It's written in Python. Python is my number one choice. Review collected by and hosted on G2.com.
Syntax is somehow complex. It could be much simpler like others. Review collected by and hosted on G2.com.
Keras is very easy to use, easy to learn with a great community backing it and the code is readable Review collected by and hosted on G2.com.
It's a high level API which makes it a little slow. It only runs on python programming language unlike other deep learning frameworks that run on different languages Review collected by and hosted on G2.com.
Keras is the best deep-learning framework for beginner level data scientists and machine-learning/artificial intelligence researchers. I like its high-level features that one can use to program a neural network, for instance, write an image classification task in a few lines. Additionally, the availability of pre-trained networks is a plus which allows practitioners to use DL models even without the need for training or an intelligent initialization for researchers. The feature I like the most is high-level function for training, which automatically keeps track of training and validation loss and also keeps an estimate of training progress. Loss functions are already implemented, so in most cases one do not have to implement these functionalities by themselves. Training and validation set queues are an efficient way of data handling and that feature is already implemented in Keras. Review collected by and hosted on G2.com.
I only have one complaint that once you start working with Keras it is difficult for researchers to move to Tensorflow type libraries where one can implement flexible functionalities specialized to one's research. Review collected by and hosted on G2.com.
Every ML starter should use keras as it directly binds with theano and tensorflow Review collected by and hosted on G2.com.
New version of keras has some incompatability with older version of Keras Review collected by and hosted on G2.com.
Very simple, clean and easy framework to implement deep learning models compared to tensorflow, caffe and PyTorch Review collected by and hosted on G2.com.
Even though it provides a simple interface and high-level implementation of basic deep learning blocks it becomes difficult for custom functions Review collected by and hosted on G2.com.
The API they provide are simple to use and are easy to understand with the detailed documentation available on the official Keras website. Generally Deep Learning Models are complex and hard to implement but with Keras API a newbie can also build his/her own model with great ease. Review collected by and hosted on G2.com.
Keras is a amazing Deep Learning library that fits with TensorFlow and many more Deep Learning libraries but the disadvantage of Keras is that it is only available for Python whereas TensorFlow and the other libraries have multi-language support. Review collected by and hosted on G2.com.
The API they provide is efficient where one can make ones own deep learning models with only some basic knowledge. Also they provide support for other machine learning/deep learning libraries like TensorFlow which really makes work easy. One more thing I would like to mention is that they provide module wise API, which means one can play around with each module while building a deep net like the layers, activation functions, cost functions, e.t.c. Review collected by and hosted on G2.com.
This library is pretty good and only option for me to carry out my deep learning tasks. So I personally have no dislikes for it. Review collected by and hosted on G2.com.
Keras makes making machine learning models super easy. It has an easy to learn syntax that runs on top of Tensorflow which is the backend I use, which compiles down into native Tensorflow that allows the user to create fast performing machine learning models without worrying about the exact Tensorflow syntax, which can get quite complicated at times. Review collected by and hosted on G2.com.
Sometimes, if you want to do something custom, you really can't unless you go down to the base Tensorflow layer and add raw Tensorflow code, which isn't too great. There should be greater extensibility within Keras itself. Review collected by and hosted on G2.com.
Fast and flexible way of testing and building different networks, easy loading and saving of model and training with a similar interface to sklearn. Review collected by and hosted on G2.com.
When using with tensor board (with TF background) model looks clumsy, model is shown with additional blocks Review collected by and hosted on G2.com.
The best thing about Keras is it can work with the well known deep learning and machine learning libraries like TensorFlow and Theano. The API is very easy to use and one can create a whole CNN with less than 20 lines of code. Review collected by and hosted on G2.com.
It is the best deep learning library I have ever found out for python, no dislikes. Review collected by and hosted on G2.com.
Averages based on real user reviews.
1 month
