TensorFlow Reviews

In one of my latest projects, I used convolutional neural networks along with several other models for Finance; The objective was to predict future Close Price of S&P500 index. And in the end, I discovered that an ensemble model of convolutional neural networks works the best; I got a very low error and pretty good accuracy. That was my most recent project.

Another project that I used the solution for was using convolutional neural networks was in visual recognition in which the goal was to take a picture of somethin. Then the model would recognize what the images are. That's the pretty standard use case of convolutional neural networks. Along with that, I also used generative adversarial networks and style transfer in TensorFlow. 

The primary feature that I personally like is the fact that TensorFlow allows us to utilize GPUs. At present, in data-driven deep learning, the most important thing is the usage of GPUs which accelerate the training of the model by many folds. What I love about TensorFlow is that it allows me to use GPUs.

TensorFlow is a framework that makes it really easy (and quick) to use deep learning. For example, it has an API, which is called 'Sequential API' , and using that, you can create a whole Deep learning model in about five lines of code. That's another core benefit from my perspective.

TensorFlow is primarily geared towards Python community at present. JavaScript is a different thing and all the websites, web apps and all the mobile apps are built-in JavaScript. JavaScript is the core of that. What can be improved with TensorFlow is how it can mix in. How the JavaScript developers can use TensorFlow. 

There's a huge gap currently. If you are a web developer, then using Machine Learning with TF is not as straightforward as using a regular Javascript library by reading its documentation. TensorFlow should provide a way to do that easily. 

TensorFlow is very stable. It's very reliable. Training a model won't halt in the middle for unknown backend issues. Behavior of each TF layer is predicable based on its documentation. It's one of the top frameworks. If you want to do deep learning, then TensorFlow is the way to go. 

On the other hand, the environment on cloud you use for TensorFlow is a separate story: there are both paid and free versions. For example, Google Colab provodes free aceess to unlimited GPUs yet it's not stable. You might see frequent disconnections or reset of runtimes. Luckily, the paid versions resolve all such issues for a small price.

If you're building large models, then TensorFlow is going to be scalable partially because it allows you to use GPUs. If you have a huge data set, and if you want to train it on your local computer, then it is going to take a whole lot of time if you are not using GPUs with TensorFlow.

I've never used their technical support services. I can't speak to how knowledgable or responsive they are.

TenserFlow and PyTorch are the two big giants of deep learning and they are widely used. If you want to use deep learning, your first choice is either TensorFlow or PyTorch.

I actually recently developed a whole video tutorial on NLP and all the code was written using PyTorch. I also work in PyTorch.

TensorFlow was recently updated to Version Two. Until a year ago, it was TensorFlow Version One and TensorFlow One and PyTorch were hugely different. PyTorch is more Python-like. If you're writing code in PyTorch, you'll feel like you're simply writing your regular Python code as a developer. TensorFlow-1 was a different design in which the developer will first build the computational graph and later creating a session to execute.

Recently, TensorFlow has been updated to TensorFlow-2. It has been made more like PyTorch due to its popularity. At present, when you are writing code at a high level, for example, in a sequential API, you won't even notice, from a programmer's perspective, that there are many differences in TensorFlow and PyTorch. At the same time, both of them allow us to use GPUs, which is the primary use case.

At present, if you're working on Colab, you do not need to set up or anything or install TensorFlow as Colab is specifically for TensorFlow and PyTorch and they are pretty much built-in and everything is there already. If you're working on a cloud too, you can just write TensorFlow.

In terms of maintenance, from the developer's perspective, it doesn't require any maintenance, however, from the creators of TensorFlow's perspective, they are obviously building TensorFlow and maintaining and optimizing it all the time.

This is an open-source solution.

I primarily work on Google Colab in which everything is installed. The most recent versions of TensorFlow are already installed on the Colab.

I have written a deep learning library, which is like very much TensorFlow and PyTorch. It's like my own miniature version of TensorFlow, which I have written as it was an academic project. TensorFlow hides all the details of like nitty-gritty details like how is it working, how the matrices are being multiplied, how is it being handled on GPUs? All these details have been abstracted. If you're writing a model in TensorFlow, you will write just five lines of code (in Sqequential API), in these five lines everything is happening. When I was developing that library, I learned that those five lines of code acutally map to, say, a 1000 lines of code underneaththe hood. I actually wrote all that in order to learn how exactly it works. I just learned what those a thousand lines are.

TensorFlow is just a tool for deep learning. You can use a complete model, which will recognize images in five lines of code. However, to really do deep learning you have to go underneath the hood and understand how exactly things are working. If you are coming from a different background and you write five lines and you can do a model, that's great, however, for example, if a debugging problem comes, you will be in much better shape if you have learned what's underneath. You will be much better shaped to debug your code. If you better understand your code you can better optimize your code. TensorFlow provides you a layer of abstraction, however, that layer of abstraction is bad in some ways.

Primarily, I'm a machine learning engineer. Most of my projects are on using TensorFlow, from my perspective I use TensorFlow a lot and PyTorch occasionally. I also am a full stack developer, I develop apps using React, Django, and D3 yet I don't work a lot in that area. Primarily I work from TensorFlow. From my perspective, they are widely used.

Overall, I would rate the solution ten out of ten. It's very good.

I worked for a French company. They used TensorFlow for image classification. after that, I started working with a Russian-American Company who used TensorFlow mainly for object detection. TensorFlow is very good at object detection. We also used it once for natural language processing and audio processing, but I was not directly involved in that project. I was just assisting with deployment issues. We have some clients which wanted us to deploy on the cloud. Alternatively, some clients are releasing Tenserflow on some new edge devices, as an alternative to deploying on the cloud. It is going to be called NextGen AI or something like that from AWS. We use it for all aspects. including data processing, training, and sometimes deployment, but sometimes the use cases differ in practice for ML. As a result, we sometimes stay with TensorFlow or move into AWS specific architectures.

TensorFlow has benefitted us by enabling faster training and deployment. With TensorFlow, we don't really need any more DevOps to do the deployments. Even data scientists can do the deployment part. This has saved about 30% of the time we used to take for deployments.

Our clients were not aware they were using TensorFlow, so that aspect was transparent. I think we personally chose TensorFlow because it provided us with more of the end-to-end package that you can use for all the steps regarding billing and our models. So basically data processing, training the model, evaluating the model, updating the model, deploying the model and all of these steps without having to change to a new environment. Especially the part where you could train the model again, then evaluate it if it's better than the previous versions. It will do the deployment on its own. The end-users will not really see the change, as the update takes place without any downtime.

It doesn't allow for fast the proto-typing. So usually when we do proto-typing we will start with PyTorch and then once we have a good model that we trust, we convert it into TensorFlow. So definitely, TensorFlow is not very flexible.

We have been using Tensorflow since 2017, so three years.

It's very stable. So we usually don't get any problems. Once any bugs are fixed, you shouldn't have any problems with TensorFlow. Once the deployment process is completed, you can monitor your model and datasets. You can monitor your model to ensure it is correctly deployed and it's working as it's supposed to do, including services.

TensorFlow is very scalable.

I've never really contacted TensorFlow support, but definitely, I can say you don't really need to do that because the support, like the community is pretty strong. Whatever problem you face, there's always going to be some Stack Overflow answer for it or at least some GitHub issue where you can find your solution.

We used PyTorch and MXNet. A couple of my friends actually used MXNet, but I did not use it personally. Now I work mainly between PyTorch and TenserFlow.

Setup got easier with TensorFlow 2. With TensorFlow 1 it was a bit more complicated. There are fewer compatibility issues with the newer version. Training takes at least a week. Deployment usually takes one or two days. With regards to deployment, this changes depending on the client. The usual method is to get our TensorFlow models up and running and then we have to convert them into specific formats depending on the client's requirements. Some clients actually require AWS specific formats. To incorporate that we usually just convert our TensorFlow models to AWS compatible models.

We use in-house teams. I think the ML team has around 20 people. There is a team in Russia, Ukraine and the U.S. I am part of the team in Russia. In Russia, we have around 30 people who have used TensorFlow, including data analysts. They basically handle data pre-processing. We also have ML engineers and ML Ops.

TensorFlow is free, so cost is not an issue.

PyTorch is very flexible, and definitely more flexible than TensorFlow. However, TensorFlow allowed us to deploy our models faster and more robustly. With TensorFlow, you don't encounter a lot of errors. PyTorch is faster to prototype with. So it's very flexible. You can do all the changes that you want, but it's not very stable, whereas TensorFlow is more of a stable solution but is not very flexible. We looked at other alternatives, but they don't scale to large problems.

Have a look at TensorFlow extended. It's very useful. Especially if you know how to use the old system. It will speed up the process of deploying your model. Don't reinvent the wheel. There's always going to be a good GitHub repo out there which kind of answers your solution. You shouldn't really spend a lot of time trying to build the new models where there is some other open source project that actually did a good job of the modelling part. You definitely need to have your own pipelines for this process. Try to build the pipelines that automate most of the tasks for you. Then all you need to concern yourself with is just the architecture. Obtain a pipeline template from GitHub of what you are trying to achieve, amend it for your needs and then you are ready to go. Your model is training already. I would rate TensorFlow 8 out of 10.

I use the product mostly for machine learning and creating AI models.

We encountered version mismatch errors while using the product. It sometimes does not integrate well with other libraries in Python, which can be problematic. Additionally, it can be less intuitive when creating neural networks than PyTorch.

I have been using TensorFlow for around two years.

I haven't experienced any significant stability issues or bugs.

Around three to four executives use the product in my current organization.

We chose TensorFlow because we have developed our skills around it. Based on our expertise, it was the best option for us.

Sometimes, working on the setup process can be difficult because loading models takes a significant amount of time compared to PyTorch.

I recommend using TensorFlow. Its libraries make Python programming smoother and reduce the workload. It is easy to use and learn. 

I rate it an eight. 

I use this solution out of personal interest — for AI.

Our company doesn't use it too much. We tend to use OpenML, which is included in the OpenCV package, OpenML. 

It's got quite a big community, which is useful.

I tend to find it to be a bit too much orientated to AI itself for other use cases, which is fine — that's what it's designed for. Personally, I find it to be a bit too much AI-oriented.

We have been using TensorFlow for roughly five years.

We haven't had any issues stability-wise or scalability-wise. I tend to use it with Python. It seems okay. It works fine. 

The initial setup was straightforward.

I would happily recommend TensorFlow to people who are looking to use it for AI. Overall, on a scale from one to ten, I would give this solution a rating of eight.