Processing of missing data by neural networks

This is the training code for our paper "Processing of missing data by neural networks".

In order to repeat the experiments from column "our":

in Figure 2:
```
    python ae.py
```
This code will download MNIST database, run training with default (parameters) and test reconstruction on random test images. In order to use non-default parameters, modify initial lines in file "ae.py".
in Table 3 (except the first row):
```
    python rbfn.py --data_dir ./data_rbfn/bands/ --learning_rate 0.25 --batch_size 75 --training_epochs 500 --n_hidden_1 25 --n_distribution 3
```
This command works for data set "bands", other dat sets attached to this code are as follows:
- hepatitis
- horse
- kidney_disease
- mammographics
- pima
- winconsin

marblet/Processing-of-missing-data-by-neural-networks