Keras

What do you like best?

Keras is very easy to use, even for beginners that have basic Python programming skills.

Even complex deep learning models can be built just with a few lines of codes. The biggest advantage is running time: the codes execute pretty fast.

Besides, code examples are intuitives and readily availables. The documentation is built with care and attention and there are answers for almost every issues. I always find what I need to solve my problem.

What do you dislike?

Sometimes it is not easy to find code examples for some advanced features. Also there were some errors in the code execution and I had difficulty to understand where they came from. Since Keras is pretty simple code, sometimes it is hard to customize models that have been built by someone else. In such case, I would rather use other packages that might be more complex but do the job.

Recommendations to others considering the product:

I would say: let's give it a try. I first knew about Keras when I was a Python beginner and was immediately impressed by its user-friendly aspect. Besides that, the documentation is complete and the community is there to help.

What problems are you solving with the product? What benefits have you realized?

I used Keras to build several deep learning models in different topics: image classification, time series prediction, categorical variable prediction, object detection in images. I used it in both my academic research as well as in professional projects.

I successfully built a CNN combined with a RNN model that was later deployed in a mobile application. Even though the original dataset was big, Keras has an impressive running time. That helped me a lot to speed up the execution time of my codes, which is crucial for the project's success.