1. Train a two layer simple perceptron on the financial data normalized
2. Use as measure standard IR measures, precission and recall, micro and macro
3. Load other statements for analysys, without performance data
4. Divide companies between underweight and overweight according to the predicted gain