► 代码示例 / 结构化数据 / 从零开始的结构化数据分类

从零开始的结构化数据分类

作者： fchollet
创建日期 2020/06/09
上次修改 2020/06/09
描述： 包含数值和分类特征的结构化数据的二元分类。

ⓘ 本示例使用 Keras 3

在 Colab 中查看 • GitHub 源代码

简介

本示例演示如何从原始 CSV 文件开始进行结构化数据分类。我们的数据包括数值和分类特征。我们将使用 Keras 预处理层来归一化数值特征并向量化分类特征。

请注意，此示例应使用 TensorFlow 2.5 或更高版本运行。

数据集

我们的数据集 由克利夫兰诊所基金会提供，用于心脏病研究。它是一个包含 303 行的 CSV 文件。每行包含有关一名患者的信息（一个样本），每列描述患者的一个属性（一个特征）。我们使用这些特征来预测患者是否患有心脏病（二元分类）。

以下是每个特征的描述

列 描述 特征类型
年龄 以年为单位的年龄 数值
性别 (1 = 男性；0 = 女性) 分类
CP 胸痛类型 (0, 1, 2, 3, 4) 分类
Trestbpd 静息血压（入院时以 mmHg 为单位） 数值
Chol 血清胆固醇，以 mg/dl 为单位 数值
FBS 空腹血糖，以 120 mg/dl 为单位 (1 = 是；0 = 否) 分类
RestECG 静息心电图结果 (0, 1, 2) 分类
Thalach 达到的最大心率 数值
Exang 运动诱发的 angina (1 = 是；0 = 否) 分类
Oldpeak 运动诱发的 ST 段压低，相对于静息状态 数值
斜率 峰值运动 ST 段的斜率 数值
CA 通过荧光造影显示的冠状动脉主要血管数 (0-3) 数值和分类
Thal 3 = 正常；6 = 固定缺陷；7 = 可逆缺陷 分类
目标 心脏病诊断 (1 = 是；0 = 否) 目标

设置

import os

# TensorFlow is the only backend that supports string inputs.
os.environ["KERAS_BACKEND"] = "tensorflow"

import tensorflow as tf
import pandas as pd
import keras
from keras import layers

准备数据

让我们下载数据并将其加载到 Pandas 数据帧中

file_url = "http://storage.googleapis.com/download.tensorflow.org/data/heart.csv"
dataframe = pd.read_csv(file_url)

数据集包含 303 个样本，每个样本有 14 列（13 个特征，加上目标标签）

dataframe.shape

(303, 14)

以下是一些样本的预览

dataframe.head()
年龄 性别 cp trestbps chol fbs restecg thalach exang oldpeak 斜率 ca thal 目标
0 63 1 1 145 233 1 2 150 0 2.3 3 0 固定 0
1 67 1 4 160 286 0 2 108 1 1.5 2 3 正常 1
2 67 1 4 120 229 0 2 129 1 2.6 2 2 可逆 0
3 37 1 3 130 250 0 0 187 0 3.5 3 0 正常 0
4 41 0 2 130 204 0 2 172 0 1.4 1 0 正常 0

最后一列“目标”表示患者是否患有心脏病 (1) 或没有 (0)。

让我们将数据拆分为训练集和验证集

val_dataframe = dataframe.sample(frac=0.2, random_state=1337)
train_dataframe = dataframe.drop(val_dataframe.index)

print(
    f"Using {len(train_dataframe)} samples for training "
    f"and {len(val_dataframe)} for validation"
)

Using 242 samples for training and 61 for validation

让我们为每个数据帧生成 tf.data.Dataset 对象

def dataframe_to_dataset(dataframe):
    dataframe = dataframe.copy()
    labels = dataframe.pop("target")
    ds = tf.data.Dataset.from_tensor_slices((dict(dataframe), labels))
    ds = ds.shuffle(buffer_size=len(dataframe))
    return ds


train_ds = dataframe_to_dataset(train_dataframe)
val_ds = dataframe_to_dataset(val_dataframe)

每个 Dataset 都产生一个元组 (输入, 目标)，其中 输入 是一个特征字典，目标 是值 0 或 1

for x, y in train_ds.take(1):
    print("Input:", x)
    print("Target:", y)

Input: {'age': <tf.Tensor: shape=(), dtype=int64, numpy=64>, 'sex': <tf.Tensor: shape=(), dtype=int64, numpy=1>, 'cp': <tf.Tensor: shape=(), dtype=int64, numpy=4>, 'trestbps': <tf.Tensor: shape=(), dtype=int64, numpy=128>, 'chol': <tf.Tensor: shape=(), dtype=int64, numpy=263>, 'fbs': <tf.Tensor: shape=(), dtype=int64, numpy=0>, 'restecg': <tf.Tensor: shape=(), dtype=int64, numpy=0>, 'thalach': <tf.Tensor: shape=(), dtype=int64, numpy=105>, 'exang': <tf.Tensor: shape=(), dtype=int64, numpy=1>, 'oldpeak': <tf.Tensor: shape=(), dtype=float64, numpy=0.2>, 'slope': <tf.Tensor: shape=(), dtype=int64, numpy=2>, 'ca': <tf.Tensor: shape=(), dtype=int64, numpy=1>, 'thal': <tf.Tensor: shape=(), dtype=string, numpy=b'reversible'>}
Target: tf.Tensor(0, shape=(), dtype=int64)

让我们对数据集进行批处理

train_ds = train_ds.batch(32)
val_ds = val_ds.batch(32)

使用 Keras 层进行特征预处理

以下特征是作为整数编码的分类特征

  • 性别
  • cp
  • fbs
  • restecg
  • exang
  • ca

我们将使用独热编码对这些特征进行编码。我们这里有两个选择

  • 使用 CategoryEncoding()，它需要知道输入值的范围，并且会在输入超出范围时报错。
  • 使用 IntegerLookup()，它将为输入构建一个查找表，并为未知输入值保留一个输出索引。

对于本示例，我们想要一个简单的解决方案，它将在推理时处理超出范围的输入，因此我们将使用 IntegerLookup()。

我们还有一个作为字符串编码的分类特征：thal。我们将创建所有可能特征的索引，并使用 StringLookup() 层对输出进行编码。

最后，以下特征是连续数值特征

  • 年龄
  • trestbps
  • chol
  • thalach
  • oldpeak
  • 斜率

对于这些特征中的每一个，我们将使用 Normalization() 层来确保每个特征的均值为 0，标准差为 1。

下面，我们定义了两个实用函数来执行这些操作

  • encode_numerical_feature 用于对数值特征应用特征级归一化。
  • encode_categorical_feature 用于对字符串或整数分类特征进行独热编码。
def encode_numerical_feature(feature, name, dataset):
    # Create a Normalization layer for our feature
    normalizer = layers.Normalization()

    # Prepare a Dataset that only yields our feature
    feature_ds = dataset.map(lambda x, y: x[name])
    feature_ds = feature_ds.map(lambda x: tf.expand_dims(x, -1))

    # Learn the statistics of the data
    normalizer.adapt(feature_ds)

    # Normalize the input feature
    encoded_feature = normalizer(feature)
    return encoded_feature


def encode_categorical_feature(feature, name, dataset, is_string):
    lookup_class = layers.StringLookup if is_string else layers.IntegerLookup
    # Create a lookup layer which will turn strings into integer indices
    lookup = lookup_class(output_mode="binary")

    # Prepare a Dataset that only yields our feature
    feature_ds = dataset.map(lambda x, y: x[name])
    feature_ds = feature_ds.map(lambda x: tf.expand_dims(x, -1))

    # Learn the set of possible string values and assign them a fixed integer index
    lookup.adapt(feature_ds)

    # Turn the string input into integer indices
    encoded_feature = lookup(feature)
    return encoded_feature

构建模型

完成这些操作后，我们可以创建我们的端到端模型

# Categorical features encoded as integers
sex = keras.Input(shape=(1,), name="sex", dtype="int64")
cp = keras.Input(shape=(1,), name="cp", dtype="int64")
fbs = keras.Input(shape=(1,), name="fbs", dtype="int64")
restecg = keras.Input(shape=(1,), name="restecg", dtype="int64")
exang = keras.Input(shape=(1,), name="exang", dtype="int64")
ca = keras.Input(shape=(1,), name="ca", dtype="int64")

# Categorical feature encoded as string
thal = keras.Input(shape=(1,), name="thal", dtype="string")

# Numerical features
age = keras.Input(shape=(1,), name="age")
trestbps = keras.Input(shape=(1,), name="trestbps")
chol = keras.Input(shape=(1,), name="chol")
thalach = keras.Input(shape=(1,), name="thalach")
oldpeak = keras.Input(shape=(1,), name="oldpeak")
slope = keras.Input(shape=(1,), name="slope")

all_inputs = [
    sex,
    cp,
    fbs,
    restecg,
    exang,
    ca,
    thal,
    age,
    trestbps,
    chol,
    thalach,
    oldpeak,
    slope,
]

# Integer categorical features
sex_encoded = encode_categorical_feature(sex, "sex", train_ds, False)
cp_encoded = encode_categorical_feature(cp, "cp", train_ds, False)
fbs_encoded = encode_categorical_feature(fbs, "fbs", train_ds, False)
restecg_encoded = encode_categorical_feature(restecg, "restecg", train_ds, False)
exang_encoded = encode_categorical_feature(exang, "exang", train_ds, False)
ca_encoded = encode_categorical_feature(ca, "ca", train_ds, False)

# String categorical features
thal_encoded = encode_categorical_feature(thal, "thal", train_ds, True)

# Numerical features
age_encoded = encode_numerical_feature(age, "age", train_ds)
trestbps_encoded = encode_numerical_feature(trestbps, "trestbps", train_ds)
chol_encoded = encode_numerical_feature(chol, "chol", train_ds)
thalach_encoded = encode_numerical_feature(thalach, "thalach", train_ds)
oldpeak_encoded = encode_numerical_feature(oldpeak, "oldpeak", train_ds)
slope_encoded = encode_numerical_feature(slope, "slope", train_ds)

all_features = layers.concatenate(
    [
        sex_encoded,
        cp_encoded,
        fbs_encoded,
        restecg_encoded,
        exang_encoded,
        slope_encoded,
        ca_encoded,
        thal_encoded,
        age_encoded,
        trestbps_encoded,
        chol_encoded,
        thalach_encoded,
        oldpeak_encoded,
    ]
)
x = layers.Dense(32, activation="relu")(all_features)
x = layers.Dropout(0.5)(x)
output = layers.Dense(1, activation="sigmoid")(x)
model = keras.Model(all_inputs, output)
model.compile("adam", "binary_crossentropy", metrics=["accuracy"])

让我们可视化我们的连接图

# `rankdir='LR'` is to make the graph horizontal.
keras.utils.plot_model(model, show_shapes=True, rankdir="LR")

png

训练模型

model.fit(train_ds, epochs=50, validation_data=val_ds)

Epoch 1/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 5s 46ms/step - accuracy: 0.3932 - loss: 0.8749 - val_accuracy: 0.3303 - val_loss: 0.7814
Epoch 2/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 1s 7ms/step - accuracy: 0.4262 - loss: 0.8375 - val_accuracy: 0.4914 - val_loss: 0.6980
Epoch 3/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 8ms/step - accuracy: 0.4835 - loss: 0.7350 - val_accuracy: 0.6541 - val_loss: 0.6320
Epoch 4/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.5932 - loss: 0.6665 - val_accuracy: 0.7543 - val_loss: 0.5743
Epoch 5/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.5861 - loss: 0.6600 - val_accuracy: 0.7683 - val_loss: 0.5360
Epoch 6/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.6489 - loss: 0.6020 - val_accuracy: 0.7748 - val_loss: 0.4998
Epoch 7/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.6880 - loss: 0.5668 - val_accuracy: 0.7699 - val_loss: 0.4800
Epoch 8/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.7572 - loss: 0.5009 - val_accuracy: 0.7559 - val_loss: 0.4573
Epoch 9/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.7492 - loss: 0.5192 - val_accuracy: 0.8060 - val_loss: 0.4414
Epoch 10/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 8ms/step - accuracy: 0.7212 - loss: 0.4973 - val_accuracy: 0.8077 - val_loss: 0.4259
Epoch 11/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.7616 - loss: 0.4704 - val_accuracy: 0.7904 - val_loss: 0.4143
Epoch 12/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8374 - loss: 0.4342 - val_accuracy: 0.7872 - val_loss: 0.4061
Epoch 13/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.7863 - loss: 0.4630 - val_accuracy: 0.7888 - val_loss: 0.3980
Epoch 14/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.7742 - loss: 0.4492 - val_accuracy: 0.7996 - val_loss: 0.3998
Epoch 15/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8083 - loss: 0.4280 - val_accuracy: 0.8060 - val_loss: 0.3855
Epoch 16/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8058 - loss: 0.4191 - val_accuracy: 0.8217 - val_loss: 0.3819
Epoch 17/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8071 - loss: 0.4111 - val_accuracy: 0.8389 - val_loss: 0.3763
Epoch 18/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 8ms/step - accuracy: 0.8533 - loss: 0.3676 - val_accuracy: 0.8373 - val_loss: 0.3792
Epoch 19/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8170 - loss: 0.3850 - val_accuracy: 0.8357 - val_loss: 0.3744
Epoch 20/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8207 - loss: 0.3767 - val_accuracy: 0.8168 - val_loss: 0.3759
Epoch 21/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8151 - loss: 0.3596 - val_accuracy: 0.8217 - val_loss: 0.3685
Epoch 22/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.7988 - loss: 0.4087 - val_accuracy: 0.8184 - val_loss: 0.3701
Epoch 23/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8180 - loss: 0.3632 - val_accuracy: 0.8217 - val_loss: 0.3614
Epoch 24/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8295 - loss: 0.3504 - val_accuracy: 0.8200 - val_loss: 0.3683
Epoch 25/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8386 - loss: 0.3864 - val_accuracy: 0.8200 - val_loss: 0.3655
Epoch 26/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8482 - loss: 0.3345 - val_accuracy: 0.8044 - val_loss: 0.3639
Epoch 27/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 8ms/step - accuracy: 0.8340 - loss: 0.3470 - val_accuracy: 0.8077 - val_loss: 0.3616
Epoch 28/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8418 - loss: 0.3684 - val_accuracy: 0.8060 - val_loss: 0.3629
Epoch 29/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8309 - loss: 0.3147 - val_accuracy: 0.8060 - val_loss: 0.3637
Epoch 30/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8722 - loss: 0.3151 - val_accuracy: 0.8044 - val_loss: 0.3672
Epoch 31/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 8ms/step - accuracy: 0.8746 - loss: 0.3043 - val_accuracy: 0.8060 - val_loss: 0.3637
Epoch 32/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8794 - loss: 0.3245 - val_accuracy: 0.8200 - val_loss: 0.3685
Epoch 33/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 8ms/step - accuracy: 0.8644 - loss: 0.3541 - val_accuracy: 0.8357 - val_loss: 0.3714
Epoch 34/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8867 - loss: 0.3007 - val_accuracy: 0.8373 - val_loss: 0.3680
Epoch 35/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8737 - loss: 0.3168 - val_accuracy: 0.8357 - val_loss: 0.3695
Epoch 36/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8191 - loss: 0.3298 - val_accuracy: 0.8357 - val_loss: 0.3736
Epoch 37/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8613 - loss: 0.3543 - val_accuracy: 0.8357 - val_loss: 0.3745
Epoch 38/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8835 - loss: 0.2835 - val_accuracy: 0.8357 - val_loss: 0.3707
Epoch 39/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8784 - loss: 0.2893 - val_accuracy: 0.8357 - val_loss: 0.3716
Epoch 40/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8919 - loss: 0.2587 - val_accuracy: 0.8168 - val_loss: 0.3770
Epoch 41/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8882 - loss: 0.2660 - val_accuracy: 0.8217 - val_loss: 0.3674
Epoch 42/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8790 - loss: 0.2931 - val_accuracy: 0.8200 - val_loss: 0.3723
Epoch 43/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8851 - loss: 0.2892 - val_accuracy: 0.8200 - val_loss: 0.3733
Epoch 44/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8504 - loss: 0.3189 - val_accuracy: 0.8200 - val_loss: 0.3755
Epoch 45/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8610 - loss: 0.3116 - val_accuracy: 0.8184 - val_loss: 0.3788
Epoch 46/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 8ms/step - accuracy: 0.8956 - loss: 0.2544 - val_accuracy: 0.8184 - val_loss: 0.3738
Epoch 47/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.9080 - loss: 0.2895 - val_accuracy: 0.8217 - val_loss: 0.3750
Epoch 48/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8706 - loss: 0.2993 - val_accuracy: 0.8217 - val_loss: 0.3757
Epoch 49/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 7ms/step - accuracy: 0.8724 - loss: 0.2979 - val_accuracy: 0.8184 - val_loss: 0.3781
Epoch 50/50
 8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 8ms/step - accuracy: 0.8609 - loss: 0.2937 - val_accuracy: 0.8217 - val_loss: 0.3791

<keras.src.callbacks.history.History at 0x7efc32e01780>

我们很快就能达到 80% 的验证准确率。

对新数据进行推理

要获取新样本的预测结果，您可以简单地调用 model.predict()。您只需要做两件事

  1. 将标量包装到列表中，以便具有批次维度（模型只处理数据批次，而不是单个样本）
  2. 对每个特征调用 convert_to_tensor
sample = {
    "age": 60,
    "sex": 1,
    "cp": 1,
    "trestbps": 145,
    "chol": 233,
    "fbs": 1,
    "restecg": 2,
    "thalach": 150,
    "exang": 0,
    "oldpeak": 2.3,
    "slope": 3,
    "ca": 0,
    "thal": "fixed",
}

input_dict = {name: tf.convert_to_tensor([value]) for name, value in sample.items()}
predictions = model.predict(input_dict)

print(
    f"This particular patient had a {100 * predictions[0][0]:.1f} "
    "percent probability of having a heart disease, "
    "as evaluated by our model."
)

 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 252ms/step
This particular patient had a 27.6 percent probability of having a heart disease, as evaluated by our model.
从零开始的结构化数据分类
◆ 简介
数据集
◆ 设置
◆ 准备数据
◆ 使用 Keras 层进行特征预处理
◆ 构建模型
◆ 训练模型
◆ 对新数据进行推理