---
title: "Use `unifiedml` with caret without tuning"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Use `unifiedml` with caret without tuning}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r fig.width=7}
library(unifiedml) # this package
require(caret)

set.seed(123)
```

# Random Forest

```{r fig.width=7}
iris_binary <- iris[iris$Species %in% c("setosa", "versicolor"), ]
X_binary <- iris_binary[, 1:4]
y_binary <- as.factor(as.character(iris_binary$Species))  # factor → classification

datasplit <- unifiedml::train_test_split(X_binary, y_binary, 
                                         test_size = 0.3, seed = 42)

mod <- Model$new(caret::train) 
mod$fit(datasplit$X_train, datasplit$y_train,
        method = "rf",
        trControl = caret::trainControl(method = "none"))
print(head(mod$predict(datasplit$X_test)))
print(head(mod$predict(datasplit$X_test, type="prob")))
```


# Logistic Regression with `glmnet`

```{r fig.width=7}
X <- iris[, 1:4]
y <- iris$Species  # factor → classification

datasplit <- unifiedml::train_test_split(X, y, 
                                         test_size = 0.3, seed = 42)

mod <- Model$new(caret::train) 
mod$fit(datasplit$X_train, datasplit$y_train,
        method = "glmnet",
        tuneGrid = data.frame(alpha = 0,  # ridge regression
                               lambda = 0.01),  # fixed lambda
        trControl = caret::trainControl(method = "none"))
print(head(mod$predict(datasplit$X_test)))
print(head(mod$predict(datasplit$X_test, type="prob")))

(cv <- cross_val_score(mod, datasplit$X_train, datasplit$y_train, cv = 5L, 
                       fit_params=list(method = "glmnet",
                       tuneGrid = data.frame(alpha = 0,  # ridge regression
                       lambda = 0.01),  # fixed lambda
                       trControl = caret::trainControl(method = "none"))))  # Auto-uses accuracy
cat("\nMean Accuracy:", mean(cv), "\n")


(cv <- cross_val_score(mod, datasplit$X_train, datasplit$y_train, cv = 5L, 
                       fit_params=list(method = "glmnet",
                       tuneGrid = data.frame(alpha = 0.5,  # ridge regression
                       lambda = 0.01),  # fixed lambda
                       trControl = caret::trainControl(method = "none"))))  # Auto-uses accuracy
cat("\nMean Accuracy:", mean(cv), "\n")
```