Dear fellow researchers I would like to know the best library to do a neural network modeling in Python. I know programming (Python included) to an intermediate level. I have used MATLAB NN toolbox and I loved the interface and all but when I got to some advanced algorithm implementations, I faced some constraints owing to the fact that I could not edit the source code of the in-built functions.After wrestling with the interface I gave up and I am considering shifting to Python. I know Python and can operate the Scipy array functions. However, there are many machine learning libraries in Python like Pybrain, Lmfit, Scikit learn and so so. My requirement is very small, it is just modeling a NN model to predict an output from a number of inputs. I learned MATLAB NN toolbox quite easily because it had a manual enriched with examples and all. However, for Python all i found was documentation with links directing me to GIThub to Stack overflow and the likes where I got confused with way to advanced levels. If someone has such an experience, please suggest me an appropriate library and some kind of a systematic way to learn it. I learned Python from Zed Shaw's The hard way series. I am looking for something nicely documented like that. Sorry for the long post. Thank you immensely.
My personal preference would be the Theano package in Python. It would function as a black box as in the other machine learning packages. So you do have to do your fair share of code. But this tutorial should be well worth the effort.
http://deeplearning.net/tutorial/mlp.html
It explains ANN from scratch along with all the code you require to run a fully fledged program on it. Its step-by-step instruction would make you learn everything you need to know to be able to tweak its parameters.
It would be advisable to start from here
http://deeplearning.net/software/theano/
What's great about this implementation is that all your code is optimized transparently and can even be better if you have a NVIDIA GPU with almost no change to the code.
Thank you for the prompt answer Ebenezer. I will look into the suggested sources.
My recommendation will be the keras library. It is actually an application layer or API that is used on top of Tensorflow, Theano or CNTK. You can specify which kernel to work with.
The cool thing about keras is that it is easy to use. You don't need to define variables and placeholders.
You will find the documentation here https://keras.io/
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python