does seem to go up over time
look at that above chance accuracy for everything
versus truth
looks reasonable – confusing the two kneeling-ish ones with each other, the two less feature-y ones
decent correlation, model is selectively bad at a couple
## [[1]]
##
## [[2]]
## [1] "Correlation between mlp best model and human 0.603"
Item level correlation between correct and wrong responses (model & per-human response)
## [[1]]
##
## [[2]]
##
## Pearson's product-moment correlation
##
## data: for_corr$human_correct and for_corr$model_correct
## t = 16.167, df = 7291, p-value < 2.2e-16
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
## 0.1637851 0.2080999
## sample estimates:
## cor
## 0.1860371
##
##
## [[3]]