SDAE_pytorch

Training and visualization

For training and visualization, go into folder SDAE_pytorch. Place the training data (observation.data) in dataset folder, and use following commands for training. SDAE model is trained and stored as chept.pt in folder model, and will be used for visualization and feature extraction.

Training:

usage: python train.py [-h] [--epoch EPOCH] [--data_size DATA_SIZE] [--lr LR]
                [--workers WORKERS] [--batchSize BATCHSIZE] [--in_dim IN_DIM]
                [--out_dim OUT_DIM] [--stack_num STACK_NUM]
                [--noise_r NOISE_R]

optional arguments:
  -h, --help              show this help message and exit
  --epoch EPOCH           number of training epochs
  --data_size DATA_SIZE   size of training data
  --lr LR                 learning rate, default=0.0002
  --workers WORKERS       number of data loading workers
  --batchSize BATCHSIZE   input batch size
  --in_dim IN_DIM         input dimension
  --out_dim OUT_DIM       output(feature) dimension
  --stack_num STACK_NUM   number of hidden layers
  --noise_r NOISE_R       Ratio of noise

Visualization:

usage: python visualize.py [-h] [--data_size DATA_SIZE]

optional arguments:
  -h, --help              show this help message and exit
  --data_size DATA_SIZE   size of data used for visualization

Feature extraction

Place the module in the root folder of the project. Use from SDAE_pytorch.extract import Autoencoder to import the feature extraction class.

Example:

from SDAE_pytorch.extract import Autoencoder
import numpy as np

SDAE = Autoencoder()

for i in range(10):
	dumb_data = np.random.randn(48)
	dumb_feature = SDAE.extract(dumb_data)

	print(dumb_feature)