Merge pull request #9 from Machine-Learning-Foundations/src_code

Added Todos with numerations to the first part

Author: tenosel

src/pegel_bonn.py

@@ -20,7 +20,10 @@
         day, month, year = ts_date.split(".")
         hour, minute = ts_time.split(":")
         datetime_list.append(datetime(int(year), int(month), int(day)))
+        if hour == " 04" or hour == " 05":
+            keep_indices.append(idx)
 
+    # 2.3.0
     # uncomment to start in the year 2000.
     # levels = levels[:-30_000:4]
     # datetime_list = datetime_list[:-30_000:4]
@@ -37,3 +40,42 @@
     plt.ylabel("Water level [cm]")
     plt.title("Rhine water level in Bonn")
     plt.show()
+
+    # 2.1 Regression
+
+    # 2.1.1 Set up the point matrix for n=2
+    #       The corresponding function that you have implemented in Part 1 has already been imported and is ready to use!
+
+    # 2.1.2 Estimate the coefficients
+
+    # 2.1.3 Evaluate the straight line and plot your result
+
+    # 2.1.4 Calculate the data where the Rhine will be dried out
+
+
+
+    # 2.2 Fitting a higher-order Polynomial
+
+    # 2.2.1 Take a look at the datetime_stamps and scale the axis
+
+    # 2.2.2 Set up the point matrix with n=20
+
+    # 2.2.3 Compute the coefficients for the polynomial
+
+    # 2.2.4 Evaluate the polyomial
+
+    # 2.2.5 Plot the results
+
+
+
+    # 2.3 Regularization
+
+    # 2.3.0 Uncomment the lines of code above
+
+    # 2.3.1 Compute the SVD of the point-matrix for n=20
+
+    # 2.3.2 Compute the filter matrix
+
+    # 2.3.3 Estimate the coefficients by applying regularization
+
+    # 2.3.4 Plot the evaluated, regularized polynomial

src/regularization.py

@@ -24,17 +24,69 @@ def set_up_point_matrix(axis_x: np.ndarray, degree: int) -> np.ndarray:
         np.ndarray: The polynomial point matrix A.
     """
     mat_a = np.zeros((len(axis_x), degree))
-    # TODO implement me!
+    # TODO 1.1.1 implement me!
     return mat_a
 
 
 if __name__ == "__main__":
     b_noise_p = pandas.read_csv("./data/noisy_signal.tab", header=None)
-    b_noise = np.asarray(b_noise_p)
+    b_noise = np.asarray(b_noise_p) # This is the artificial data we work with
 
     x_axis = np.linspace(0, 1, num=len(b_noise))
 
     plt.plot(x_axis, b_noise, ".", label="b_noise")
     plt.show()
 
-    # TODO put your code here!
+    # TODO put your code for Part 1 here!
+
+    # 1.1 Regression
+
+    # 1.1.2 Create the point-matrix A for n=2
+
+    # 1.1.3 Calculate the estimated coefficients for the polynomial
+    #       You can use np.linalg.pinv to compute the Pseudo-Inverse
+
+    # 1.1.4 Plot the original data as well as the estimated polynomial by evaluating it.
+
+
+
+    # 1.2 Higher order Polynomial
+
+    # 1.2.1 Create the point-matrix A for n=300
+
+    # 1.2.2 Calculate the estimated coefficients for the polynomial
+
+    # 1.2.3 Plot the original data as well as the estimated polynomial by evaluating it.
+
+
+
+    # 1.3 Regularization
+
+    # 1.3.1 Compute the SVD of A
+
+    # 1.3.2 Compute the filter matrix
+
+    # 1.3.3 Estimate the coefficients by applying regularization
+
+    # 1.3.4 Plot your results
+
+    #----------------------------------------------------------------------------------------#
+    # Optional Task 1.4 Model Complexity
+
+    # For every degree from 1 to 20:
+
+    # 1.4.1 Set up the point matrix for the current degree
+
+    # 1.4.2 Estimate the coefficients for the polynomial via the pseudoinverse
+
+    # 1.4.3 Compute the predictions by evaluating the estimated polynomial at the x-values
+
+    # 1.4.4 Compute the MSE between the predictions and the original b-values
+
+    # 1.4.5 Plot the MSE-error against the degree  
+
+    # 1.4.6 Take a look at the graph with all 20 MSE's and see if there is a link between degree and MSE
+    #       Estimate the optimal degree of polynomial and fit the polynomial with this new degree
+    #             --> so perform the usual steps but only once with the optimal degree 
+    
+    #       Plot the result