Current implementation of calibrate_no_booking uses following formula to calculate weights of products with no observed demand:
Wi = Pi * (1 - Ai)
where Pi is a probability of a customer selecting a product i, Ai is availability of the product i.
This formula penalizes products with high availability and small probability but it favours products with small availability even if their probability is low. For example, let P1 = 0.5, A1 = 0.9, P2 = 0.1 and A2 = 0, in this case W1 = 0.05 and W2 = 0.1. This means that product 2 demand will be higher than product 1 demand even the probability of this product being selected by a customer is much smaller than probability of the product 1.
Current implementation of
calibrate_no_bookinguses following formula to calculate weights of products with no observed demand:Wi = Pi * (1 - Ai)where Pi is a probability of a customer selecting a product i, Ai is availability of the product i.
This formula penalizes products with high availability and small probability but it favours products with small availability even if their probability is low. For example, let P1 = 0.5, A1 = 0.9, P2 = 0.1 and A2 = 0, in this case W1 = 0.05 and W2 = 0.1. This means that product 2 demand will be higher than product 1 demand even the probability of this product being selected by a customer is much smaller than probability of the product 1.