133 lines
5.4 KiB
Python
133 lines
5.4 KiB
Python
import torch
|
|
import torch.nn as nn
|
|
import torch.optim as optim
|
|
from torch.utils.data import DataLoader, TensorDataset
|
|
import numpy as np
|
|
import pandas as pd
|
|
from sklearn.preprocessing import LabelBinarizer
|
|
import numpy as np
|
|
from sklearn.preprocessing import MinMaxScaler
|
|
from sklearn.model_selection import train_test_split
|
|
|
|
|
|
|
|
class MyNeuralNetwork(nn.Module):
|
|
def __init__(self, *args, **kwargs) -> None:
|
|
super(MyNeuralNetwork, self).__init__(*args, **kwargs)
|
|
self.fc1 = nn.Linear(12, 64)
|
|
self.relu = nn.ReLU()
|
|
self.fc1 = nn.Linear(12, 64)
|
|
self.relu = nn.ReLU()
|
|
self.fc2 = nn.Linear(64, 1)
|
|
self.sigmoid = nn.Sigmoid()
|
|
|
|
def forward(self, x):
|
|
x = self.fc1(x)
|
|
x = self.relu(x)
|
|
x = self.fc2(x)
|
|
x = self.sigmoid(x)
|
|
return x
|
|
|
|
def prepare_df_for_nn(df: pd.DataFrame):
|
|
|
|
id_column_name_list: list[str] = [column for column in df.columns.to_list() if 'id' in column.lower()]
|
|
if len(id_column_name_list) == 0:
|
|
pass
|
|
else:
|
|
df.drop(id_column_name_list[0], inplace=True, axis=1)
|
|
encoder: LabelBinarizer = LabelBinarizer()
|
|
df.reset_index(inplace=True)
|
|
for column in df.columns:
|
|
if str(df[column].dtype).lower() == 'object':
|
|
encoded_column: np.ndarray = encoder.fit_transform(df[column])
|
|
df[column] = pd.Series(encoded_column.flatten(), dtype=pd.Int16Dtype)
|
|
return df
|
|
|
|
def load_data(path: str):
|
|
df: pd.DataFrame = pd.read_csv('home_loan_train.csv')
|
|
train_dataset: pd.DataFrame = prepare_df_for_nn(df)
|
|
x: np.ndarray = train_dataset.iloc[:, :-1].values.astype(float)
|
|
y: np.ndarray = train_dataset.iloc[:, -1].values.astype(float)
|
|
x_tensor: torch.Tensor = torch.tensor(x, dtype=torch.float32)
|
|
y_tensor: torch.Tensor = torch.tensor(y, dtype=torch.float32)
|
|
dataset: TensorDataset = TensorDataset(x_tensor, y_tensor)
|
|
return dataset
|
|
|
|
def train(epochs: int, dataloader_train: DataLoader, dataloader_val: DataLoader):
|
|
model: MyNeuralNetwork = MyNeuralNetwork()
|
|
criterion: nn.BCELoss = nn.BCELoss()
|
|
optimizer = optim.Adam(model.parameters(), lr=0.001)
|
|
for epoch in range(epochs):
|
|
total_correct_train = 0
|
|
total_samples_train = 0
|
|
total_correct_val = 0
|
|
total_samples_val = 0
|
|
|
|
for inputs, labels in dataloader_train:
|
|
outputs = model(inputs)
|
|
labels = labels.reshape((labels.shape[0], 1))
|
|
loss = criterion(outputs, labels)
|
|
predicted_labels = (outputs > 0.5).float()
|
|
total_correct_train += (predicted_labels == labels).sum().item()
|
|
total_samples_train += labels.size(0)
|
|
|
|
optimizer.zero_grad()
|
|
loss.backward()
|
|
optimizer.step()
|
|
|
|
with torch.no_grad():
|
|
for inputs, labels in dataloader_val:
|
|
outputs_val = model(inputs)
|
|
predicted_labels_val = (outputs_val > 0.5).float()
|
|
labels = labels.reshape((labels.shape[0], 1))
|
|
total_correct_val += (predicted_labels_val == labels).sum().item()
|
|
total_samples_val += labels.size(0)
|
|
|
|
accuracy_val = total_correct_val / total_samples_val
|
|
accuracy_train = total_correct_train / total_samples_train
|
|
print(f"Epoch {epoch+1}/{epochs}, Loss: {loss.item():.4f}, Accuracy train: {accuracy_train:.4f}, Accuracy val: {accuracy_val:.4f}")
|
|
|
|
return model
|
|
|
|
def create_dataset():
|
|
home_loan_train = pd.read_csv('/Users/wojciechbatruszewicz/InformatykaStudia/SEMESTR8/IUM/ZADANIA/createDataset/loan_sanction_train.csv')
|
|
home_loan_test = pd.read_csv('/Users/wojciechbatruszewicz/InformatykaStudia/SEMESTR8/IUM/ZADANIA/createDataset/loan_sanction_test.csv')
|
|
|
|
home_loan_train_final, home_loan_test = train_test_split(home_loan_train, test_size=0.2, random_state=1)
|
|
home_loan_test_final, home_loan_val_final = train_test_split(home_loan_test, test_size=0.5, random_state=1)
|
|
|
|
numeric_cols_train = home_loan_train_final.select_dtypes(include='number').columns
|
|
numeric_cols_test = home_loan_test_final.select_dtypes(include='number').columns
|
|
numeric_cols_val = home_loan_val_final.select_dtypes(include='number').columns
|
|
|
|
scaler = MinMaxScaler()
|
|
|
|
home_loan_train_final[numeric_cols_train] = scaler.fit_transform(home_loan_train_final[numeric_cols_train])
|
|
home_loan_test_final[numeric_cols_test] = scaler.fit_transform(home_loan_test_final[numeric_cols_test])
|
|
home_loan_val_final[numeric_cols_val] = scaler.fit_transform(home_loan_val_final[numeric_cols_val])
|
|
|
|
home_loan_train_final = home_loan_train_final.dropna()
|
|
home_loan_test_final = home_loan_test_final.dropna()
|
|
home_loan_val_final = home_loan_val_final.dropna()
|
|
|
|
home_loan_train_final.to_csv('home_loan_train.csv', index=False)
|
|
home_loan_test_final.to_csv('home_loan_test.csv', index=False)
|
|
home_loan_val_final.to_csv('home_loan_val.csv', index=False)
|
|
|
|
def main() -> None:
|
|
# create_dataset()
|
|
train_dataset = load_data("home_loan_train.csv")
|
|
val_dataset = load_data("home_loan_val.csv")
|
|
|
|
batch_size: int = 32
|
|
dataloader_train = DataLoader(train_dataset, batch_size = batch_size, shuffle=True)
|
|
dataloader_val = DataLoader(val_dataset, batch_size = batch_size)
|
|
|
|
model = train(20, dataloader_train, dataloader_val)
|
|
torch.save(model.state_dict(), 'model.pt')
|
|
|
|
if __name__ == "__main__":
|
|
main()
|
|
|
|
|
|
|