Model with sqrt scaled strength

Code

library(tidyverse)
library(targets)
tar_source(here::here("R"))
tar_load(
  c(exp1_data_agg, exp2_data_agg),
  store = here::here("_targets")
)

let’s fit the sqrt version of the model

Code

est <- estimate_model(
  start = paper_params(),
  data = exp1_data_agg,
  two_step = TRUE,
  simplify = TRUE,
  lambda = 0.5
)

est

# A tibble: 1 × 8
   prop prop_ltm   rate   tau  gain deviance convergence fit       
  <dbl>    <dbl>  <dbl> <dbl> <dbl>    <dbl>       <int> <list>    
1 0.490    0.662 0.0317 0.427  6.98     145.           0 <srl_rcl_>

Code

exp1_data_agg$pred <- predict(est$fit[[1]], data = exp1_data_agg, group_by = c("chunk", "gap"), lambda = 0.5)

exp1_data_agg |>
  ggplot(aes(x = gap, y = p_correct, color = chunk)) +
  geom_point() +
  geom_line() +
  geom_line(aes(y = pred), linetype = "dashed") +
  facet_wrap(~itemtype)

exclude sp1

Code

est <- estimate_model(
  start = paper_params(),
  data = exp1_data_agg,
  two_step = TRUE,
  simplify = TRUE,
  exclude_sp1 = TRUE,
  lambda = 0.5
)

est

# A tibble: 1 × 8
    prop prop_ltm    rate   tau  gain deviance convergence fit       
   <dbl>    <dbl>   <dbl> <dbl> <dbl>    <dbl>       <int> <list>    
1 0.0978    0.575 0.00842 0.290  69.2     31.7           0 <srl_rcl_>

predict

Code

exp1_data_agg$pred <- predict(est$fit[[1]], data = exp1_data_agg, group_by = c("chunk", "gap"), lambda = 0.5)

exp1_data_agg |>
  ggplot(aes(x = gap, y = p_correct, color = chunk)) +
  geom_point() +
  geom_line() +
  geom_line(aes(y = pred), linetype = "dashed") +
  facet_wrap(~itemtype)

replicate with different starting params

Code

est <- estimate_model(
  start = paper_params(),
  data = exp1_data_agg,
  two_step = TRUE,
  simplify = TRUE,
  exclude_sp1 = TRUE,
  lambda = 0.5,
  priors = list(rate = list(mean = 0.2, sd = 0.01))
)

est

# A tibble: 1 × 8
   prop prop_ltm  rate   tau  gain deviance convergence fit       
  <dbl>    <dbl> <dbl> <dbl> <dbl>    <dbl>       <int> <list>    
1 0.688    0.352 0.156 0.468  3.51     83.4           0 <srl_rcl_>

Code

exp1_data_agg$pred <- predict(est$fit[[1]], data = exp1_data_agg, group_by = c("chunk", "gap"), lambda = 0.5)

exp1_data_agg |>
  ggplot(aes(x = gap, y = p_correct, color = chunk)) +
  geom_point() +
  geom_line() +
  geom_line(aes(y = pred), linetype = "dashed") +
  facet_wrap(~itemtype)