Best works with LearnCard: NRows ≳ 500 and 5 ≲ NCols ≲ 30
LearnCard should have less than 2000 rows and 30 columns, InCard can have at most 10k rows
First row is intended for titles and is always removed
Example 'House prices': Each row corresponds to a selling house features: number of bedrooms, bathrooms, their sizes, garage, etc. The last column in LearnCard is set to 1 if the house was overpriced (not sold in the first 3 months), and to 0 otherwise. Therefore, the Real State agent can use this learning to make his/her business work better for upcoming houses. See example LearnCard_House.xlsx and InCard_Houses.xlsx.
Example 'Increasing': Each row consists of random numbers which increase randomly by steps of 1 to 10. However, some times there is a fail and the number decreases instead of increasing. We call it category 1 if there is no fail, and category 0 if there is a fail. See example LearnCard_Increasing.xlsx and InCard_Increasing.xlsx.
Example 'Topological Islands': Each row consists of 1 and 0's; if all the 1's are always within only one compact island -with no 0's in between-, we call it category 1, otherwise, category 0. See example LearnCard_Islands.xlsx and InCard_Islands.xlsx.
The Neural Network consists in 2 layers with 32 Neurons each, optimized through SGD.