Tabular Autoencoder

Upload a table of training samples

Output sample importances (optional)

Samples are: rows columns

Autoencoder specs:

features encoder decoder

extra (variational) features
uniformly distributed (0,1)
normally distributed (0 mean, σ=1)

Architecture options

Max weights:
Max neurons:
Max layers:
Max layer skip:

Sparse weights
Include bias

Instructions: This tool builds neural networks that:
1) extract features from data (encoder);
2) reconstruct source data from features (decoder).

Upload a table^* of sample source data. These examples will train the network. Specify whether each sample is a row or column in the table and click Submit.
The trained network opens in a new page, where you can a) run the encoder/decoder, and b) export it as computer code for use in your own programs (no attribution needed).

Try it using example data

* Sample table must be a fully numeric, comma-separated .csv file with no empty fields. Max allowed: 100 kB file size, 1000 rows/columns, 1 header row/column. Generation time limited to 1 sec.