# this is a recap

# a training pipeline generally consists of 3 steps:
# 1. Design model (input, output size, forward pass (layers))
# 2. Construct loss and optimizer
# 3. Training loop
#     - forward pass: compute prediction
#     - backward pass: gradient computation
#     - update parameters

# (iterate step 3)

import torch
import torch.nn as nn
import numpy as np
from sklearn import datasets
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
# import matplotlib.pyplot as plt

# 0. data preparation
bc = datasets.load_breast_cancer()
X, y = bc.data, bc.target

n_samples, n_features = X.shape # 569 30

# split data into training and test data
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=1234)

# scale
sc = StandardScaler()
X_train = sc.fit_transform(X_train)
X_test = sc.transform(X_test)

X_train = torch.from_numpy(X_train.astype(np.float32))
X_test = torch.from_numpy(X_test.astype(np.float32))
y_train = torch.from_numpy(y_train.astype(np.float32))
y_test = torch.from_numpy(y_test.astype(np.float32))

# reshape y
y_train = y_train.view(-1, 1) # y_train.view(y_train.shape[0], 1) 
y_test = y_test.view(-1, 1)

X_train = X_train.to('cuda')
X_test = X_test.to('cuda')
y_train = y_train.to('cuda')
y_test = y_test.to('cuda')


# 1. model
# f = wx + b, sigmoid at the end
class LogReg(nn.Module):
    def __init__(self, n_input_features):
        super(LogReg, self).__init__()
        self.linear = nn.Linear(n_input_features, 1)

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

model = LogReg(n_features)
model = model.to('cuda')

# 2. loss + optimizer
learning_rate = .0001
criterion = nn.BCELoss() # binary cross entropy                             #accuracy at lr = .0001, n_epochs = 100000
# optimizer = torch.optim.SGD(model.parameters(), lr = learning_rate)       # 0.93
# optimizer = torch.optim.Adam(model.parameters(), lr = learning_rate)      # 0.96
# optimizer = torch.optim.AdamW(model.parameters(), lr = learning_rate)     # 0.96
optimizer = torch.optim.SGD(model.parameters(), lr = learning_rate, momentum=0.9)     # 0.95

# 3. training
n_epochs = 100000

for epoch in range(n_epochs):
    # forward pass and loss
    y_pred = model(X_train)
    loss = criterion(y_pred, y_train)

    # backward pass
    loss.backward()

    # updates and zero gradients
    optimizer.step()
    optimizer.zero_grad()

    if (epoch+1) % (10000) == 0:
        print(f'{(epoch+1)/n_epochs*100:.1f}% Epoch {epoch+1:>8}: loss = {loss.item():.5f}')

# evaluation
with torch.no_grad(): # don't track these operations for the gradient computation
    y_pred = model(X_test)
    y_pred_cls = y_pred.round()
    acc = y_pred_cls.eq(y_test).sum() / float(y_test.shape[0])
    print(f'Accuracy = {acc:.4f}')