RegressionModel.cpp

regression model i've made in c++ :)
just input your desired coordinates and it'll send out the slope using gradient descent

Author: AbePineda

RegressionModel.cpp

@@ -0,0 +1,102 @@
+#include <iostream>
+#include <vector>
+#include <string>
+#include <regex>
+#include <sstream>
+#include <cstdlib>
+using namespace std;
+
+class input_handler {
+    public:
+    vector<double> x_data;
+    vector<double> y_data;
+    string userInput;
+    input_handler() {
+        getline(cin, userInput);
+        input_pusher(userInput);
+    }
+
+    ~input_handler() {
+        x_data.clear();
+        y_data.clear();
+    }
+
+    void input_pusher (const string& userInput) {
+        //regex digitGrabber("\\d+\\.?\\d*");
+        regex puncRemover("[^0-9.-]");
+        string cleaned = regex_replace(userInput, puncRemover, " ");
+        istringstream stream(cleaned); 
+        string tokenized_input;
+        int interval = 0;
+        while(stream >> tokenized_input) {
+                interval++;
+                if(interval % 2 != 0)
+                    x_data.push_back(stod(tokenized_input));
+                else
+                    y_data.push_back(stod(tokenized_input));
+        }
+        if(y_data.size() != x_data.size()) {
+            cout << "Error: unbalanced values" << endl;
+            exit(1);
+        }
+        
+    }
+};
+
+class regressionModel : public input_handler{
+    public:
+    double m = 0;
+    double b = 0;
+    double learning_rate = 0.01; //learning rate is like the how much the slope is rotating towards its desired location
+    int iterations = 1000; //controls how many times the learning gets iterated
+
+    ~regressionModel() {
+        cout << "Predicted Slope: y = " << m << "x + " << b <<  endl;
+        m = 0;
+        b = 0;
+    }
+
+    regressionModel() {
+        train();
+    }
+
+    void train() {
+        int n  = x_data.size(); //amount of coordinates in order to get averages
+        double prev_difference = 0.0; //holds the previous difference
+        double curr_difference = 0.0; //holds the current difference
+        for(int i = 0; i < iterations; i++) {
+            double compilation_m = 0; // set these to zero to create new instances each iteration
+            double compilation_b = 0;
+            vector<double> y_prediction = predict(m, b); //execute 'y = mx + b' based off of compilation values
+            curr_difference = 0; //creates new instances of the current difference
+            for(int j = 0; j < n; j++) {
+                //gradient descent
+                compilation_m += -2 * x_data[j] * (y_data[j] - y_prediction[j]); //m has both x/y-components
+                compilation_b += -2 * (y_data[j] - y_prediction[j]); //b only has a y-component
+
+                curr_difference += (y_data[j] - (m * x_data[j] + b)) * (y_data[j] - (m * x_data[j] + b)); //square error method - takes the origianal y-value to the predicted y-vale and finds the difference
+                //we multiple it to avoid negatives and to make it more zoomed in and be more accurate 
+            }
+            curr_difference /= n; //gets the average
+
+            if (abs(curr_difference - prev_difference) < 1e-6) break; //if the difference is small between the last iteration, break out of the loop
+
+            prev_difference = curr_difference; //do recursion for the previous difference
+        
+            m -= (learning_rate * compilation_m) / n; //gets the average with respect of the learning rate (this gets repeatadly updated till we decide to print it)
+            b -= (learning_rate * compilation_b) / n;
+
+        }
+    }
+    vector<double> predict(double m, double b) {
+        vector<double> y_prediction;
+        for(double X: x_data)
+            y_prediction.push_back(m * X + b); 
+        return y_prediction;
+    }
+};
+
+int main () {
+    regressionModel model;
+    return 0;
+}