For people who grew up learning the math of backprop, who enjoy thinking about syntax trees and computation graphs, Tensorflow will allow you to make full use of you that insight. Interesting loss functions like Wasserstein loss (where the gradient itself enters as part of the loss function) enter naturally. Review collected by and hosted on G2.com.

The mix between Tensorflow v1 and v2 code is somewhat difficult to learn, if you only get into it now. Tensorflow v2 is modeled much more on Keras, and is designed for you to particular architectures and pipelines. That's great, but if you then want to mix that with the flexibility of v1, you run into a lot of pain. Review collected by and hosted on G2.com.

There are so many points which I liked about TensorFlow i.e. It is fast and its scalability on the larger dataset. With the help of TensorFlow, I am able to write customizable evaluation functions. Review collected by and hosted on G2.com.

Overall performance of TensorFlow is good but the documentation of TensorFlow can be improved. Sometimes I felt inconsistency in the algorithm which can be further optimized. Review collected by and hosted on G2.com.

Python based and API is intuitive. Keras is great and uses the Tensorflow library. I used scikit-learn prior and it was so much harder to understand and require way more code to get the same things done. The user-friendly interface is honestly the best part of Tensorflow/Keras. Review collected by and hosted on G2.com.

Not a lot, but for Tensorflow Lite, a user manual to port to other boards would be great. I wanted to use Tensorflow Lite on my TM4C123GXL board, but it's not a supported platform. I am sure there is a way to get it running on any board, I just do not know how. Review collected by and hosted on G2.com.

It was easy to get on with Tensorflow compared to other machine learning libraries. There are tons of community support, tutorials, videos, and even pre-build models to learn and get maximum out of it in a short time. And what's more, it's completely free and open-source. Review collected by and hosted on G2.com.

There is nothing to say in this section. Tensorflow has it all, and I love it. I haven't faced a serious issue yet, and even If I did, the community is there to solve them happily. Review collected by and hosted on G2.com.

The main important thing I like about Tensorflow is, it is Open Source. Anyone can use it and can create multiple Machine Learning applications. I can also visualize my machine learning model in TensorFlow by using Tensorboard. Tensorflow also supports Keras, so we can easily create ML and CNN models using it. Tensorflow is compatible with many programming languages like Java, Python, C++, Ruby etc. Review collected by and hosted on G2.com.

One thing I don't like about Tensorflow is, it gives updates regularly. So it becomes a little bit difficult to install new versions. Because sometimes, whatever application I developed may not be supported on a more recent version of Tensorflow. Review collected by and hosted on G2.com.

Tensorflow is a mature framework that offers many valuable features such as Keras, Tensorboard, data processing modules, easy-to-implement multiprocessing, integration with HDFS, and more. Tensorflow has a strong community and very robust documentation. Tensorflow has many time-saving features, such as easily integrated pre-trained model layers. The TensorFlow model hub is one of the best I have seen in terms of ease of finding and using pre-trained models. There are many demos and example notebooks that demonstrate how to use complex and straightforward concepts. Review collected by and hosted on G2.com.

Tensorflow has gone through many iterations over the past years, so code maintenance has been an issue. In my opinion, eager execution is a preferred method of developing and debugging networks; however, experience with legacy TensorFlow makes the switch more challenging. Since the deep learning research community favors PyTorch over TensorFlow, researchers generally find state-of-art models and new methodologies implemented in PyTorch. Review collected by and hosted on G2.com.

The most valuable part of TensorFlow is the Tensorboard. While training the AI model development, it provides better visualization for debugging and error handling. Review collected by and hosted on G2.com.

The least liked part of TensorFlow is its implementation speed. In comparison to another deep learning framework, development time is higher in TensorFlow. Review collected by and hosted on G2.com.

Tensorflow is very mature deep learning library which is heavily used in production scenarios. I particularly like the tensorflow-lite version which comes along which reduces the size of the model and is good to deploy in edge devices. Review collected by and hosted on G2.com.

It requires a little more coding as compared to pytorch. Pytorch is more pythonic and hence is easier to learn and implement Review collected by and hosted on G2.com.

The fact is that you can create the network and then do computation all at once. The computation is well optimized to run on GPU. The tensorboard support enables us to view the metric like accuracy and weights during the training which is absent in other deep learning packages Review collected by and hosted on G2.com.

The high level api is not present in the package itself. For that we need to use keras or other packages which is build on top of this but these high level API is not native to tensorflow Review collected by and hosted on G2.com.

It works wonders when processing image, text and audio data. The documentation is very good and easy to use. With Keras, you can do your deep learning work simply and quickly. Open source. The best software library on the market for deep learning. It's reassuring to have Google behind it. Documentation is being updated. Functional. Review collected by and hosted on G2.com.

It's forcing the video card. Detecting and resolving errors is sometimes difficult. Review collected by and hosted on G2.com.