Table 1 Overall model results, model comparison, and fitted parametric values.

From: Exploring Feature Dimensions to Learn a New Policy in an Uninformed Reinforcement Learning Task

 

Model Comparison

Model parameters

Log-likelihood

AIC

BIC

# of param

α

β

Learned val ini

−121.35 ± 32.05

246.70 ± 64.10

253.79 ± 64.10

2

0.25 ± 0.25

4.47 ± 5.19

Zero ini

−125.38 ± 32.79

254.77 ± 65.58

261.86 ± 65.58

2

0.22 ± 0.25

4.30 ± 2.83

Naïve + Softmax

−119.42 ± 30.30

266.85 ± 60.59

316.43 ± 60.59

14

0.22 ± 0.28

10.26 ± 19.25

Naïve + HMM

−123.30 ± 31.38

274.59 ± 62.77

324.17 ± 62.77

14

0.21 ± 0.26

14.69 ± 38.55

  1. Table of overall models, model comparison and fitted parametric values. Learned val ini: Value transfer learning model initialised with learned value. Zero ini: Value transfer learning model initialised with zero. Naïve + Softmax: Probabilistic policy exploration model with softmax function-based policy search. Naïve + HMM: Probabilistic policy exploration model with HMM based policy search. Log-likelihood value: Larger values indicate a better fit. AIC and BIC: Smaller values indicate a better fit. All values are represented as mean ± SD, except number of parameters. HMM: hidden Markov model; AIC: Akaike information criterion; BIC: Bayesian information criterion.
Back to article page