From 77b934db71686f95254d1bf5c6d5d79b79cc1bc2 Mon Sep 17 00:00:00 2001
From: Philip Smith <philsmith990@hotmail.co.uk>
Date: Thu, 21 Mar 2024 13:35:05 +0000
Subject: [PATCH] add svm model

---
 R/fitInteractive.R       |  2 +-
 inst/misc_ml_fitting.Rmd | 68 +++++++++++++++++++++++++++++++++++++++-
 2 files changed, 68 insertions(+), 2 deletions(-)

diff --git a/R/fitInteractive.R b/R/fitInteractive.R
index f6028b8..3d479fb 100644
--- a/R/fitInteractive.R
+++ b/R/fitInteractive.R
@@ -37,7 +37,7 @@ fitInteractive <- function(data=NULL,metadata=NULL,autoSave=FALSE,autoSaveInt=15
 
     if (!requireNamespace("colourpicker", quietly = TRUE)) {
         stop(
-            "Package \"shinyjs\" must be installed to use interactive fitting",
+            "Package \"colourpicker\" must be installed to use interactive fitting",
             call. = FALSE
         )
     }
diff --git a/inst/misc_ml_fitting.Rmd b/inst/misc_ml_fitting.Rmd
index 9a9dcd7..ccddeb8 100644
--- a/inst/misc_ml_fitting.Rmd
+++ b/inst/misc_ml_fitting.Rmd
@@ -37,8 +37,8 @@ cellQcLong <- cellQC %>%
 
 ```{r datapre}
 ## Data preview
-head(cellQC)
 dim(cellQC)
+head(cellQC)
 ```
 
 ```{r plots}
@@ -99,6 +99,65 @@ summary(modelRecipe)
 folds <- vfold_cv(trainingData, v = 10,strata = use)
 ```
 
+## linear SVM
+
+```{r linsvmModel}
+#set linear svm
+linsvm_mod <- svm_rbf(cost = tune(), rbf_sigma = tune()) %>%
+  set_mode("classification") %>%
+  set_engine("kernlab")
+```
+
+```{r linsvmworkflow}
+linsvm_WF <- workflow() %>%
+            add_model(linsvm_mod) %>%
+            add_recipe(modelRecipe)
+```
+
+```{r linsvmtune}
+#lr_reg_grid <- tibble(penalty = 10^seq(-4, -1, length.out = 30))
+
+linsvm_res <- linsvm_WF %>% 
+            tune_grid(resamples = folds,
+                #grid = lr_reg_grid,
+                control = control_grid(save_pred = TRUE),
+                metrics = metric_set(accuracy))
+```
+
+```{r linsvmselectTuning}
+linsvm_best <- linsvm_res %>% 
+    select_best("accuracy",n = 15)
+
+linsvm_WF <- finalize_workflow(linsvm_WF,linsvm_best)
+```
+
+```{r linsvmlastfit}
+linsvm_res_final <- last_fit(linsvm_WF,dataSplit)
+collect_metrics(linsvm_res_final)
+```
+
+```{r linsvmpred}
+linsv_auc <- linsvm_res_final %>%
+  collect_predictions() %>% 
+  roc_curve(use, .pred_FALSE) %>% 
+  mutate(model = "SVM (linear)")
+```
+
+```{r modelPerflinsvm}
+linsvm_res_final %>%
+    collect_metrics()
+
+# linsvm_res_final %>%
+#     extract_fit_parsnip() %>% 
+#     vip(num_features = 20) +
+#     theme_bw()
+
+linsvm_res_final %>% 
+    collect_predictions() %>% 
+    roc_curve(use, .pred_FALSE) %>% 
+    autoplot()
+```
+
 ## logistic regression
 
 ```{r lgModel}
@@ -294,11 +353,18 @@ final_xgb_res %>%
 
 ```{r compare}
 metrics <- metric_set(precision,accuracy,recall,f_meas,roc_auc)
+
 lr_metrics <- lr_res_final %>%
     collect_predictions() %>%
     metrics(truth = use,estimate = .pred_class,.pred_FALSE) %>%
     mutate(model = "logistic regression")
 
+linsvm_metrics <- linsvm_res_final %>%
+    collect_predictions() %>%
+    metrics(truth = use,estimate = .pred_class,.pred_FALSE) %>%
+    mutate(model = "SVM (linear)")
+
+
 rf_metrics <- final_rf_res %>%
     collect_predictions() %>%
     metrics(truth = use,estimate = .pred_class,.pred_FALSE) %>%