Support for scipy.sparse matrices

[Defasium]

Hi, @olegranmo!
It would be very cool if methods fit, predict and transform could accept scipy.sparse matrices like most of the sklearn api models: LogisticRegression, MultinomialNB, RandomForest, etc...
For example in
https://github.com/cair/tmu/blob/main/examples/MNISTDemo.py#L18
Converting binarized X_train into scipy.sparse.csr_matrix can lower RAM consumption by a large factor.
This would be very convinient in case of relatively large datasets (with over 1 million examples). Or when there are a lot of features (like high-res images).

[olegranmo]

Great point, @Defasium! Will add support for sparse matrixes at the first opportunity. Currently, the class TMAutoEncoder uses sparse input matrixes to deal with large text datasets.