R Notebook

library(tidyverse)
library(tidymodels)
tidymodels_prefer()

Load the Data

read_delim("../development_gene_expression.txt") -> data

# The predicted variable needs to be factor
data %>%
  mutate(Development=factor(Development)) %>%
  select(Development,everything()) -> data

# We want to randomly shuffle the rows so there is no structure
set.seed(123)
data %>%
  sample_frac() -> data

head(data)

We’re going to do proper cross validation so we need to split into 10 chunks

data %>%
  vfold_cv(v = 10) -> cv_data

Set the main options

number_of_trees_to_build = 100
random_predictors_per_node = 20
minimum_measures_per_node = 5

Random Forest

rand_forest(trees=number_of_trees_to_build, min_n=minimum_measures_per_node, mtry=random_predictors_per_node) %>%
  set_mode("classification") %>%
  set_engine("ranger") -> model

model %>% translate()

Train the model

model %>%
  fit_resamples(Development ~ ., resamples = cv_data) -> model_fit

model_fit

collect_metrics(model_fit) -> model_metrics

model_metrics