part1&2.Rmd

---
title: "hw07"
author: "shiyi"
date: "2/27/2022"
output: html_document
---

```{r setup, include=FALSE}

library(tidyverse)
library(tidymodels)
library(rcfss)



```


# This is part one, Student debt


```{r spiltdata}
set.seed(100)

data("scorecard") 
scorecard <- scorecard %>%
  select(-all_of(c("unitid","name"))) %>%
  modify_if(is.character, as.factor)

#spilt it
score_spilt <- initial_split(scorecard, prop = 3/4)
score_train <- training(score_spilt)
score_test <- testing(score_spilt)

```




```{r trianmodel}


#train model
line_model <- linear_reg() %>%
  set_engine("lm") %>%
  set_mode("regression")

line_model %>%
  fit(debt ~ . - state, 
      data = score_train) %>%
  predict(new_data = score_test) %>%
  mutate(truescore = score_test$debt) %>%
# report RMSE as required
  rmse( truth = truescore, estimate = .pred)

```
## linear regression
Root-mean_square deviation means deviation of the estimation of the model
RMSE for the  model is 3108.3261071


```{r fold10}

fold10 <- vfold_cv(score_train , v = 10, strata = debt)

line_model %>%
  fit_resamples(debt ~ . - state, resamples = fold10) %>%
  collect_metrics()

```

## cross-valid
RMSE for the ten-fold cross-validated model is 3108.3261071

```{r fold}

fold10 <- vfold_cv(score_train, v = 10, strata = debt)

line_model %>%
  fit_resamples(debt ~ . - state, resamples = fold10) %>%
  collect_metrics()

```

```{r decision_treee}
treemodel <- decision_tree() %>%
  set_engine( engine = "rpart") %>%
  set_mode (mode = "regression")

treemodel %>%
 fit_resamples(debt ~ . - state, resample = fold10)%>%
  collect_metrics()


```
## Tree model
RMSE for the tree model ten-fold cross-validated model is 3108.3261071

# Part 2 Predicting attitudes towards racist college professors

``` {r data2}

data("gss", package = "rcfss")
data("gss", package = "rcfss")
final_data <- gss %>%
  select(-all_of(c("id","wtss")))  

```

``` {r spilt}
set.seed(100)

gss_split <- initial_split(gss, strata = colrac, prop = 3/4)
gss_train <- training(gss_split)
gss_test <- testing(gss_split)


```


``` {r  logistic_regression_model}

#logistic regression model
lrmod <- logistic_reg(mode = "classification") %>% 
  set_engine(engine = "glm") 
# cross validation
gss_folds <- vfold_cv(data = gss_train, v = 10, strata = colrac)
# train it
lrmod %>%
  fit_resamples(colrac ~ age + black + degree + partyid_3 + sex + south, resamples = gss_folds) %>% 
  collect_metrics()

```

##accuracy
accuracy of the model is 0.5291205, roc_auc is 0.5475663		.


```{r forest}


# cross validation
gss_folds <- vfold_cv(gss_train, v = 10, strata = colrac)

forest_mod <- rand_forest(
  mode = "classification",
  engine = "ranger"
)

#recipe
gss_rec <- recipe(colrac ~ ., data = gss_train) %>%
        update_role(id, wtss, new_role = "ID") %>%
        step_naomit(colrac, skip = TRUE) %>%
        step_impute_median(all_numeric_predictors()) %>%
        step_impute_mode(all_nominal_predictors()) 

# workflow
workflow() %>%
  add_recipe(gss_rec) %>%
  add_model(forest_mod) %>%
  fit_resamples(resamples = gss_folds) %>% 
  collect_metrics()


```

##accuracy for  forest model 
accuracy of the model is 0.8066150, for roc_auc is 0.8851214

```{r neigh }


knn_mod <- nearest_neighbor(neighbors = 5) %>%              
  set_engine("kknn") %>%             
  set_mode("classification")

gss_rec_nei <- recipe(colrac ~ ., data = gss_train) %>%
  update_role(id, wtss, new_role = "ID") %>% 
  step_naomit(colrac, skip = TRUE) %>% 
  step_medianimpute(all_numeric()) %>%
  step_modeimpute(all_nominal(), -all_outcomes()) %>% 
  step_normalize(all_numeric()) 

gss_wf_nei <- workflow() %>% 
  add_recipe(gss_rec_nei) %>% 
  add_model(knn_mod) 


gss_wf_nei %>%
  fit_resamples(resamples = gss_folds) %>% 
  collect_metrics()


```
##accuracy for  five-nearest neighbors model
accuracy for  five-nearest neighbors model is accuracy for  five-nearest neighbors model,
roc_auc is 0.7992496

``` {r ridge logistic regression model }

ridge_mod <- logistic_reg(mode = "classification",
                          penalty = .01,
                          mixture = 0) %>% 
  set_engine(engine = "glm")

#workflow, tired of naming it
gss_wf3 <- workflow() %>% 
  
#just use previous one
  add_recipe(gss_rec_nei) %>% 
  add_model(ridge_mod) 

# training the model
gss_wf3 %>%
  fit_resamples(resamples = gss_folds) %>% 
  collect_metrics()


```

## ridge model

ridge logistic regression model, accuracy is accuracy, roc_auc is 0.8558371

```{r best}

rec_best <- recipe(colrac ~ ., data = gss_test) %>%
  update_role(id, wtss, new_role = "ID") %>%
  step_naomit(colrac, skip = TRUE) %>% 
  step_impute_median(all_numeric()) %>% 
  step_impute_mode(all_nominal(), -all_outcomes()) %>% 
  step_dummy(all_nominal(), -all_outcomes())

workflow() %>%
  add_recipe(rec_best) %>%
  add_model(forest_mod) %>%
  fit(data = gss_train) %>%
  predict(new_data = gss_test) %>%
  mutate(true_racist = gss_test$colrac) %>%
  accuracy(truth = true_racist, estimate = .pred_class)

```

## final

accuracy is 0.7943925