Clean set_3 data

[ninanikitina]

• remove outliers
• split data train/test = 80/20
• create a script that loads data and extract dataset for a specific frame

[ninanikitina]

Timing: ~ 4h

It took me so long time since I wanted to use sklearn.model_selection.train_test_split, but after installing scikit-learn, I got "ValueError:numpy.ndarray size changed, may indicate binary incompatibility." error. So I ended up installing a new environment.