先是data_processing文件import pandas as pd from urllib.request import urlretrieve def load_data(downloadTrue): # download data from : http://archive.ics.uci.edu/ml/datasets/CarEvaluation if download: data_path, _ urlretrieve(http://archive.ics.uci.edu/ml/machine-learning-databases/car/car.data, car.csv) print(Downloaded to car.csv) # use pandas to view the data structure col_names [buying, maint, doors, persons, lug_boot, safety, class] data pd.read_csv(car.csv, namescol_names) return data def convert2onehot(data): # covert data to onehot representation return pd.get_dummies(data, prefixdata.columns) if __name__ __main__: data load_data(downloadTrue) new_data convert2onehot(data) print(data.head()) print(\nNum of data: , len(data), \n) # 1728 # view data values for name in data.keys(): print(name, pd.unique(data[name])) print(\n, new_data.head(2)) new_data.to_csv(car_onehot.csv, indexFalse)1.pd.read_csv功能给数据添加一行列索引也就是横在数据上面的表头返回一个诡异的数据类型data: class ‘pandas.core.frame.DataFrame’col_names [buying, maint, doors, persons, lug_boot, safety, class] data pd.read_csv(car.csv, namescol_names)2.get_dummies(data, prefixdata.columns)pandas的功能真的很强大可以直接以data.columns为标准生成独热码形式https://blog.csdn.net/qq_35290785/article/details/91415240return pd.get_dummies(data, prefixdata.columns)3.data.head()读取前五行数据https://blog.csdn.net/qq_18649781/article/details/89033749print(data.head())4.pd.unique用法自己琢磨把…for name in data.keys(): print(name, pd.unique(data[name]))5.to_csvdt.to_csv() #默认dt是DataFrame的一个实例indexFalse表示不保留行索引也就是不保留最左侧的竖列索引new_data.to_csv(car_onehot.csv, indexFalse)综上所述data_processing代码完成了one-hot的csv文件的建立接下来 我们来看model文件import numpy as np import tensorflow as tf import matplotlib.pyplot as plt import data_processing data data_processing.load_data(downloadTrue) new_data data_processing.convert2onehot(data) # prepare training data new_data new_data.values.astype(np.float32) # change to numpy array and float32 np.random.shuffle(new_data) sep int(0.7*len(new_data)) train_data new_data[:sep] # training data (70%) test_data new_data[sep:] # test data (30%) # build network tf_input tf.placeholder(tf.float32, [None, 25], input) tfx tf_input[:, :21] tfy tf_input[:, 21:] l1 tf.layers.dense(tfx, 128, tf.nn.relu, namel1) l2 tf.layers.dense(l1, 128, tf.nn.relu, namel2) out tf.layers.dense(l2, 4, namel3) prediction tf.nn.softmax(out, namepred) loss tf.losses.softmax_cross_entropy(onehot_labelstfy, logitsout) accuracy tf.metrics.accuracy( # return (acc, update_op), and create 2 local variables labelstf.argmax(tfy, axis1), predictionstf.argmax(out, axis1))[1] opt tf.train.GradientDescentOptimizer(learning_rate0.1) train_op opt.minimize(loss) sess tf.Session() sess.run(tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())) # training plt.ion() fig, (ax1, ax2) plt.subplots(1, 2, figsize(8, 4)) accuracies, steps [], [] for t in range(4000): # training batch_index np.random.randint(len(train_data), size32) sess.run(train_op, {tf_input: train_data[batch_index]}) if t % 50 0: # testing acc_, pred_, loss_ sess.run([accuracy, prediction, loss], {tf_input: test_data}) accuracies.append(acc_) steps.append(t) print(Step: %i % t,| Accurate: %.2f % acc_,| Loss: %.2f % loss_,) # visualize testing ax1.cla() for c in range(4): bp ax1.bar(c0.1, heightsum((np.argmax(pred_, axis1) c)), width0.2, colorred) bt ax1.bar(c-0.1, heightsum((np.argmax(test_data[:, 21:], axis1) c)), width0.2, colorblue) ax1.set_xticks(range(4), [accepted, good, unaccepted, very good]) ax1.legend(handles[bp, bt], labels[prediction, target]) ax1.set_ylim((0, 400)) ax2.cla() ax2.plot(steps, accuracies, labelaccuracy) ax2.set_ylim(ymax1) ax2.set_ylabel(accuracy) plt.pause(0.01) plt.ioff() plt.show()1.tf.nn.softmax将所选择的一维数组归一化也就是把每一个元素都变成0-1之间的数字2.tf.metrics.accuracytf.metrics.accuracy返回两个值accuracy为到上一个batch为止的准确度update_op为更新本批次后的准确度。https://blog.csdn.net/lyb3b3b/article/details/83047148因为有局部变量所以要初始化局部变量sess.run(tf.local_variables_initializer())这就是为什么最后有初始化局部变量的原因3.plt.ion()打开交互模式若想动态显示图像则需要使用交互interactive模式。https://www.cnblogs.com/wmy-ncut/p/10172601.html4.plt.subplots前两个数字是子图的行列数figsize(8, 4)估计就是确定figure的大小fig, (ax1, ax2) plt.subplots(1, 2, figsize(8, 4))5.numpy.random.randint(low, highNone, sizeNone, dtype‘l’)函数的作用是返回一个随机整型数范围从低包括到高不包括即[low, high)。如果没有写参数high的值则返回[0,low)的值。size: int or tuple of ints(可选)输出随机数的尺寸比如size (m * n* k)则输出同规模即m * n* k个随机数。默认是None的仅仅返回满足要求的单一随机数。https://blog.csdn.net/u011851421/article/details/83544853即整数生成随机元素的个数或者元组数组的行和列也就是这里生成的是array([1,5,6,8,4,3…]) 共32个元素batch_index np.random.randint(len(train_data), size32)6.现在不用feed_dict{}这种形式了用sess.run(train_op, {tf_input: train_data[batch_index]})7.ax1.cla()plt.cla() # 清除axes轴即清楚当前 figure 中的活动的axes但其他axes保持不变。暂时存疑8.ax1.barhttps://blog.csdn.net/liangzuojiayi/article/details/78187704leftx轴的位置序列一般采用arange函数产生一个序列heighty轴的数值序列也就是柱形图的高度一般就是我们需要展示的数据alpha透明度width为柱形图的宽度一般这是为0.8即可color或facecolor柱形图填充的颜色edgecolor图形边缘颜色label解释每个图像代表的含义linewidth or linewidths or lw边缘or线的宽度bp ax1.bar(c0.1, heightsum((np.argmax(pred_, axis1) c)), width0.2, colorred)后面的先不写了matplotlib有点麻烦补充1.dataframe格式建立https://blog.csdn.net/qq_39161737/article/details/78866399https://blog.csdn.net/weixin_40240670/article/details/80506402加入了save并且单独摘2个测试 data_processing:import pandas as pd from urllib.request import urlretrieve import numpy as np def load_data(downloadTrue): # download data from : http://archive.ics.uci.edu/ml/datasets/CarEvaluation if download: data_path, _ urlretrieve(http://archive.ics.uci.edu/ml/machine-learning-databases/car/car.data, car.csv) print(Downloaded to car.csv) # use pandas to view the data structure col_names [buying, maint, doors, persons, lug_boot, safety, class] data pd.read_csv(car.csv, namescol_names) data_test pd.read_csv(car_test.csv, namescol_names) # print(data_test) #class pandas.core.frame.DataFrame return data, data_test def convert2onehot(data, train): # covert data to onehot representation if train True: print(type(pd.get_dummies(data, prefixdata.columns))) return pd.get_dummies(data, prefixdata.columns) else: x [[0,0,0,1,0,0,0,1,1,0,0,0,1,0,0,0,0,1,0,1,0,0,0,1,0],[0,0,0,1,0,0,0,1,1,0,0,0,1,0,0,0,0,1,0,0,1,0,0,1,0]] return pd.DataFrame(x) if __name__ __main__: data, data_test load_data(downloadTrue) new_data convert2onehot(data, trainTrue) new_data_test convert2onehot(data_test, trainFalse) # new_data_test new_data_test.values.astype(np.float32) # print(自己建立的是, new_data_test) # print(new:, new_data) print(data.head()) print(\nNum of data: , len(data), \n) # 1728 也就是1728行 # view data values for name in data.keys(): print(name, pd.unique(data[name])) print(\n, new_data.head(2)) new_data.to_csv(car_onehot.csv, indexFalse) # new_data_test.to_csv(car_onehot_test.csv, indexFalse)model:import numpy as np import tensorflow as tf import matplotlib.pyplot as plt import data_processing import os tf.app.flags.DEFINE_integer(is_train, 2, 指定是否是训练模型还是拿数据去预测) FLAGS tf.app.flags.FLAGS data, data_test data_processing.load_data(downloadTrue) new_data data_processing.convert2onehot(data, trainTrue) new_data_test data_processing.convert2onehot(data, trainFalse) # prepare training data new_data new_data.values.astype(np.float32) # change to numpy array and float32 转换为numpy数组形式并规定为浮点数 new_data_test new_data_test.values.astype(np.float32) #print(打印, new_data) np.random.shuffle(new_data) sep int(0.7*len(new_data)) train_data new_data[:sep] # training data (70%) test_data new_data[sep:] # test data (30%) realtest_data new_data_test[::] np.random.shuffle(new_data_test) # build network tf_input tf.placeholder(tf.float32, [None, 25], input) #其中最后4个是汽车状态评价 tfx tf_input[:, :21] tfy tf_input[:, 21:] # tfx_test tf_input[:, :21] # tfy_test tf_input[:, 21:] l1 tf.layers.dense(tfx, 128, tf.nn.relu, namel1) l2 tf.layers.dense(l1, 128, tf.nn.relu, namel2) out tf.layers.dense(l2, 4, namel3) prediction tf.nn.softmax(out, namepred) # print(prediction:, prediction) loss tf.losses.softmax_cross_entropy(onehot_labelstfy, logitsout) accuracy tf.metrics.accuracy( # return (acc, update_op), and create 2 local variables labelstf.argmax(tfy, axis1), predictionstf.argmax(out, axis1))[1] opt tf.train.GradientDescentOptimizer(learning_rate0.1) train_op opt.minimize(loss) #进行模型保存 # (2)收集要显示的变量 # 先收集损失和准确率 tf.summary.scalar(losses, loss) tf.summary.scalar(accuracy, accuracy) # (3)合并所有变量op merged tf.summary.merge_all() # 创建模型保存与加载 saver tf.train.Saver() with tf.Session() as sess: sess.run(tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())) # (1)创建一个events文件实例 file_writer tf.summary.FileWriter(./tmp/summary2/, graphsess.graph) # 加载模型 if os.path.exists(./tmp/modelckpt2/checkpoint): saver.restore(sess, ./tmp/modelckpt2/cnn_model) # 注意modelckpt2这个文件夹要自己建立 # 也就是说 模型保存和加载的时候 也就是saver.save或saver.restore的路径需要自己建立 否则会蓝屏 # 但是创建envents实例化的路径可以不用自己建立 # trainingtest1 if FLAGS.is_train 1: accuracies, steps [], [] for t in range(4000): # training batch_index np.random.randint(len(train_data), size32) #生成32个随机数返回array数组 sess.run(train_op, {tf_input: train_data[batch_index]}) #注意feeddict可以不用了 # 运行合变量op写入事件文件当中 summary sess.run(merged, {tf_input: train_data[batch_index]}) file_writer.add_summary(summary, t) if t % 2 0: saver.save(sess, ./tmp/modelckpt2/cnn_model) if t % 50 0: # testing acc_, pred_, loss_ sess.run([accuracy, prediction, loss], {tf_input: test_data}) accuracies.append(acc_) steps.append(t) print(Step: %i % t,| Accurate: %.2f % acc_,| Loss: %.2f % loss_,) # test2 else: for t in range(2): # training batch_index np.random.randint(len(realtest_data), size1) # sess.run(train_op, {tf_input_test: test_real[batch_index]}) acc_, pred_, loss_ sess.run([accuracy, prediction, loss], {tf_input: realtest_data[batch_index]}) print(真实值为, sess.run(tf.argmax(realtest_data[batch_index][:, 21:], axis1))) print(预测值为, sess.run(tf.argmax(pred_, axis1)))