机器学习实验和结果

nn_ex_1/train_1.py

@@ -0,0 +1,117 @@
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+import torch
+import torch.nn as nn
+import torch.optim as optim
+from sklearn.model_selection import train_test_split
+from torch.autograd import Variable
+
+NUM_EPOCH = 512
+DTYPE = torch.float
+DEVICE = torch.device("cpu")
+DATA_PATH = './wdbc.data'
+
+# 读取数据
+data = pd.read_csv(DATA_PATH)
+
+data = data.iloc[:, 1:32].values
+
+# 数据预处理
+# 将标记转换为1和0，并转换数据类型为fp32
+data[data == 'M'] = 1
+data[data == 'B'] = 0
+data = data.astype(np.float32)
+# 划分训练集
+x_train, x_test, y_train, y_test = train_test_split(data[:, 1:31],
+                                                    data[:, 0],
+                                                    test_size=0.3)
+# 数据类型和维度转换
+x_train = Variable(torch.Tensor(x_train))
+x_test = Variable(torch.Tensor(x_test))
+y_train = Variable(torch.Tensor(y_train))
+y_test = Variable(torch.Tensor(y_test))
+
+y_train = y_train.reshape([-1, 1])
+y_test = y_test.reshape([-1, 1])
+
+
+# 定义torch模型
+class torch_model(nn.Module):
+    def __init__(self, input, n1, n2, output):
+        super(torch_model, self).__init__()
+        self.layer1 = nn.Sequential(nn.Linear(input, n1), nn.Sigmoid())
+        self.layer2 = nn.Sequential(nn.Linear(n1, n2), nn.Sigmoid())
+        self.layer3 = nn.Sequential(nn.Linear(n2, output), nn.Sigmoid())
+
+    def forward(self, x):
+        x = self.layer1(x)
+        x = self.layer2(x)
+        x = self.layer3(x)
+        return x
+
+# 准确率
+def Acc(y_pred, y_real):
+
+    y_pred = np.array(y_pred.data)
+    y_real = np.array(y_real)
+    pred_mean = np.mean(y_pred)
+    y_pred[y_pred > pred_mean] = 1.0
+    y_pred[y_pred <= pred_mean] = 0.0
+    error = np.sum(np.abs(y_pred - y_real))
+    return 1 - (error / y_pred.shape[0])
+
+
+# 模型，优化器，损失函数
+model = torch_model(30, 90, 45, 1)
+optimizer = optim.Adam(model.parameters(), lr=0.001)
+crierion = nn.MSELoss()
+
+E_list = []
+L_list = []
+A_list = []
+# 训练并保存数据
+for epoch in range(NUM_EPOCH):
+    y_pred = model(x_train)
+    loss = crierion(y_pred, y_train)
+    accuracy = Acc(y_pred, y_train)
+    print("epoch:%d loss:%f accuracy:%f" % (epoch, loss.data.item(), accuracy))
+    optimizer.zero_grad()
+    loss.backward()
+    optimizer.step()
+    E_list.append(epoch)
+    L_list.append(loss.data.item())
+    A_list.append(accuracy)
+# 预测数据并评价结果
+model.eval()
+eval_loss = 0.0
+eval_acc = 0.0
+
+y_pred = model(x_test)
+loss = crierion(y_pred, y_test)
+eval_loss = loss.data.item()
+
+y_test = np.array(y_test)
+y_pred = np.array(y_pred.data)
+pred_mean = np.mean(y_pred)
+y_pred[y_pred > pred_mean] = 1
+y_pred[y_pred <= pred_mean] = 0
+
+for i in range(y_pred.shape[0]):
+    y_p = y_pred[i]
+    y_t = y_test[i]
+    if y_p == y_t:
+        eval_acc += 1
+eval_acc_rare = eval_acc / y_pred.shape[0]
+print('accuracy %d in %d' % (int(eval_acc), y_pred.shape[0]))
+print('accuracy rate is %f' % (eval_acc / y_pred.shape[0]))
+
+# 训练中损失函数的变化
+
+E = np.array(E_list)
+L = np.array(L_list)
+A = np.array(A_list)
+plt.figure()
+plt.plot(E, L)
+plt.plot(E, A)
+plt.show()