---
title: "exam2"
author: "Mark Pearl"
date: "7/28/2021"
output: html_document
---

```{r import libraries}
library(glmnet)
library(tidyverse)
library(ggplot2)
library(caret)
library(glmnet)
library(tidyr)
library(dplyr)
library(ggrepel)
```

```{r read data and split into train/test}
beans <- read_csv('./Question1.csv')

#Split data into test and train
train <- beans[0:2800,]
test <- beans[2801:3871,]

#Modify class variable to map back to a factor
train <-
  train %>% mutate(ClassFactor = case_when(
        Class == 1 ~ "Barbunya",
        Class == 2 ~ "Bombay",
        TRUE ~ "Seker"))

train$Class <- as.factor(train$Class)

test <-
  test %>% mutate(ClassFactor = case_when(
        Class == 1 ~ "Barbunya",
        Class == 2 ~ "Bombay",
        TRUE ~ "Seker"))

test_x <- test[,0:16]
y_actual <- test[,17]
```

```{r part 1 muliti-nomial logistic regression}
#Use glmnet function and plot the results
fit <- glmnet(x = as.matrix(train[,0:16]),y = train$Class, family = "multinomial",alpha=0,lambda=0)
options(scipen=999)
fit$beta
```

```{r part 1 muliti-nomial logistic regression}
create_ggplot <- function(beta_index, lambda_var) {
  betas = matrix(mod$glmnet.fit$beta)[beta_index,][[1]]
  lambdas = mod$lambda
  names(lambdas) = betas@Dimnames[[2]]
  as.data.frame(as.matrix(betas)) %>% 
  tibble::rownames_to_column("variable") %>% 
  pivot_longer(-variable) %>% 
  mutate(lambda=lambdas[name]) %>% 
  ggplot(aes(x=lambda,y=value,col=variable)) + 
  geom_line() + 
  geom_vline(xintercept = lambda_var,color='red')+
  geom_label_repel(data=~subset(.x,lambda==min(lambda)),
  aes(label=variable),nudge_x=-0.5) +
  scale_x_log10()
}
```


```{r part 1 muliti-nomial logistic regression get prediction / confusion matrix result}
cnf <- confusion.glmnet(fit, newx=as.matrix(test_x), newy = y_actual[[1]])
cnf
```


```{r part 2 ridge muliti-nomial logistic regression}
#Create cv.fit object for ridge regression
#p=9
mod <- cv.glmnet(x = as.matrix(train[,0:16]),y = train$Class, family = "multinomial",alpha = 0, nfolds=20, grouped = FALSE,type.measure = 'class')
lambda_ridge = mod$lambda.min
ridge_final = glmnet(as.matrix(train[,0:16]), y = train$Class, family = "multinomial",alpha = 0,lambda=lambda_ridge)
ridge_final$beta
```
```{r get output ridge}
create_ggplot(1, lambda_ridge)
create_ggplot(2, lambda_ridge)
create_ggplot(3, lambda_ridge)
```


```{r part 1 muliti-nomial logistic regression get prediction / confusion matrix result}
cnf_ridge <- confusion.glmnet(ridge_final, newx=as.matrix(test_x), newy = y_actual[[1]])
cnf_ridge
```

```{r part 3 lasso muliti-nomial logistic regression}
#Create cv.fit object for ridge regression
#p=9
mod <- cv.glmnet(x = as.matrix(train[,0:16]),y = train$Class, family = "multinomial",alpha = 1, nfolds=20)
lambda_lasso = mod$lambda.min
lasso_final = glmnet(as.matrix(train[,0:16]), y = train$Class, family = "multinomial",alpha = 1,lambda=lambda_lasso)
lasso_final$beta
```

```{r get output lasso}
create_ggplot(1, lambda_lasso)
create_ggplot(2, lambda_lasso)
create_ggplot(3, lambda_lasso)
```

```{r part 1 muliti-nomial logistic regression get prediction / confusion matrix result}
cnf_lasso <- confusion.glmnet(lasso_final, newx=as.matrix(test_x), newy = y_actual[[1]])
cnf_lasso
```
```{r adaptive lasso}
gamma = 0.5
b.ridge = do.call(cbind, coef(ridge_final))
w1 <- 1 / abs(as.matrix(b.ridge)[-1,])^gamma
mod <- cv.glmnet(x = as.matrix(train[,0:16]),y = train$Class,nfold = 20,family = "multinomial",alpha = 1,penalty.factor = rowNorms(w1))
lambda_alasso = mod$lambda.min
alasso_final = glmnet(as.matrix(train[,0:16]), y = train$Class, family = "multinomial",alpha = 1,lambda=lambda_alasso)
alasso_final$beta
```

```{r get output lasso}
create_ggplot(1, lambda_alasso)
create_ggplot(2, lambda_alasso)
create_ggplot(3, lambda_alasso)
```

```{r part 1 muliti-nomial logistic regression get prediction / confusion matrix result}
cnf_alasso <- confusion.glmnet(alasso_final, newx=as.matrix(test_x), newy = y_actual[[1]])
cnf_alasso
```