It's a little late, but I think it can't be excluded from this year's Python keywords. So, looking back on this year, I would like to introduce the Deep Learning Python library that was taken care of.
Keras Various frameworks such as chainer, tensorflow, mxnet are available. However, if you simply look at the number of stars on Github, it seems to be popular in the following order. (As of December 25)
You can see that tensorflow is very popular. I like keras because it requires less lines of code, but tensorflow-slim has also appeared in the future. It seems that there will be a library that can be written more easily based on tensorflow.
keras-rl Deep Reinforcement Learing library using keras. It corresponds to the environment of OpenAI Gym. The following basic deep reinforcement learning algorithms are implemented.
Keras also has a deep reinforcement learning library called kerlym that supports OpenAI Gym. In the tensorflow based library, there is rllab. This is made by OpenAI, is large in scale, and has the following algorithms implemented in addition to the above.
This one seems to be good, but keras-rl is simpler, so I use it.
keras-resnet Resnet has made it possible to dramatically increase the layer of the network. Even quite a lot of people have implemented Github. Not limited to keras, I will list the main implementations.
GAN GAN was also popular. It seems that new GANs are proposed one after another and will be implemented on Github soon, so the competition is fierce.
There was also a Keras GAN-only library. It seems that the model definition and learning method for GAN are easier to do.
quiver There are various drawing tools such as learning results. quiver draws Convnet in Keras.
Other drawing tools.
I personally like Keras, so I have collected tools that interest me around Keras.
Recommended Posts