GC

Verified User in Computer Software

Mid-Market (51-1000 emp.)

1/17/2018

"Keras: Tensorflow made simpler"

5/5

What do you like best about Keras?

Keras makes working with Tensorflow and doing deep-learning very easy: it provides an user friendly API that hides most of the complexity of using TF. Models can be quickly defined by piping Keras layers together and tweaking the parameters through the layer constructor. The learning process can be easily customized by tweaking the various rates. Since training a deep-learning model takes a lot of time/data and things can easily go wrong, you will appreciate the availability of the training callbacks: the ReduceLROnPlateau, ModelCheckpoint or TensorBoard, to name a few. Definitely a library the must be used for quickly putting together prototypes before diving too much into the details. Review collected by and hosted on G2.com.

What do you dislike about Keras?

Keras doesn't make it easier to debug your models: but this is not a problem strictly tied to the library and it's common in the deep learning world. Review collected by and hosted on G2.com.

Recommendations to others considering Keras:

If you need to quickly put together a prototype for a new idea, Keras is the go-to library to get started. You might need to do without it to have a greater control over your model in the next phases, but this is definitely the way to start.

Documentation is widely available and there is plenty of tutorials to get yourself up to speed with the API. Review collected by and hosted on G2.com.

What problems is Keras solving and how is that benefiting you?

Training a deep-learning model for performing anomaly detection in time series and recommendation systems. Review collected by and hosted on G2.com.

GC

Verified User in Consumer Services

Small-Business (50 or fewer emp.)

12/27/2017

"Excellent Platform for Deep Learning"

5/5

What do you like best about Keras?

The Keras API makes experimenting with deep learning models a breeze. It's easy to stack layers and to experiment with different deep learning architectures for a wide variety of problems including natural language processing and activity recognition. Review collected by and hosted on G2.com.

What do you dislike about Keras?

Keras is still in development. Although it works great the majority of the time, don't be surprised if you find a bug or that a specific architecture/deep learning layer you want to use is not implemented yet. Review collected by and hosted on G2.com.

Recommendations to others considering Keras:

There are plenty of tutorials online on how to get started with Keras. However, it's always good to understand the math behind any neural network architecture before implementing it to make it easier to debug and understand why it does what it does. Review collected by and hosted on G2.com.

What problems is Keras solving and how is that benefiting you?

I have used Keras to build natural language processing models. The application of these models has helped us to realize better engagement with our users and better performance from our employees. Review collected by and hosted on G2.com.

SC

Subit C.

Graduate Research Assistant

Defense & Space

Small-Business (50 or fewer emp.)

10/30/2017

"Usability: Maximum, Configurability: Minimum"

3/5

What do you like best about Keras?

Keras is all about ease of use. It should take a true beginner (someone who theoretically understands neural network but has never coded one) less than an hour to code their first network. And it works ! I also like that it works with both Theano and Tensorflow (although that might change soon). Review collected by and hosted on G2.com.

What do you dislike about Keras?

There is minimum configurability and things that are essential and easily accomplished in other frameworks takes a while. For example, building a layer that is not in one of the predefined templates or checking the weights on each layers. Review collected by and hosted on G2.com.

Recommendations to others considering Keras:

Keras is super easy to use and a good way to learn neural networks. However, once you start building complicated models/networks Keras is rarely enough Review collected by and hosted on G2.com.

What problems is Keras solving and how is that benefiting you?

Keras tries to simplify the building of deep neural networks. Its audience is people who are not hardcore software engineers/data scientists who breathe Tensorflow/Theano every day. This it does perfectly. Review collected by and hosted on G2.com.

GC

Verified User in Computer Software

Mid-Market (51-1000 emp.)

5/29/2018

"Very recomendable for deep learning in medical imaging"

4.5/5

What do you like best about Keras?

Different backends, relies on python. Works in 3D images. Is easy to install. Review collected by and hosted on G2.com.

What do you dislike about Keras?

Since it is python based, it's not compatible with 3d image visualisation. When using tensor flow, the only way to follow the training is using tensor board, which is slow. Review collected by and hosted on G2.com.

GT

Verified User in Telecommunications

Small-Business (50 or fewer emp.)

5/21/2018

"Easy and flexible"

4.5/5

What do you like best about Keras?

Easy setup, flexibility, different backends, python-based Review collected by and hosted on G2.com.

What do you dislike about Keras?

There's no interface such as nvidia digits. There is tensorboard but it's slow and not as intuitive as digits Review collected by and hosted on G2.com.