知乎讲解tensorflow2.0之regression详解

最新推荐文章于 2021-09-12 15:32:57 发布

翻译最新推荐文章于 2021-09-12 15:32:57 发布 · 830 阅读

该博客介绍了在TensorFlow 2.0中进行回归分析的步骤，包括数据预处理、数据划分、模型构建以及利用早停法优化训练过程。通过示例展示了如何使用pandas进行数据清洗，利用sns.pairplot进行数据可视化，并用describe()函数进行数据统计分析。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

首先导入需要的库

import pandas as pd
import tensorflow as tf
import tensorflow.keras as keras
import numpy as np
import seaborn as sns
import tensorflow.keras.layers as layers
#import tensorflow.keras.model as Model
import matplotlib.pyplot as plt

下载数据

dataset_path = keras.utils.get_file('auto-mpg.data'， 'https://archive.ics.uci.edu/ml/machine-learning-databases/auto-mpg/auto-mpg.data')
print(dataset_path)
``
结果：
C:\Users\john\.keras\datasets\auto-mpg.data
`
数据读取


column_names = ['MPG','Cylinders','Displacement','Horsepower','Weight',
                'Acceleration', 'Model Year', 'Origin']
raw_dataset = pd.read_csv(dataset_path, names=column_names,
                         na_values='?', comment='\t',
                         sep=' ', skipinitialspace=True)
dataset = raw_dataset.copy()
dataset.tail()   #只输出最后几行

结果:


MPG	Cylinders	Displacement	Horsepower	Weight	Acceleration	Model Year	Origin
102	26.0	4	97.0	46.0	1950.0	21.0	73.0	2.0
103	11.0	8	400.0	150.0	4997.0	14.0	73.0	1.0
104	12.0	8	400.0	167.0	4906.0	12.5	73.0	1.0
105	13.0	8	360.0	170.0	4654.0	13.0	73.0	1.0
106	12.0	8	350.0	180.0	NaN	NaN	NaN	NaN

清洗数据

#清洗数据
print(dataset.tail().isna())

：
MPG Cylinders Displacement Horsepower Weight Acceleration
101 False False False False False False
102 False False False False False False
103 False False False False False False
104 False False False False False False
105 False False False False False False

Model Year USA Europe Japan
101 False False False False
102 False False False False
103 False False False False
104 False False False False
105 False False False False

print(dataset.tail().isna().sum())

:
MPG 0
Cylinders 0
Displacement 0
Horsepower 0
Weight 0
Acceleration 0
Model Year 0
USA 0
Europe 0
Japan 0
dtype: int64
常用的清洗数据方法

dataset = dataset.dropna()  #用dropna()清洗数据
origin = dataset.pop('Origin')
dataset['USA'] = (origin == 1)*1.0
dataset['Europe'] = (origin == 2)*1.0
dataset['Japan'] = (origin == 3)*1.0
dataset.tail()

数据如何划分训练集与测试集

#划分测试集合训练集
train_dataset = dataset.sample(frac=0.8,random_state=0)
test_dataset = dataset.drop(train_dataset.index)

sns.pairplot用法

#不太懂为什么图画出来是这个样子
sns.pairplot(train_dataset[["MPG", "Cylinders", "Displacement", "Weight"]], diag_kind="kde")

在这里插入图片描述
pandas的describe()函数详解

#整体数据统计
train_stats = train_dataset.describe()
train_stats.pop("MPG")     #把MPG去出
train_stats = train_stats.transpose()
train_stats

#标准化数据
def norm(x):
    return (x - train_stats['mean']) / train_stats['std']
normed_train_data = norm(train_dataset)
normed_test_data = norm(test_dataset)
example_batch = normed_train_data[:10]
example_result = model.predict(example_batch)
example_result

结果：
array([[ 9.936947], [13.620609], [17.324503], [13.222853], [13.925085], [24.418594], [19.316984], [22.01549 ], [23.534117], [21.643145]], dtype=float32)

#构建模型
def build_model():
    model = keras.Sequential([
        layers.Dense(64, activation='relu', input_shape=[len(train_dataset.keys())]),
        layers.Dense(64, activation='relu'),
        layers.Dense(1)
    ])
    optimizer = tf.keras.optimizers.RMSprop(0.001)
    model.compile(loss='mse',
                 optimizer=optimizer,
                 metrics=['mae', 'mse'])  #metrics怎么用的？
    return model
model = build_model()
model.summary()


#训练模型
class PrintDot(keras.callbacks.Callback):
    def on_epoch_end(self, epoch, logs):
        if epoch % 100 == 0: print('')
        print('.', end='')

EPOCHS = 1000

history = model.fit(
  normed_train_data, train_labels,
  epochs=EPOCHS, validation_split = 0.2, verbose=0,
  callbacks=[PrintDot()]   #callbacks怎么用的？
  
)

#查看记录
hist = pd.DataFrame(history.history)
hist['epoch'] = history.epoch
hist.tail()
``

def plot_history(history):
hist = pd.DataFrame(history.history)
hist[‘epoch’] = history.epoch

plt.figure()
plt.xlabel('Epoch')
plt.ylabel('Mean Abs Error [MPG]')
plt.plot(hist['epoch'], hist['mae'],
       label='Train Error')
plt.plot(hist['epoch'], hist['val_mae'],
       label = 'Val Error')
plt.ylim([0,5])
plt.legend()

plt.figure()
plt.xlabel('Epoch')
plt.ylabel('Mean Square Error [$MPG^2$]')
plt.plot(hist['epoch'], hist['mse'],
       label='Train Error')
plt.plot(hist['epoch'], hist['val_mse'],
       label = 'Val Error')
plt.ylim([0,20])
plt.legend()
plt.show()

plot_history(history)

#使用早停法
model = build_model()

early_stop = keras.callbacks.EarlyStopping(monitor=‘val_loss’, patience=10)

history = model.fit(normed_train_data, train_labels, epochs=EPOCHS,
validation_split = 0.2, verbose=0, callbacks=[early_stop, PrintDot()])

plot_history(history)