labs-14.Rmd

---
title: "Week 14 - SVM"
---

Download template [here](templates/labs-14.Rmd)

```{r setup, include=FALSE}
# Delete this chunk!
knitr::opts_chunk$set(eval = FALSE)
```

We will be using tidymodels and the `flights` data set from {nycflights13}.

```{r}
library(tidymodels)
library(nycflights13)
```

We will do the same transformation as we have done before.

```{r}
flights1 <- flights %>%
  mutate(delay = factor(arr_delay > 0, c(TRUE, FALSE),
                        c("Delayed", "On time"))) %>%
  filter(month == 1, !is.na(delay)) %>%
  select(delay, hour, minute, dep_delay, carrier, distance)

set.seed(1234)
flights_split <- initial_split(flights1)

flights_train <- training(flights_split)
flights_test <- testing(flights_split)
```

# The models

We will start using a `svm_linear()` model . These can be used for both regression and classification so we need to specify it for this model. We will be using the `kernlab` package as the engine.

```{r}
svm_lin_spec <- svm_linear() %>%
  set_mode("classification") %>%
  set_engine("kernlab")
```

and then we will fit it right away. The fitting might take a minute or two but we shouldn't worry.

```{r}
svm_lin_fit <- fit(svm_lin_spec, delay ~ ., data = flights_train)
svm_lin_fit
```

We can get the confusion matrix

```{r}
svm_lin_fit %>%
  augment(new_data = flights_train) %>%
  conf_mat(delay, .pred_class) %>%
  autoplot(type = "heatmap")
```

and calculate the accuracy.

```{r}
bean_tree %>%
  augment(new_data = beans_train) %>%
  accuracy(class, .pred_class)
```

They are not doing well.

Let us try a polynomial SVM model to see if that helps at all.

```{r}
svm_poly_spec <- svm_poly(degree = 2) %>%
  set_mode("classification") %>%
  set_engine("kernlab")

svm_poly_fit <- fit(svm_poly_spec, delay ~ ., data = flights_train)
svm_poly_fit
```

calculating another confusion matrix doesn't give us much luck.

```{r}
svm_poly_fit %>%
  augment(new_data = flights_train) %>%
  conf_mat(delay, .pred_class) %>%
  autoplot(type = "heatmap")
```

# Your turn

But wait, we didn't do any preprocessing. Let us do some proper preprocessing to see if we can improve on the model. We also have a `cost` parameter we could tune. Let us try that as well.