A_기민준_lab_week3 - Githubissues

lab01.py

import pandas as pd
import torch
import torch.nn.functional as F

filename = "Breast_cancer_train.csv"
df = pd.read_csv(filename)

x_train = torch.tensor(df.loc[:, ['mean_radius', 'mean_texture', 'mean_perimeter', 'mean_area', 'mean_smoothness']].values.tolist(), dtype=torch.float32)
y_train = torch.tensor(df['diagnosis'].values.tolist(), dtype=torch.float32).unsqueeze(1)

def get_prediction(x_predict):
    prediction = torch.sigmoid(x_predict.matmul(W) + b)
    prediction = (prediction >= torch.tensor([0.5])).int()
    return prediction

W = torch.zeros((5,1), requires_grad=True, dtype=torch.float32)
b = torch.zeros(1, requires_grad=True, dtype=torch.float32)

lr = 1e-5

sgd_optim = torch.optim.SGD([W, b], lr=lr)

nb_epochs = 2000
for epoch in range(nb_epochs):
    hypothesis = torch.sigmoid(x_train.matmul(W) + b)
    cost = F.binary_cross_entropy(hypothesis, y_train)

    sgd_optim.zero_grad()
    cost.backward()
    sgd_optim.step()

    if epoch % 100 == 0:
        correct = get_prediction(x_train) == y_train.int()
        acc = correct.sum().item()/len(correct)
        print(f'epoch [{epoch}]: cost: {cost}, acc: {acc:.5f}')

if __name__=="__main__":
    x_test = torch.tensor([[17.99,10.38,122.8,1001,0.1184],[13.54,14.36,87.46,566.3,0.09779]], dtype=torch.float32)
    print(get_prediction(x_test))

...
...
...
epoch [19700]: cost: 0.3231341540813446, acc: 0.85928
epoch [19800]: cost: 0.323068767786026, acc: 0.85928
epoch [19900]: cost: 0.32300370931625366, acc: 0.85928
tensor([[0],
        [1]], dtype=torch.int32)

lab02.py

import pandas as pd
import torch
import torch.nn as nn
import torch.nn.functional as F

filename = "titanic_train.csv"
df = pd.read_csv(filename, index_col=0)
x_train = torch.tensor((df_x := df.drop(columns=['PassengerId', 'Survived'])).values.tolist(), dtype=torch.float32)
y_train = torch.tensor(df['Survived'].values.tolist(), dtype=torch.float32).unsqueeze(1)

def get_prediction(x_predict):
    prediction = model(x_predict)
    prediction = (prediction >= torch.tensor([0.5])).int()
    return prediction

# df_x = df_x.drop(columns=['Emb_1', 'Emb_2', 'Emb_3'])
# x_train = torch.tensor(df_x.values.tolist(), dtype=torch.float32)

class BinaryClassifier(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(14, 1)
        self.sigmoid = nn.Sigmoid()

    def forward(self, x):
        return self.sigmoid(self.linear(x))

lr = 1e-1

model = BinaryClassifier()

adam_optim = torch.optim.Adagrad(model.parameters(), lr=lr)
# sgd_optim = torch.optim.SGD(model.parameters(), lr=lr, momentum=0.9)

nb_epochs = 1000
for epoch in range(nb_epochs):
    hypothesis = model(x_train)
    cost = F.binary_cross_entropy(hypothesis, y_train)

    adam_optim.zero_grad()
    cost.backward()
    adam_optim.step()

    if epoch % 10 == 0:
        correct = get_prediction(x_train) == y_train.int()
        acc = correct.sum().item()/len(correct)
        print(f'epoch [{epoch}]: cost: {cost}, acc: {acc:.5f}')

...
...
...
epoch [970]: cost: 0.4183887541294098, acc: 0.82197
epoch [980]: cost: 0.4183367192745209, acc: 0.82197
epoch [990]: cost: 0.41828653216362, acc: 0.82197

vlab-kaist / NeuralNet101_2022

A_기민준_lab_week3 #22