Variational auto-encoder for anomaly detection/features extraction, with lstm cells (stateless or statefull).
$ pip install --upgrade git+https://github.com/Danyleb/Lstm-Variational-Auto-encoder.git
from LstmVAE import LSTM_Var_Autoencoder
from LstmVAE import preprocess
preprocess(df) #return standardized and normalized df, check NaN values replacing it with 0
df = df.reshape(-1,timesteps,n_dim) #use 3D input, n_dim = 1 for 1D time series.
vae = LSTM_Var_Autoencoder(intermediate_dim = 15,z_dim = 3, n_dim=1, stateful = True) #default stateful = False
vae.fit(df, learning_rate=0.001, batch_size = 100, num_epochs = 200, opt = tf.train.AdamOptimizer, REG_LAMBDA = 0.01,
grad_clip_norm=10, optimizer_params=None, verbose = True)
"""REG_LAMBDA is the L2 loss lambda coefficient, should be set to 0 if not desired.
optimizer_param : pass a dict = {}
"""
x_reconstructed, recons_error = vae.reconstruct(df, get_error = True) #returns squared error
x_reduced = vae.reduce(df) #latent space representationPull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.
Tutorial on variational Autoencoders
A Multimodal Anomaly Detector for Robot-Assisted Feeding Using an LSTM-based Variational Autoencoder
Variational Autoencoder based Anomaly Detection using Reconstruction Probability
The Generalized Reparameterization Gradient
- Notebook with an example :)
- Probability of reconstruction
- Reparametrization trick for non-normal densities