import matplotlib.pyplot as plt import torch from IPython import display from d2l import torch as d2l batch_size = 256 train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size) # Return the iterator of the training set and test set # Will flatten each image, treating them as vectors of length 784. The image is 28*28, stretched to 784 # Because the data set has 10 categories, the grid output dimension is 10 num_inputs = 784 num_output = 10 w = torch.normal(0, 0.01, size=(num_inputs, num_output), requires_grad=True) # Weight size: shape --num_inputs: row, num_output: column b = torch.zeros(num_output, requires_grad=True) # Deviation ''' # Review: Given a matrix, we can sum all elements x = torch.tensor([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]) print(x) print(x.sum(0, keepdim=True), '\\ ', x.sum(1, keepdim=True)) # tensor([[1., 2., 3.], # [4., 5., 6.]]) # tensor([[5., 7., 9.]]) # tensor([[ 6.], # [15.]]) ''' # Implement softmax. Using softmax on a matrix is to perform it on each row of the matrix. """ exponentiate each term (using exp); Sum each row (each sample in the mini-batch is one row) to get the normalization constant for each sample; Divide each row by its normalizing constant, making sure the results sum to 1. """ def softmax(x): x_exp = torch.exp(x) partition = x_exp.sum(1, keepdim=True) return x_exp / partition # The broadcast mechanism is applied # test x = torch.normal(0, 1, (2, 5)) x_prob = softmax(x) # print(x_prob, '\\ ', x_prob.sum(1)) #Test result: Turn each element into a non-negative number. Furthermore, according to the principle of probability, the sum of each row is 1 # tensor([[0.1488, 0.0353, 0.1059, 0.5232, 0.1868], # [0.1903, 0.0968, 0.3103, 0.1855, 0.2172]]) # tensor([1.0000, 1.0000]) # Implement softmax regression model def net(x): return softmax(torch.matmul(x.reshape((-1, w.shape[0])), w) + b) # x.reshape((-1, w.shape[0]): # -1 means automatically calculating the dimension size and adjusting x to the same shape as w # is wx + b # Demonstrate how to take out my corresponding predicted value based on my label in my predicted value. # Create a data y_hat, which contains the predicted probabilities of 2 samples in 3 categories, using y as the index of the probability in y_hat y = torch.tensor([0, 2]) # represents two real labels y_hat = torch.tensor([[0.1, 0.3, 0.6], [0.3, 0.2, 0.5]]) # Predicted value # print(y_hat[[0, 1], y]) # Here [0, 1] represents the 0th and 1st rows of y_hat, y is [0, 2], which represents the 0th and 2nd columns, The combination is (0,0) and (1,2) # Implement the cross entropy loss function -ln(y_hat[range(len(y_hat)), y]) def cross_entropy(y_hat, y): return -torch.log(y_hat[range(len(y_hat)), y]) # print(cross_entropy(y_hat, y)) # tensor([2.3026, 0.6931]) # Compare predicted categories to true y elements def accuracy(y_hat, y): """Calculate the number of correct predictions""" if len(y_hat.shape) > 1 and y_hat.shape[1] > 1: # The number of rows and columns of y_hat is greater than one y_hat = y_hat.argmax(axis=1) # The subscript with the largest element value in each row is stored in y_hat cmp = y_hat.type(y.dtype) == y # Convert y_hat to the data type of y and then compare it to become a bool tensor return float(cmp.type(y.dtype).sum()) # Then convert cmp into the same shape as y and calculate the sum # print(accuracy(y_hat, y) / y) # tensor([ inf, 0.5000]) # We can evaluate the accuracy of any model net def evaluate_accuracy(net, data_iter): """Calculate the accuracy of the model on the specified data set""" if isinstance(net, torch.nn.Module): net.eval() # Set the model to evaluation mode metric = Accumulator(2) # Number of correct predictions, total number of predictions Accumulator object, which is used to accumulate two values: the number of correct predictions and the total number of predictions. for x, y in data_iter: metric.add(accuracy(net(x), y), y.numel()) return metric[0] / metric[1] class Accumulator: # @save """Accumulate on n variables""" def __init__(self, n): self.data = [0.0] * n def add(self, *args): self.data = [a + float(b) for a, b in zip(self.data, args)] def reset(self): self.data = [0.0] * len(self.data) def __getitem__(self, idx): return self.data[idx] def train_epoch_ch3(net, train_iter, loss, updater): # @save """Train the model for one iteration cycle (see Chapter 3 for definition)""" # Set the model to training mode if isinstance(net, torch.nn.Module): net.train() # Sum of training losses, sum of training accuracy, number of samples metric = Accumulator(3) for X, y in train_iter: # Calculate gradient and update parameters y_hat = net(X) l = loss(y_hat, y) if isinstance(updater, torch.optim.Optimizer): # Use PyTorch’s built-in optimizer and loss function updater.zero_grad() l.mean().backward() updater.step() else: # Use custom optimizer and loss function l.sum().backward() updater(X.shape[0]) metric.add(float(l.sum()), accuracy(y_hat, y), y.numel()) # Return training loss and training accuracy return metric[0] / metric[2], metric[1] / metric[2] class Animator: # @save """Draw data in animation""" def __init__(self, xlabel=None, ylabel=None, legend=None, xlim=None, ylim=None, xscale='linear', yscale='linear', fmts=('-', 'm--', 'g-.', 'r:'), nrows=1, ncols=1, figsize=(3.5, 2.5)): # Draw multiple lines incrementally if legend is None: legend = [] d2l.use_svg_display() self.fig, self.axes = d2l.plt.subplots(nrows, ncols, figsize=figsize) if nrows * ncols == 1: self.axes = [self.axes, ] # Use lambda function to capture parameters self.config_axes = lambda: d2l.set_axes( self.axes[0], xlabel, ylabel, xlim, ylim, xscale, yscale, legend) self.X, self.Y, self.fmts = None, None, fmts def add(self, x, y): # Add multiple data points to the chart if not hasattr(y, "__len__"): y = [y] n = len(y) if not hasattr(x, "__len__"): x = [x] * n if not self.X: self.X = [[] for _ in range(n)] if not self.Y: self.Y = [[] for _ in range(n)] for i, (a, b) in enumerate(zip(x, y)): if a is not None and b is not None: self.X[i].append(a) self.Y[i].append(b) self.axes[0].cla() for x, y, fmt in zip(self.X, self.Y, self.fmts): self.axes[0].plot(x, y, fmt) self.config_axes() display.display(self.fig) plt.draw() plt.pause(0.001) display.clear_output(wait=True) def train_ch3(net, train_iter, test_iter, loss, num_epochs, updater): # @save """Training model (see Chapter 3 for definition)""" animator = Animator(xlabel='epoch', xlim=[1, num_epochs], ylim=[0.3, 0.9], legend=['train loss', 'train acc', 'test acc']) for epoch in range(num_epochs): train_metrics = train_epoch_ch3(net, train_iter, loss, updater) test_acc = evaluate_accuracy(net, test_iter) animator.add(epoch + 1, train_metrics + (test_acc,)) train_loss, train_acc = train_metrics assert train_loss < 0.5, train_loss assert 1 >= train_acc > 0.7, train_acc assert 1 >= test_acc > 0.7, test_acc lr = 0.1 def updater(batch_size): return d2l.sgd([w, b], lr, batch_size) if __name__ == '__main__': # print(evaluate_accuracy(net, test_iter)) num_epochs = 10 train_ch3(net, train_iter, test_iter, cross_entropy, num_epochs, updater)