Common use of Training data Clause in Contracts

Training data. The three aforementioned aspects make up the training examples. However these are all ‘positive’ instances, participants whom were activated after the series of events. To complete the training data, ‘negative’ instances are needed as well. These are constructed from parts of the sequence of events that did not (yet) led to contact with the participant. As an example: a participant consecutively received Campaign A, got married and received Campaigns B and C and subsequently logged in to his personal page (see Figure 1a). The training data then includes four sequences for this participant, three with no response and one with response (see Figure 1b). (a) Example of a sequence of events. (b) Transforming sequence of events into four records for training data. Figure 1: Construction of training data. Randomly 50,000 participants were selected out of those whom satisfied the requirements of being passive for at least two years before seeking contact with the fund. This resulted in 135,189 potential training sequences (an average of 2.7 sequences per participant and events leading up to activation). Table 2 displays the number of sequences that resulted in an activated participant versus the number of sequences that did result in activation. Activated Number of Sequences Yes 50,000 (37.0%) No 85,189 (63.0%) Total 135,189 (100%)