0
$\begingroup$

I wish to apply non-linear dimensionality reduction on a very small dataset (less than 100 observations). The dataset is very sparse, of approx 20 columns, each containing either 0 or 1. It's the result of a set of non-exclusive on-hot encoding of features. Observations are of this kind:

---------------------------------
| a | b | c | d | e | f | g | h |
---------------------------------
| 0   1   1   1   0   0   0   0 |
| 1   1   0   1   0   1   1   0 |
| 0   1   1   1   1   0   0   1 |

...

I tried with t-SNE, but its stochastic nature leads to very different results each time I run my code.

Any suggestions for what shall I use? I also need a technique that is strictly non-hierarchical (such as PCA, for example, in which each factors explains just the "leftovers" of the variance of the previous factor; I don't want that).

$\endgroup$
2
  • 1
    $\begingroup$ How about using an autoencoder to make it into a smaller dimensional latent space? $\endgroup$ – 1tan Oct 8 '19 at 2:18
  • $\begingroup$ I agree, that's my main choice at the moment. My "problem" is that it's such a small and simple task I thought using Deep Learning is an overkill... I hoped I could use a quicker tool, something more sklearn style $\endgroup$ – Leevo Oct 8 '19 at 7:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.