Rewritten Deep Embedded Clustering (DEC) and Improved DEC (IDEC) algorithms from keras to the current version of pytorch. Obtained code was implemented then for the purposes of text feature extraction. (See this paper and this repo for details). See .ipynb notebook for full code and plots.
Original papers:
- DEC Unsupervised Deep Embedding for Clustering Analysis (2016) by Junyuan Xie et al.
- IDEC Improved Deep Embedded Clustering with Local Structure Preservation (2017) by Xifeng Guo et al.
Perfomance of the algortihms was tested on the MNIST. For both DEC and IDEC cluster accuracy achieves around 85% (Hungarian algorithm, see scipy desc) and highly depends on accuracy of initially generated centroids (kmeans on the encoder outputs)
