, at some point, move other stuff to a less costly thing. It is very scalable.

My clients are mostly mid-level to enterprise businesses. I have worked with huge clients, but our clients may not be that huge. When we worked as a product engineering partner to AWS for two years, I dealt with huge clients, but that was very specific. I was working with Amazon SageMaker Ground Truth. I have worked with Nissan, Sony, and Samsung, but even though it was ML, it was a very niche kind of thing. We were doing labeling support for the ML models and training data. We were doing training data and labeling by partly using ML. Sometimes, we used to use Amazon SageMaker as well, but that was very niche. If we were to embark on a complete ML journey with some clients, then I would say that I have dealt with small to mid-scale customers.

The technical support of the tool was good. I rate the technical support a ten out of ten.

Positive

My company takes a tool-agnostic approach and utilizes a variety of AI and ML tools across the ecosystem, focusing on Amazon, Microsoft, Databricks, and Snowflake.

Speaking about the product's initial setup phase, I am in an Amazon ecosystem. I find the tool good because it depends on whether you are very command-line or user-interface-driven. If you are user interface-driven, I think Azure is good. I am okay with the command lines and shell kind of an interface. I am more used to AWS than other cloud providers. I think it's good for me. For a programmer, it's easy. As a technical person, it's easy. As a business person, I think Azure is a little easier than AWS.

The tool is deployed on a private cloud. Sometimes, we don't use it ourselves and then just give access to the apps to the client, meaning we host it. We host the model, and then the clients use the model but don't have access to infrastructure. We mostly used AWS along with some other cloud providers too.

The tool's deployment was fast, and it took a week.

For the deployment, whatever we do in general for on-premise, we do almost the same thing in SageMaker. Just that it offers parallelism. We do EDA and all that. We first have the data in the S3 bucket, and then we do the EDAs. Then we do the training. Then, we do the grid optimization. First, we mostly start with the pilot. When I say that deployment takes one week, it's just for having it up and running and showing some results with the model. It's not for the complete production model, as it is a different cycle. We start the PoC with the pilot model first, and then we kind of do it. Whatever steps are involved for on-premises, we do the same steps for SageMaker. Just that the computing cost is less, and we don't have to spin up a server. For the rest of the steps, like getting and cleaning the data, we don't do a lot in SageMaker. We kind of do it outside, or we do it with EC2 instances or use PySpark, maybe, and Databricks because that is a little easier compared to SageMaker libraries. You can do it, but it's kind of expensive. We do all that data transformation, and then we do the modeling. Most of the time, 100 percent deployment is done in SageMaker, but the rest of the parts are a mix of technologies. There is an influenced pipeline.

The cost offers a pay-as-you-go pricing model. It depends on the instance that you do.

I used Amazon SageMaker as a customer from the client side, though I have worked as a contractor with Amazon for two years. I work for Persistent, but the engagement with our client was over last year. When I was working at Amazon for two years, I was working at Amazon. Now, we are partners as well. We are strategic partners for Amazon, Microsoft, Google, Databricks, Snowflake, and most of the ecosystems.

If you want to try out something with, say, for instance, you want to do a PoC for testing more, I mean, say, for instance, that you're doing a data annotation project. You want to see how it goes. You don't want to invest a lot, and you want to try it out and see whether it works or not. For those kinds of typical PoC situations, I would say Amazon SageMaker is good. You are using a microservices architecture, and you want to go serverless. That is the first use case. The second use case is that you want to go serverless and plug and play a lot of components rather than having a bulk of computing like EC2 and all that. You would rather have an Amazon setup that is serverless.

We use it for tuning, but it's just like any other tool, except for the fact that it's serverless. It's not that it significantly boosts anything; it's just a choice. Either we tune it on-premise or we tune it on the cloud. We use Azure, AWS, and all that. So, in terms of tuning, it's not special. It's just the way you tune any model in any environment, and that is not a huge thing. It is a good tool that works well with its components and other components. There's nothing special about the tuning itself. You can either use PySpark or other cloud technologies. It's not that we get a huge boost just because it's AWS.

The serverless feature and the complete lifecycle that can be handled inside SageMaker are important. It covers everything from training the model to deploying it and sometimes using it for data pipelines. However, we generally don't use it for pipelining and data transformations because it's expensive inside SageMaker. We do use it for model training, although sometimes we train outside. We also utilize model training and Amazon SageMaker JumpStart, which is pretty handy because you don't have to train the model from scratch. You can use it, especially for LLM settings, right out of the box. There are models inside SageMaker that make it a little faster, both from a computing perspective and from a bandwidth deployment perspective, so you don't have to spend a lot of time training before deployment. Amazon SageMaker JumpStart is definitely valuable, along with the whole lifecycle for ML as well.

I would recommend others if they want to do a quick PoC workload, or proof of concept, and if they want to do something very quick, then I would definitely recommend it. If it's a very huge production workload, then I might want to consider other options. But for anything where there is a PoC kind of thing, I would recommend products in such areas.

Speaking about AI, I can say that it's kind of quick to set up and get it running. I can't say specifically. We have worked on a lot of projects. We have worked with document processing projects a lot. In those cases, if you were asking about specific projects, I can remember a recent project where we were trying to digitize documents using manual annotation and automated ML models, but there, we didn't use SageMaker. We used Amazon SageMaker Ground Truth, which is under the umbrella of SageMaker. If you use Ground Truth, it's a SageMaker product. We were using SageMaker Ground Truth, which is pretty handy because it sits well in the environment. If you are specifically asking how it accelerated the process, it was easy to set up, and we just got going in less than a week. So, yeah, I can think of that example.

I rate the tool an eight out of ten.

The primary use case for Amazon SageMaker involves developing on the cloud to utilize flexible compute resources. I use it when dealing with really large datasets that do not fit on my local laptop. I leverage SageMaker's more powerful GPU, CPU, and memory for these tasks.

Amazon SageMaker, in combination with Posit, helped develop an internal application that aids our business partners in prioritizing workflow, saving us approximately three million dollars.

The most valuable feature of Amazon SageMaker is its flexibility in selecting resources according to my needs. It is not a one-size-fits-all solution and is highly adaptable to my requirements. 

Additionally, the intuitive interface and streamlined user experience make it easy to navigate and set up various tools like Visual Studio Code or Jupyter Notebook. Its integration with our Snowflake environment also contributes to its overall ease of use.

There are two areas in SageMaker that need improvement. 

Firstly, when starting a new session, the waiting time can be quite long, ranging from two to five minutes. 

Secondly, the integration with Snowflake is not optimal. I face challenges in authentication, requiring multiple steps to achieve single sign-on. This process feels somewhat cumbersome.

I have used Amazon SageMaker for about one and a half years.

Amazon SageMaker has been quite stable overall for the past one and a half years. I cannot recall an instance where it broke down.

The scalability of SageMaker is highly impressive. Its compute resources are incredibly flexible, allowing us to scale from a basic setup to one with multiple GPUs and extensive RAM, depending on our requirements.

The technical support for Amazon SageMaker is not very responsive. It takes time for them to respond, and setting up a meeting with a support person can be challenging. Moreover, they often request additional fees for more attentive support.

Neutral

Previously, we used an Azure Kubernetes cluster to host Posit Workbench. We switched due to adding new team members skilled in AWS and to reduce infrastructure costs, which were over one hundred thousand dollars annually with Azure Kubernetes.

The initial setup was rated eight out of ten. It was straightforward for the tech team to install Workbench and SageMaker, following AWS documentation. However, configuring security roles and access for different users was more complex.

I worked with the tech team for the initial setup, and they found the process very straightforward using AWS's guidelines.

The use of Amazon SageMaker, along with Pawsit, led to developing an application that saved the organization about three million dollars.

Amazon SageMaker itself is relatively inexpensive, primarily charging for compute resources. However, there are additional costs for enhanced customer support services.

Amazon SageMaker is an excellent product for users already within the AWS ecosystem, but its integration capabilities outside AWS could be improved.

I'd rate the solution eight out of ten.

I am currently working with AWS services like SageMaker, S3, EC2, VPC, load balancing, auto scaling, and RDS. SageMaker is used primarily for AI projects to build, train, and deploy AI models from scratch.

SageMaker supports building, training, and deploying AI models from scratch, which is crucial for my ML project. It provides lifecycle configurations, similar to EC2's user data, for running scripts on instances. It also offers VPC features to isolate SageMaker instances when needed, which is a valuable use case.

The entry point can be a bit difficult. Having all documentation easily accessible on the front page of SageMaker would be a great improvement.

I have been working with SageMaker for about one week due to a project related to AI.

I have not encountered any stability issues with SageMaker so far.

Scaling SageMaker has not been an issue. If required, I can increase the instance size to a higher tier as needed.

I haven't needed to contact AWS technical support for this project.

Positive

When I first started working with SageMaker, I was unfamiliar with it. I consulted AWS articles and searched online to understand how SageMaker works.

I was responsible for deploying SageMaker for my client and configuring notebook instances with VPC and subnets.

Before deploying SageMaker, I reviewed the pricing, especially for notebook instances. The cost for small to medium instances is not very high.

I recommend SageMaker for ML projects if you need to build models from scratch. If you do not want to maintain the instances, consider using Datablock instances. On a scale of one to ten, I would rate SageMaker a nine.

The primary use cases for Amazon SageMaker are for EDA processing, ML model building, setting up MLOps, predictive analysis, customer churn models, fraud detection, image and video analysis, as well as NLP projects. It is a versatile tool in the machine-learning landscape.

The most valuable features in Amazon SageMaker are its AutoML, feature store, and automated hyperparameter tuning capabilities. These features allow for generating high-quality models without needing extensive coding knowledge, making it accessible for non-experts. SageMaker helps in end-to-end machine learning, incorporating data preparation, model deployment, and continuous monitoring.

Improvements are needed in terms of complexity, data security, and access policy integration in Amazon SageMaker. It is considered complex to integrate these aspects, and adjustments need to be made in multiple places, which should be more user-friendly. A centralized interface for managing these configurations is desired.

I have been working with Amazon SageMaker for nearly three years.

Amazon SageMaker's stability depends on how well-configured the entire setup is. Due to the interconnected dependencies within the system, the learning curve may be steep for new users. However, with proper configuration, the overall stability is adequate.

Amazon SageMaker offers a high level of scalability. It allows dynamic resource allocation and supports large datasets through various features like multi-model endpoints and flexible instance configuration, scaling up or down according to requirements.

Technical support for Amazon SageMaker involves communication through web chats or telephone based on the support agreement.

Positive

In AWS, we used to build the whole pipeline model by ourselves, component by component. With Amazon SageMaker, costs have been optimized as it includes pre-configured components that reduce overall expenses.

The initial setup of Amazon SageMaker can be achieved quickly if the default configuration is used. However, setting it up more customizable, such as for specific requirements, can make the process time-consuming, earning an eight out of ten in terms of ease.

A single knowledgeable person with expertise in ML and cloud can handle the deployment and maintenance of Amazon SageMaker.

We have seen a significant reduction in costs using Amazon SageMaker. Building any ML lifecycle benefits from SageMaker's pre-configured components, which bring down the overall cost compared to setting up all components separately.

While Amazon SageMaker is expensive compared to other cloud vendors, certain cost optimizations can be made with proper setup and configuration knowledge. Greater visibility from AWS regarding cost-impacting configurations would be beneficial.

No other solutions were evaluated outside of AWS, as we were setting everything up within AWS before opting to use Amazon SageMaker.

I rate SageMaker eight out of ten. 

New users should conduct a pilot or proof of concept with Amazon SageMaker to see if it aligns with their business use cases. Evaluate and understand the integration with other AWS services and ensure the team has adequate knowledge to handle monitoring, model performance, and managing costs efficiently. Engaging with the community to remain updated on any misconfigurations is also advisable.