Experiment to determine what is better: composing or tuning?

[valer1435]

Draft of experiment (may be extended)

1. Select 5 datasets for classification/regression/time-series-forecasting
2. For each dataset run such experiment 2.1) Run FEDOT for hour with set of parameters: [0% time for tuning, 25% time for tuning, 50% time for tuning, 75% time for tuning, 100% for tuning (use only initial assumptions)] 2.2) Run it 10-20 times for stability of results 2.3) Save and plot results (time for composing and tuning, metrics)