2 min read

Tidy Models

2021-08-05

Exploring machine learning models using the Tidy Models package

Following the guidance from https://www.tidymodels.org/start/models/

library(tidymodels)  # for the parsnip package, along with the rest of tidymodels
library(tidyverse)       
library(broom.mixed) # for converting Bayesian models to tidy tibbles
library(dotwhisker)  # for visualizing regression results

Reading in the Urchin dataset

urchins <- read_csv("https://tidymodels.org/start/models/urchins.csv")  |>  
  setNames(c("food_regime", "initial_volume", "width")) |>  
  mutate(food_regime = factor(food_regime, levels = c("Initial", "Low", "High")))
## 
## -- Column specification ------------------------------------------------------------------------------------------------
## cols(
##   TREAT = col_character(),
##   IV = col_double(),
##   SUTW = col_double()
## )
glimpse(urchins)
## Rows: 72
## Columns: 3
## $ food_regime    <fct> Initial, Initial, Initial, Initial, Initial, Initial, I~
## $ initial_volume <dbl> 3.5, 5.0, 8.0, 10.0, 13.0, 13.0, 15.0, 15.0, 16.0, 17.0~
## $ width          <dbl> 0.010, 0.020, 0.061, 0.051, 0.041, 0.061, 0.041, 0.071,~

Getting a visual overview of the dataset

urchins |> 
  ggplot(aes(initial_volume, width, colour = food_regime))+
  geom_point()+
  geom_smooth(method = lm, se = FALSE)+
  theme_minimal()
## `geom_smooth()` using formula 'y ~ x'

Building a predictive linear model

linear.model <- linear_reg()  |>  
  parsnip::set_engine("lm")

linear.fit <- linear.model |> 
  parsnip::fit(width ~ initial_volume * food_regime, data = urchins)

linear.fit |> 
  tidy()
## # A tibble: 6 x 5
##   term                            estimate std.error statistic  p.value
##   <chr>                              <dbl>     <dbl>     <dbl>    <dbl>
## 1 (Intercept)                     0.0331    0.00962      3.44  0.00100 
## 2 initial_volume                  0.00155   0.000398     3.91  0.000222
## 3 food_regimeLow                  0.0198    0.0130       1.52  0.133   
## 4 food_regimeHigh                 0.0214    0.0145       1.47  0.145   
## 5 initial_volume:food_regimeLow  -0.00126   0.000510    -2.47  0.0162  
## 6 initial_volume:food_regimeHigh  0.000525  0.000702     0.748 0.457
tidy(linear.fit) |> 
  ggplot(aes(term, estimate, ymin = estimate - std.error, ymax = estimate + std.error, colour = term)) +
  geom_errorbar()+
  geom_point()+
  coord_flip()+
  theme_bw()+
  theme(legend.position = "none")

Using the model to make predictions

new_data <- expand.grid(initial_volume = 35, 
                          food_regime = c("Initial", "Low", "High"))

prediction <- predict(linear.fit, new_data = new_data)
prediction
## # A tibble: 3 x 1
##    .pred
##    <dbl>
## 1 0.0875
## 2 0.0632
## 3 0.127
prediction_conf <- predict(linear.fit, 
                         new_data = new_data, 
                         type = "conf_int")
prediction_conf
## # A tibble: 3 x 2
##   .pred_lower .pred_upper
##         <dbl>       <dbl>
## 1      0.0738      0.101 
## 2      0.0520      0.0745
## 3      0.105       0.150
plotting <- new_data |> 
  bind_cols(prediction) |> 
  bind_cols(prediction_conf)

plotting
##   initial_volume food_regime      .pred .pred_lower .pred_upper
## 1             35     Initial 0.08753404  0.07383931  0.10122876
## 2             35         Low 0.06323846  0.05202294  0.07445398
## 3             35        High 0.12733395  0.10514313  0.14952477
plotting |> 
  ggplot(aes(x = food_regime, y = .pred, colour = food_regime))+
  geom_point()+
  geom_errorbar(aes(ymin = .pred_lower, ymax = .pred_upper), width = 0.3)+
  theme_light()+
  theme(legend.position = 'none')+
  labs(title = "Size of Sea Urchins",
       y = "size of Urchin",
       x  = "Food Regime")

Using other engines