Hi,
Yes, source code of my library for Kaggle is at
https://github.com/Arcady27/KaggleLib . Sorry that I haven't yet enough
time to write good docs for it, refactor and fix all bugs, but
nevertheless the main functionality can be seen and I already
successfully used it in several Kaggle contests.
Q-learning with neural net for Flappy Bird can be found at
https://github.com/Arcady27/Q-Learning-Flappy-Bird. It's rather
difficult game compared to some Atari ones because it needs the
precision of actions taken by agent, that's why it takes a long to learn
(as I remember, several hours at least to gain >10 points).
Nevertheless, it is interesting to watch how the bird is learning at
first 10-20 minutes (score is raising to 2-3-4 points from 0).
I am quite equally interested in both of RL-project and
cross-validation/models tuning one, maybe a little bit more excited
about RL. What project would you advice me to write proposal for? Thanks!
Arkadiy.
On 21.03.2017 17:17, Marcus Edel wrote:
Hello Arkadiy,
welcome and thanks for the introduction.
I trained neural network (using Keras framework for Python) to play
famous game
Flappy Bird much better than a human can, that's why I am very
interested in
GSOC project connected with reinforcement learning. Also I
participate in Kaggle
competitions and coded my own small library in Python with most useful
functions, such as cross-validation, parameters tuning and stacking
models.
That sounds really interesting, is the code somewhere so that we could
take a
look and probably play with it?
Also, the Reinforcement learning and cross-validation project has
been discussed at on the mailing list before:
http://mlpack.org/pipermail/mlpack/2017-March/003095.html
http://mlpack.org/pipermail/mlpack/2017-March/003098.html
http://mlpack.org/pipermail/mlpack/2017-February/003087.html
http://mlpack.org/pipermail/mlpack/2017-March/003212.html
Note that there are many more posts on this in the mailing list archive
to search for; those are only some places to get started.
I hope this is helpful,
Thanks,
Marcus
On 21 Mar 2017, at 02:18, Arkadiy Dushatskiy <[email protected]
<mailto:[email protected]>> wrote:
Dear developers, I am very interested in participating in GSOC-2017
by working at one of your exciting projects.
I am a student from Moscow State University, Faculty of
Computational Mathematics and Cybernetics, currently pursuing my 1st
year of a Master’s degree in Applied Mathematics and Informatics. My
main areas of interest areArtificial Intelligence and Machine
Learning. The topic of my scientific research is GPU-accelerated
neural networks. It is focused on image recognition, I investigate
how the architecture of convolutional neural network and GPU
characteristics affect the time consumption of forward propagation
for a batch of images. The main technologies I use in scientific work
are CUDA and cuDNN libraries (both for C++) for programming neural
networks and Python for results visualization.
I have experience with reinforcement learning tasks, for
instance, inspired by DeepMing articles, I trained neural network
(using Keras framework for Python) to play famous game Flappy Bird
much better than a human can, that's why I am very interested in GSOC
project connected with reinforcement learning. Also I participate in
Kaggle competitions and coded my own small library in Python with
most useful functions, such as cross-validation, parameters tuning
and stacking models. So, I am also interested in project about
creating API for cross-validation and models tuning in mlpack.
Thank you for your attention!
_______________________________________________
mlpack mailing list
[email protected] <mailto:[email protected]>
http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack
_______________________________________________
mlpack mailing list
[email protected]
http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack
