Assignment 1
By: André Franzén (af223kr)
Course: 1DV516
Date: March-April 2022
File name microships.py
In order to run task 1.2 Where the predefined points are calculated run the perdictForPredefPoints() (In order to avoid messy output)
To run Task 1.3 and display the plots just run the python file
I have gone through many many iterations to try to make it quicker but and have to great lengths succeeded. I have cut down the time it takes from my original working solution by more than 75% But it still takes some time to run and I Think its due to how i calculate the distances
for x,y,ok in points:
#d=√((x_2-x_1)²+(y_2-y_1)²)
dis = sqrt( pow((z[0] - x),2) + pow((z[1] - y),2) )
distanceList.append( ((x, y, ok),dis) )I want to great lengths to try to understand and how to use distance matrices and there for be able to use broadcast but i did not succeed.
Time to complete (on my pc) 6.5s to compute above image
File name polynomial.py
Run the python file, (2 plots will open 1 corresponding to task 1.2) and the other with the MSE errors and regression lines
For this task I am happy about my implementation there are probably better ways as i use a fair few loops but all in all its pretty quick.
Note! the MSE should in theory be 0 for the one with K = 1 but due to the dots being to close together even a very very small stepsize is needed.
2.5: Which K gives the best regression Motivate your answer! I Think that the K = 9 is the best K this is due to it having the lowest MSE test error. training error does not really give alot of insight this is why i make my pick based on MSE test value
Time to complete (on my pc) 6s to compute above image
File name MNIST_V2.py & MNIST.py
Run the python file MNIST_V2.py this will do two things
- Plot 6 numbers from the dataset and the predicted values for them.
- Predict 100 random numbers and print the result
To complete this exercise it took me two attempts, I first started off by learning and trying PCA this is a way to reduce the amount of dimensions (in my case down to two) so that i could plot the input just like in exercise 1. This later turned out to be the wrong approach and lead to the first program only predicting about 45% correct.
My second attempt was much more successfull and much less complicated. I reuse much of what i have learnt from exercise 1 combined with my previous attempt.
To find the best value for K i tried many different ranging from 1 to 11 and found 3 to be the best. (had the best accuracy for the test values)
Here is the 1st attempt where i tried to use PCA to lower the dimensions of the input to only two. This successfully guessed the correct number about 45% of the time due to the immense data lost with PCA
Here is the 2nd attempt this is much more successfull.
This guesses the correct number around 95% of the time.
File name sckitMicroShips.py
Run the python file sckitMicroShips.py
I am happy with my implementation was fun to see how much easier it was with the library
Time to complete (on my pc) 3s to compute above image