Classification.Rmd

---
title: "Classification"
author: "Xiao Yang"
date: "8/22/2019"
output: html_document
---

```{r setup, include=FALSE}
require(tidyverse)
require(rpart)
```

```{r}
dat = read_csv("data/outputTraining_11042019.csv") %>% 
  select(-`system:index`)
  # mutate(scl_int = factor(scl_int, levels = 1:4, labels = c("Thin ice", "Cloud", "Opaque ice/snow", "Water")))

dat %>% select(class) %>% distinct()

target_classes = c("opaque_ice", "clear_ice", "FSI_clouds", "water", "snow")

datfil = dat %>% 
  filter(class %in% target_classes) %>% 
  filter(fmask != 0) %>% 
  mutate(class_factor = factor(class, levels = target_classes, labels = c("Opaque Ice", "Clear Ice", "Cloud", "Water", "Snow")),
         class_factor_sim = factor(class, levels = target_classes, labels = c("Snow/Ice", "Snow/Ice", "Cloud", "Water", "Snow/Ice")),
         fmask_factor = factor(fmask, levels = c(3, 2, 4, 1), labels = c("Snow/Ice", "Cloud", "Cloud", "Water")),
         sensor_id = factor(substr(LANDSAT, start = 3, stop = 3), levels = c("5", "7", "8"), labels = c("TM", "ETM+", "OLI")))
  # filter(sensor_id == "TM")

datfil %>% ggplot() + geom_boxplot(aes(x = class_factor, y = Nir_stdDev))
datfil %>% ggplot() + geom_boxplot(aes(x = class_factor_sim, y = Swir1))
datfil %>% ggplot() + geom_boxplot(aes(x = class_factor_sim, y = Blue_log10_grad))


fit = rpart(class_factor_sim ~ Blue + Red + Green + Nir + Swir1 + Swir2 + Nir_stdDev + Blue_stdDev + Nir_log10_grad + Blue_log10_grad + WICI + hue + saturation + value + sensor_id, method = "class", data = datfil)

fitf = randomForest(class_factor_sim ~ Blue + Red + Green + Nir + Swir1 + Swir2 + Nir_stdDev + Blue_stdDev + Nir_log10_grad + Blue_log10_grad + WICI + hue + saturation + value + sensor_id, data = datfil)

varImpPlot(fitf, type = 2)
importance(fitf, type = 1)

# plot tree 
plot(fit, uniform=F, branch = 0, compress = F,
   main="Classification Tree for lake surface types")
text(fit, use.n=TRUE, cex=.8, all = F)

printcp(fit) # display the results 
plotcp(fit) # visualize cross-validation results 
summary(fit) # detailed summary of splits

# create additional plots 
# par(mfrow=c(1,2)) # two plots on one page 
rsq.rpart(fit) # visualize cross-validation results   


datfil %>% 
  filter(Swir1 < 0.0929) %>% 
  ggplot() +
  geom_point(aes(x = Blue, y = Blue_log10_grad, color = class_factor_sim), alpha = 0.3, size = 0.2) +
  facet_wrap(~class_factor_sim)

## confusion matrix
predicted_sim_class = predict(fit, newdata = datfil, type = "class")
reference_sim_class = datfil$class_factor_sim

caret::confusionMatrix(data = predicted_sim_class, reference = reference_sim_class)

fmask_sim_class = datfil$fmask_factor

caret::confusionMatrix(data = fmask_sim_class, reference = reference_sim_class)
```

## example of using R tree in GEE
https://code.earthengine.google.com/af98af92f61b76874d55b680e0382831

## lab

```{r}
require(party)

fit_ctree = ctree(scl_int ~ ., data = dat)
# plot(fit_ctree, main="Conditional Inference Tree for lake surface types")

```