5_plot_results.rmd

---
title: Results of RCNN model to Detect Charcoal Hearths. 100 randomly selected, manually scored images.
output:
  html_document:
    df_print: paged
  html_notebook: default
  pdf_document: default
---

This is a notebook of results for selected models to analyse which one will be selected as the best.
 


```{r}

# Formulas from Carleton University DATA5000 lecture, Dr. James Green, Pattern Classification in the Presence of Class Imbalance
graph_result <- function(cfg_name, object_type, sample_type) {
  library("ggplot2")
  library("ggthemes")
  
  if(sample_type == ""){
    sample_100 = "_100"
    subtitle = "(all)"
  }else{
    sample_100 = ""
    subtitle = "(just training)"
  }
    
  image_prediction_folder <-
  paste0(
  '/home/student/',object_type,'/predictions/',
  cfg_name,'/unknown', sample_100, sample_type, '/edited/'
  )
  
  sample_file_name <- paste0(image_prediction_folder, "results_100_random_images", sample_type, "_", cfg_name, ".csv")
  print(sample_file_name)
  
  results <-
  read.table(
  sample_file_name,
  header = TRUE,
  sep = ",",
  dec = "."
  )
  results_df <- data.frame(results)
  
  g1p <- ggplot(results_df, aes(min_score)) +
  geom_line(aes(y = gt_obj_pres_pos , colour = "Ground Truth: Present"),
  linetype = 2) +
  geom_line(aes(y = tp , colour = "Prediction: TP")) +
  geom_line(aes(y = fp , colour = "Prediction: FP")) +
  theme_igray() +
  ggtitle(paste0(object_type," Present in Image. ",subtitle," Model: ", cfg_name)) +
  xlab("Minimum Prediction Confidence Score Used") +
  ylab("Number of Occurences")+
  coord_cartesian(ylim=c(0,100))
  
  g1n <- ggplot(results_df, aes(min_score)) +
  geom_line(aes(y = gt_obj_pres_neg , colour = "Ground Truth: Not Present"),
  linetype = 2) +
  geom_line(aes(y = tn , colour = "Prediction: TN")) +
  geom_line(aes(y = fn , colour = "Prediction: FN")) +
  theme_igray() +
  ggtitle(paste0(object_type," Not Present in Image. ",subtitle," Model: ", cfg_name)) +
  xlab("Minimum Prediction Confidence Score Used") +
  ylab("Number of Occurences")+
  coord_cartesian(ylim=c(0,100))

  
  g2 <- ggplot(results_df, aes(min_score)) +
  geom_line(aes(y = tp_box , colour = "Prediction Boxes: True Positive (TP)")) +
  geom_line(aes(y = fp_box , colour = "Prediction Boxes: False Positive (FP)")) + theme_minimal() +
  theme_igray() +
  ggtitle(paste0("Prediction Regions, True vs. False Positives. ",subtitle," Model: ", cfg_name)) +
  xlab("Minimum Prediction Confidence Score Used") +
  ylab("Number of Occurences")+
  expand_limits(y=0)+
  scale_y_continuous(expand=c(0,0))
  
  print(g1p)
  
  print(g1n)
  
  print(g2)
  
  results_df <- within(results_df, accuracy <- ((tp + tn) / (tp + tn + fp +
  fn)))
  results_df <- within(results_df, sn <- ((tp) / (tp + fn)))
  results_df <- within(results_df, sp <- ((tn) / (tn + fp)))
  results_df <- within(results_df, ppv <- ((tp) / (tp + fp)))
  results_df <- within(results_df, npv <- ((tn) / (tn + fn)))
  print(results_df)
  g3 <- ggplot(results_df, aes(min_score)) +
  geom_line(aes(y = accuracy , colour = "Accuracy"), linetype=2) +
  geom_line(aes(y = sn , colour = "Sensitivity/Recall/True Positive Rate")) +
  geom_line(aes(y = sp , colour = "Specificity")) +
  theme_igray() +
  ggtitle(paste0("Sensitivity and Specificity. ",subtitle," Model: ", cfg_name)) +
  xlab("Minimum Prediction Confidence Score Used") +
  ylab("% Accuracy")+
  coord_cartesian(ylim=c(0,1))
  
  print(g3)
  
  g4 <- ggplot(results_df, aes(min_score)) +
  geom_line(aes(y = accuracy , colour = "Accuracy"), linetype=2) +
  geom_line(aes(y = ppv , colour =  "Positive Predictive Value")) +
  geom_line(aes(y = npv , colour = "Negative Predictive Value")) +
  theme_igray() +
  ggtitle(paste0("Predictive Value. ",subtitle," Model: ", cfg_name)) +
  xlab("Minimum Prediction Confidence Score Used") +
  ylab("% Accuracy")+
  coord_cartesian(ylim=c(0,1))

  
  print(g4)
  
}
```

# charcoal_hearth_hill
# Model cfg20200826T2315
## Epochs = 16

```{r}
cfg_name <- 'cfg20200826T2315'
object_type <- 'charcoal_hearth_hill'
graph_result(cfg_name, object_type, "_training")

graph_result(cfg_name, object_type, "")

```