[LAB] Week 2_Problem 2_Seohyun Lee

[seo-rii]

Problem

Week 2_Problem 2

Source Code

import torch
from random import random
from typing import Callable

##                         Problem 2                          ##
##                                                            ##
##            Arbitary x_train, y_train are given.            ##
##     In this problem, x_train is list of list of float.     ##
##   Suppose that x and y have linear correlation, y=wx+b.    ##
##           (In this problem, w will be a vector.)           ##
##     In function training(), you should return [w, b].      ##
##          In function predict(), you should return          ##
##            list y_test corresponding to x_test.            ##
##                  Made by @jangyoujin0917                   ##
##                                                            ##

# NOTE : Feel free to use torch.optim and tensor.

def training(x_train: list[list[float]], y_train: list[float]) -> tuple[
    list[float], float]:  # DO NOT MODIFY FUNCTION NAME
    # data normalization
    # 1. Prevents overflow when calculating MSE
    # 2. Prevents underfitting
    # Note that you need to convert [w, b] to the original scale.
    # w = w * (y_max - y_min)
    # b = b * (y_max - y_min) + y_min
    w = torch.tensor([0., 0.], requires_grad=True)
    b = torch.tensor(0., requires_grad=True)
    y_min = min(y_train)
    y_max = max(y_train)
    normalize = lambda y: (y - y_min) / (y_max - y_min)

    ### IMPLEMENT FROM HERE
    epoch = 30000
    lr = 0.007

    x = torch.tensor(x_train, requires_grad=True)
    y = torch.tensor([normalize(y) for y in y_train], requires_grad=True)

    opt = torch.optim.Adam([w, b], lr=lr)

    for _ in range(epoch):
        y_pred = torch.matmul(x, w) + b
        loss = torch.mean((y_pred - y) ** 2)
        loss.backward()
        opt.step()
        opt.zero_grad()

    w.data = w.data * (y_max - y_min)
    b.data = b.data * (y_max - y_min) + y_min
    return w.tolist(), b.item()

def predict(x_train: list[list[float]], y_train: list[float], x_test: list[list[float]]) -> list[
    float]:  # DO NOT MODIFY FUNCTION NAME
    w, b = training(x_train, y_train)
    return [(torch.dot(torch.tensor(x), torch.tensor(w)) + b).item() for x in x_test]

if __name__ == "__main__":
    x_train = [[0., 1.], [1., 0.], [2., 2.], [3., 1.], [4., 3.]]
    y_train = [3., 2., 7., 6., 11.]  # y = x_0 + 2*x_1 + 1 # Note that not all test cases give clear line.
    x_test = [[5., 3.], [10., 6.], [8., 9.]]

    w, b = training(x_train, y_train)
    y_test = predict(x_train, y_train, x_test)

    print(w, b)
    print(y_test)

Description

응슷응

Output (Optional)

No response

[github-actions[bot]]

This is an auto-generated grading output. Your code failed to run. Please check again.

[github-actions[bot]]

This is an auto-generated grading output. Your code failed to run. Please check again.

[jangyoujin0917]

In the problem description, I expressed that w is a vector. It means that we do not know the size of the w.

[Dongyeongkim]

This issue is closed now because of the lack of progress.