Yet ANother autoencoder toolbox based on Theano
We implemented the autoencoder (AE) algorithm in Python based on the Theano library. The materials in http://deeplearning.net/ helped a lot.
yan_ae.py: the shallow AE. Various strategies can be enabled or disabled, including denoising, sparsity, L1/L2 regularization, and weight tying strategies. One can also config the loss function, activation function, and training algorithm (e.g. minibatch, CG).
yan_sae.py: the stacked AE (SAE), built upon the shallow ones. Methods include layer-wise pretraining, unsupervised and supervised fine-tuning.
yan_mlp.py: multi-layer perceptron (fully connected network), mainly implemented for fine-tuning of SAE.
yan_msda.py: marginalized denoising autoencoder.
dcae.py: drift correction autoencoder (DCAE), designed for correction of instrumental variation and time-varying drift in sensor systems. Please see the ref for details.
test_ae.py: test SAE and MSDA on benchmark datasets.
**test_corn.py: test DCAE on the corn dataset.
test_uci.py: test DCAE on the gas sensor array drift dataset. Description of these two datasets is in datasets_readme.txt.
.idea: project files of PyCharm.
ref: Ke Yan, and David Zhang, "Correcting Instrumental Variation and Time-varying Drift: A Transfer Learning Approach with Autoencoders, " accepted by Instrumentation and Measurement, IEEE Transactions on
Copyright 2016 YAN Ke, Tsinghua Univ. http://yanke23.com , [email protected]