32 lines
1.3 KiB
Markdown
Executable File
32 lines
1.3 KiB
Markdown
Executable File
# MINDLE Is Not a Deep Learning Experimentation
|
|
|
|
## Requirements
|
|
|
|
### For new C++ codes
|
|
|
|
- g++ >= 4.9.2 (of course, we won't upload any binaries here)
|
|
- libsfml-dev >= 2.4.1 (for SFML library)
|
|
- libeigen3-dev >= 3.3.2 (for Eigen library)
|
|
|
|
### For old Python codes
|
|
|
|
- Python 3.5.2
|
|
- Theano 0.8.2
|
|
- Numpy 1.11.2
|
|
- Matplotlib 1.5.3
|
|
- TensorFlow 0.11.0rc0
|
|
|
|
## Usage (examples)
|
|
|
|
1. Clone the repository : `git clone https://github.com/HorlogeSkynet/MINDLE.git`
|
|
2. Move into the folder : `cd MINDLE/`
|
|
3. Compile the C++ : `make`
|
|
4. Run the C++ programs : `./bin/linearRegression.out`
|
|
5. Run the Python programs : `python3 dev_null/Theano/LinearRegression/linearRegression.py`
|
|
|
|
## Notes to read
|
|
|
|
1. Some words about the frameworks we used with Python in the past : Theano & TensorFlow gave us a too high-level abstraction for our needs. We couldn't manage to get something to work, and this is not about getting computations on the GPU; that was just impossible, and we nearly destroy our computers many times... Now, we start from scratch with C++. At least, we would understand the errors thrown.
|
|
|
|
2. Based on [this "tutorial"](https://github.com/Newmu/Theano-Tutorials), you'll be able to perform digits classification by getting the return of [loadMNIST.py](dev_null/MNIST/loadMNIST.py). If you run it, the code will download and read the training examples automatically.
|