Table 4.
Comparison of mean score for the Q-learning network model when only certain types of inputs are provided relative to baseline. The bank A, B, and C refer to the different types of input illustrated in Figure 5 (i.e., the continuous state cue, the 11 discrete state cues, or the reward from the previous trial, respectively). Again, positive values indicate conditions where the tested model provided a better fit than did the standard while negative value indicate a worse fit. In parentheses is the percentage of participants for whom the model provides a better fit than baseline. The top row (Bank A, B, & C) refers to the full model reported in Table 2.
| Experiment 1 | |||
|---|---|---|---|
| Provided Inputs | no-cue | shuffled-cue | consistent-cue |
| Bank A, B, & C | 93.2 (0.94) | 166.7 (0.94) | 118.2 (0.94) |
|
| |||
| No Inputs | −80.2 (0.00) | −81.7 (0.00) | −189.7 (0.00) |
|
| |||
| Bank A & B only | −80.2 (0.00) | 108.8 (0.94) | 73.8 (0.88) |
| Bank A Only | −80.2 (0.00) | −17.5 (0.59) | 61.8 (0.88) |
| Bank B Only | −80.2 (0.00) | 93.9 (0.94) | −3.1 (0.47) |