This repo is for brain inspired course project, with the file structure as follows:
.
├── README.md
└── brain-inspired-project
├── SGD
│ ├── __pycache__
│ │ ├── gradient_2d.cpython-36.pyc
│ │ └── two_layer_net.cpython-36.pyc
│ ├── gradient_1d.py
│ ├── gradient_2d.py
│ ├── gradient_method.py
│ ├── gradient_simplenet.py
│ ├── train_neuralnet.py
│ ├── train_neuralnet_mlp.py
│ └── two_layer_net.py
├── common
│ ├── __init__.py
│ ├── __pycache__
│ │ ├── __init__.cpython-36.pyc
│ │ ├── functions.cpython-36.pyc
│ │ ├── gradient.cpython-36.pyc
│ │ └── util.cpython-36.pyc
│ ├── functions.py
│ ├── gradient.py
│ ├── layers.py
│ ├── multi_layer_net.py
│ ├── multi_layer_net_extend.py
│ ├── optimizer.py
│ ├── trainer.py
│ └── util.py
└── dataset
├── __init__.py
├── __pycache__
│ ├── __init__.cpython-36.pyc
│ └── mnist.cpython-36.pyc
├── lena.png
├── lena_gray.png
├── mnist.pkl
├── mnist.py
├── t10k-images-idx3-ubyte.gz
├── t10k-labels-idx1-ubyte.gz
├── train-images-idx3-ubyte.gz
└── train-labels-idx1-ubyte.gz
7 directories, 34 files
This finishes three topics:
- Topic A: How does the sample size influence the smoothness of the loss function.
- Topic B: How does the depth of network influence the smoothness of the loss function.
- Topic C: How does the number of iterations influence the smoothness of the loss function.
The dataset is just MNIST, and it is included in the dataset directory.
The main program is SGD/train_neuralnet_mlp.py. For Topic B and Topic C, please modify the parameters in main program. However, for Topic A, which relates to sample input size, please modify the parameters in dataset/mnist.py.
- matplotlib 2.2.2
- numpy 1.14.1
- sklearn 0.22