-

Sanaxen · Nov 16, 2022 · 7aef019 · 7aef019
1 parent 5a5e171
commit 7aef019
Show file tree

Hide file tree

Showing 5 changed files with 40 additions and 7 deletions.
diff --git a/README.md b/README.md
@@ -8,8 +8,11 @@ Through the C# UI, you can easily perform time series forecasting with the
 At the core of this application, tft is ![](./images/554068490.png) https://github.com/mlverse.  
 It relies heavily on https://github.com/mlverse/tft, a wonderful library implemented in R developed at  
 
-<img src="./images/image03.gif" width=70%>  
-
+<img src="./images/image03.gif" width=80%>  
+---
+### Feature importance Plot  
+The permutation feature importance algorithm based on Fisher, Rudin, and Dominici (2018)  
+<img src="./images/image01.png" width=80%>  
 # Requirements  
 
 [webview2](https://developer.microsoft.com/ja-jp/microsoft-edge/webview2/)  

diff --git a/images/image01.png b/images/image01.png
diff --git a/script/tft_util.r b/script/tft_util.r
@@ -879,7 +879,7 @@ permutationFeatureImportance<- function(fitted, test, validation=F, base_name=""
 
 	FI = NULL
 	FI_s = NULL
-	sampling_n = 10
+	sampling_n = 1
 
 	for ( k in 1:sampling_n )
 	{
@@ -958,7 +958,30 @@ permutationFeatureImportance<- function(fitted, test, validation=F, base_name=""
 	g2 <- ggplot(data = x, aes(x = date, y = key , fill = importance)) + 
 	geom_tile()+
 	scale_fill_gradient2(low = "springgreen4", mid = "yellow", high = "red", midpoint = 0.5)
-	ggsave(file = paste(base_name,"_feature_importance_time.png", sep=""), plot = g2, dpi = 100, width = 6.4, height = 4.8*length(name)/10)
+	ggsave(file = paste(base_name,"_feature_importance_time1.png", sep=""), plot = g2, dpi = 100, width = 6.4, height = 4.8*length(name)/10)
+
+	FI_s$date <- NULL
+	FI_s <- (FI_s - min(FI_s))/(max(FI_s) - min(FI_s))
+ 	FI_s$date <- test_tmp$date
+ 	x<-horizontally_to_vertically(FI_s, ids_cols=c('date'), key=name)
+ 	x$importance <- x$target
+ 	x$target <- NULL
+
+	g3 <- x %>% 
+	  ggplot(aes(x = date, y = importance, color=key))+
+	  geom_line()+
+	  scale_x_datetime(breaks = date_breaks(unit), labels = date_format("%Y-%m-%d %H")) +
+	  theme(axis.text.x = element_text(angle = 45, hjust = 1))
+
+	g4 <- ggplot(x, aes(x = date, y = importance, fill = key))
+	g4 <- g4 + geom_bar(stat = "identity", position = "fill")
+	g4 <- g4 + scale_y_continuous(labels = percent)
+	plot(g4)	  
+
+	g5 <- ggplot(x, aes(x = date, y = importance, fill = key))
+	g5 <- g5 + geom_bar(stat = "identity")
+	plot(g5)	
+	ggsave(file = paste(base_name,"_feature_importance_time.png", sep=""), plot = g4, dpi = 100, width = 6.4, height = 4.8*length(name)/10)
 
  	if ( FALSE )
  	{
@@ -975,7 +998,7 @@ permutationFeatureImportance<- function(fitted, test, validation=F, base_name=""
 		heatmap(as.matrix(FI_s2),Colv = NA, Rowv=NA, scale='col',col=c(rgb(seq(0.9,0.2,-0.001),0, seq(0.0,0.3,0.001))))
 		heatmap(as.matrix(FI_s2),Colv = NA, Rowv=NA, scale='col',col=c(rgb(seq(0.9,0.2,-0.001),0, seq(0.0,0.2,0.001))))
 	}
-	return( list(n, g1, g2))
+	return( list(n, g1, g2, g3, g4, g5))
 }
 
 

diff --git a/tft/Form1.Designer.cs b/tft/Form1.Designer.cs
diff --git a/tft/Form1.cs b/tft/Form1.cs
@@ -173,6 +173,7 @@ void Plot()
                 try
                 {
                     pictureBox6.Image = CreateImage("tft_predict_measure_" + base_name + ".png");
+                    pictureBox6.Refresh();
                 }
                 catch { }
             }
@@ -181,6 +182,7 @@ void Plot()
                 try
                 {
                     pictureBox7.Image = CreateImage( base_name + "_feature_importance.png");
+                    pictureBox7.Refresh();
                 }
                 catch { }
             }
@@ -189,6 +191,7 @@ void Plot()
                 try
                 {
                     pictureBox8.Image = CreateImage(base_name + "_feature_importance_time.png");
+                    pictureBox8.Refresh();
                 }
                 catch { }
             }
@@ -354,6 +357,7 @@ void Proc_Exited(object sender, EventArgs e)
                     {
                         pictureBox6.Image = CreateImage("tft_predict_measure_" + base_name + ".png");
                         // pictureBox6.SizeMode = PictureBoxSizeMode.StretchImage;
+                        pictureBox6.Refresh();
                     }
                     catch { }
                 }
@@ -363,6 +367,7 @@ void Proc_Exited(object sender, EventArgs e)
                     try
                     {
                         pictureBox7.Image = CreateImage(base_name + "_feature_importance.png");
+                        pictureBox7.Refresh();
                     }
                     catch { }
                 }
@@ -372,6 +377,7 @@ void Proc_Exited(object sender, EventArgs e)
                     try
                     {
                         pictureBox8.Image = CreateImage(base_name + "_feature_importance_time.png");
+                        pictureBox8.Refresh();
                     }
                     catch { }
                 }
@@ -1619,10 +1625,10 @@ string tft_test()
             cmd += "\r\n";
             cmd += "\r\n";
             cmd += "fi <- permutationFeatureImportance(fitted, test, validation=validation, base_name ='" + base_name + "')\r\n";
-            cmd += "fi_plot <- gridExtra::grid.arrange(fi[[2]], fi[[3]], ncol = 1)\r\n";
+            cmd += "fi_plot <- gridExtra::grid.arrange(fi[[2]], fi[[5]], ncol = 1)\r\n";
             cmd += "ggsave(file = \"tft_" + base_name + "_fi.png\", plot = fi_plot,dpi=100, width= 1.5*6.4,height=0.09*4.8" + "*fi[[1]], limitsize = FALSE)\r\n";
             cmd += "fi_plot1 <- ggplotly(fi[[2]])\r\n";
-            cmd += "fi_plot2 <- ggplotly(fi[[3]])\r\n";
+            cmd += "fi_plot2 <- ggplotly(fi[[5]])\r\n";
             cmd += "fi_plotly <- subplot(fi_plot1, fi_plot2, nrows = 2)\r\n";
             cmd += "print(fi_plotly)\r\n";
             cmd += "htmlwidgets::saveWidget(as_widget(fi_plotly), \"tft_" + base_name + "_fi.html\", selfcontained = F)\r\n";