Examlpe on how to use SDK for creating classifiers metrics

metrics.cpp

@@ -0,0 +1,139 @@
+/* This is an example how to add additional evaluation metrics to GraphLab Create using the SDK.
+In this example, we developed seven evaluation metrics function for binary classifier. 
+These functions get as input two SArrays with actual and predicted classifier values, and the label of the positive class, 
+and are able to calculate the classifier's: true-positive (TP), False-Positive (FP),True-Negative (TN),false-negative (FN), precision score, recall score, and f1 score.
+
+The file can be complied by using the following command:
+g++ -std=c++11 metrics.cpp -I <graphlab-sdk directory> -shared -fPIC -o metrics.so
+
+After compilation the file can be loaded to Python using the following line:
+>>> import metrics
+
+Python Code Example:
+>>> import graphlab
+>>> import metrics
+
+
+>>> actual = graphlab.SArray([1,1,0,0,0])
+>>> predicted = graphlab.SArray([0,1,0,0,1])
+>>> metrics.true_positive(actual,predicted, 1) #1
+>>> metrics.false_positive(actual,predicted, 1) # 1
+>>> metrics.true_negative(actual,predicted, 1) # 2
+>>> metrics.false_negative(actual,predicted, 1) # 2
+
+>>> actual = graphlab.SArray(['a','b','a','a','a','b'])
+>>> predicted = graphlab.SArray(['a','a','b','b','b','a'])
+>>> metrics.true_positive(actual,predicted, 'a') # 1
+>>> metrics.false_positive(actual,predicted, 'a') # 2
+>>> metrics.true_negative(actual,predicted, 'a') # 0
+>>> metrics.false_negative(actual,predicted, 'a') # 3
+
+
+For more details please read:
+https://dato.com/products/create/sdk/docs/index.html
+https://github.com/dato-code/GraphLab-Create-SDK
+
+
+Written by: Michael Fire
+*/
+
+#include <graphlab/sdk/toolkit_function_macros.hpp>
+#include <graphlab/sdk/gl_sarray.hpp>
+using namespace graphlab;
+int true_positive(gl_sarray actual, gl_sarray predicted, flexible_type pos_label) {		
+	if(actual.size() != predicted.size())
+	{
+		throw "Error: Cannot calculate True-Positive. Input arrays of different size";
+	}	
+	int tp = 0;
+	for(int i=0;i<actual.size(); i++)
+	{
+		if(actual[i] == pos_label && actual[i] == predicted[i])
+		{
+			tp += 1;
+		}
+
+	}
+	return tp;
+}
+
+int false_positive(gl_sarray actual, gl_sarray predicted, flexible_type pos_label) {		
+	if(actual.size() != predicted.size())
+	{
+		throw "Error: Cannot calculate False-Positive. Input arrays of different size";
+	}	
+	int fp = 0;
+	for(int i=0;i<actual.size(); i++)
+	{
+		if(predicted[i] == pos_label && actual[i] != pos_label)
+		{
+			fp += 1;
+		}
+
+	}
+	return fp;
+}
+
+int true_negative(gl_sarray actual, gl_sarray predicted, flexible_type pos_label) {		
+	if(actual.size() != predicted.size())
+	{
+		throw "Error: Cannot calculate True-Negative. Input arrays of different size";
+	}	
+	int tn = 0;
+	for(int i=0;i<actual.size(); i++)
+	{
+		if(predicted[i] != pos_label && actual[i] != pos_label)
+		{
+			tn += 1;
+		}
+
+	}
+	return tn;
+}
+
+int false_negative(gl_sarray actual, gl_sarray predicted, flexible_type pos_label) {		
+	if(actual.size() != predicted.size())
+	{
+		throw "Error: Cannot calculate False-Negative. Input arrays of different size";
+	}	
+	int fn = 0;
+	for(int i=0;i<actual.size(); i++)
+	{
+		if(predicted[i] != pos_label && actual[i] == pos_label)
+		{
+			fn += 1;
+		}
+
+	}
+	return fn;
+}
+
+float precision_score(gl_sarray actual, gl_sarray predicted, flexible_type pos_label) {		
+	float tp = (float) true_positive(actual,predicted,pos_label);
+	float fp = (float) false_positive(actual,predicted,pos_label);
+	return tp/(tp + fp);
+}
+
+float recall_score(gl_sarray actual, gl_sarray predicted, flexible_type pos_label) {		
+	float tp = (float) true_positive(actual,predicted,pos_label);
+	float fn = (float) false_negative(actual,predicted,pos_label);
+	return tp/(tp + fn);
+}
+
+float f1_score(gl_sarray actual, gl_sarray predicted, flexible_type pos_label) {		
+	float recall = recall_score(actual,predicted,pos_label);
+	float precision = precision_score(actual,predicted,pos_label);
+	return (2 * (precision * recall)) / (precision + recall);
+}
+
+
+BEGIN_FUNCTION_REGISTRATION
+REGISTER_FUNCTION(true_positive, "actual", "predicted", "pos_label"); // provide named parameters
+REGISTER_FUNCTION(false_positive, "actual", "predicted", "pos_label"); // provide named parameters
+REGISTER_FUNCTION(true_negative, "actual", "predicted", "pos_label"); // provide named parameters
+REGISTER_FUNCTION(false_negative, "actual", "predicted", "pos_label"); // provide named parameters
+REGISTER_FUNCTION(precision_score, "actual", "predicted", "pos_label"); // provide named parameters
+REGISTER_FUNCTION(recall_score, "actual", "predicted", "pos_label"); // provide named parameters
+REGISTER_FUNCTION(f1_score, "actual", "predicted", "pos_label"); // provide named parameters
+END_FUNCTION_REGISTRATION
+