[机器学习] 低代码机器学习工具PyCaret库使用指北

pycaret · 浏览次数 : 0

小编点评

Sure, here is the content generated based on the context: **Feature Transformation** * Feature transformation is a technique to change the distribution shape of data, resulting in data being normally distributed. * PyCaret enables feature transformation by setting the `transformation` parameter to `True` in the `setup` function. * Two methods for feature transformation are available: `yeo-johnson` and splitting. **Target Transformation** * Target transformation changes the target variable, unlike feature transformation that changes the feature's distribution. * Target transformation is only available in the `pycaret.regression` module. * Setting the `transformation` parameter to `True` in the `setup` function enables target transformation. **Output** * The output of the code will be a dictionary containing the configuration settings for the feature and target transformations. * The `X_transformed` key will contain the feature transformation settings, and the `target` key will contain the target transformation settings.

正文

PyCaret是一个开源、低代码Python机器学习库,能够自动化机器学习工作流程。它是一个端到端的机器学习和模型管理工具,极大地加快了实验周期,提高了工作效率。PyCaret本质上是围绕几个机器学习库和框架(如scikit-learn、XGBoost、LightGBM、CatBoost、spaCy、Optuna、Hyperopt、Ray等)的Python包装器,与其他开源机器学习库相比,PyCaret可以用少量代码取代数百行代码。PyCaret开源仓库地址:pycaret,官方文档地址为:pycaret-docs

PyCaret基础版安装命令如下:

pip install pycaret

完整版安装代码如下:

pip install pycaret[full]

此外以下模型可以调用GPU

  • Extreme Gradient Boosting
  • Catboost
  • Light Gradient Boosting(需要安装lightgbm)
  • Logistic Regression, Ridge Classifier, Random Forest, K Neighbors Classifier, K Neighbors Regressor, Support Vector Machine, Linear Regression, Ridge Regression, Lasso Regression(需要安装cuml0.15版本以上)
# 查看pycaret版本
import pycaret
pycaret.__version__
'3.3.2'

1 快速入门

PyCaret支持多种机器学习任务,包括分类、回归、聚类、异常检测和时序预测。本节主要介绍如何利用PyCaret构建相关任务模型的基础使用方法。关于更详细的PyCaret任务模型使用,请参考:

Topic NotebookLink
二分类BinaryClassification link
多分类MulticlassClassification link
回归Regression link
聚类Clustering link
异常检测AnomalyDetection link
时序预测TimeSeriesForecasting link

1.1 分类

PyCaret的classification模块是一个可用于二分类或多分类的模块,用于将元素分类到不同的组中。一些常见的用例包括预测客户是否违约、预测客户是否流失、以及诊断疾病(阳性或阴性)。示例代码如下所示:

数据准备

加载糖尿病示例数据集:

from pycaret.datasets import get_data
# 从本地加载数据,注意dataset是数据的文件名
data = get_data(dataset='./datasets/diabetes', verbose=False)
# 从pycaret开源仓库下载公开数据
# data = get_data('diabetes', verbose=False)
# 查看数据类型和数据维度
type(data), data.shape
(pandas.core.frame.DataFrame, (768, 9))
# 最后一列表示是否为糖尿病患者,其他列为特征列
data.head()
Number of times pregnant Plasma glucose concentration a 2 hours in an oral glucose tolerance test Diastolic blood pressure (mm Hg) Triceps skin fold thickness (mm) 2-Hour serum insulin (mu U/ml) Body mass index (weight in kg/(height in m)^2) Diabetes pedigree function Age (years) Class variable
0 6 148 72 35 0 33.6 0.627 50 1
1 1 85 66 29 0 26.6 0.351 31 0
2 8 183 64 0 0 23.3 0.672 32 1
3 1 89 66 23 94 28.1 0.167 21 0
4 0 137 40 35 168 43.1 2.288 33 1

利用PyCaret核心函数setup,初始化建模环境并准备数据以供模型训练和评估使用:

from pycaret.classification import ClassificationExperiment
s = ClassificationExperiment()
# target目标列,session_id设置随机数种子, preprocesss是否清洗数据,train_size训练集比例, normalize是否归一化数据, normalize_method归一化方式
s.setup(data, target = 'Class variable', session_id = 0, verbose= False, train_size = 0.7, normalize = True, normalize_method = 'minmax')
<pycaret.classification.oop.ClassificationExperiment at 0x200b939df40>

查看基于setup函数创建的变量:

s.get_config()
{'USI',
 'X',
 'X_test',
 'X_test_transformed',
 'X_train',
 'X_train_transformed',
 'X_transformed',
 '_available_plots',
 '_ml_usecase',
 'data',
 'dataset',
 'dataset_transformed',
 'exp_id',
 'exp_name_log',
 'fix_imbalance',
 'fold_generator',
 'fold_groups_param',
 'fold_shuffle_param',
 'gpu_n_jobs_param',
 'gpu_param',
 'html_param',
 'idx',
 'is_multiclass',
 'log_plots_param',
 'logging_param',
 'memory',
 'n_jobs_param',
 'pipeline',
 'seed',
 'target_param',
 'test',
 'test_transformed',
 'train',
 'train_transformed',
 'variable_and_property_keys',
 'variables',
 'y',
 'y_test',
 'y_test_transformed',
 'y_train',
 'y_train_transformed',
 'y_transformed'}

查看归一化的数据:

s.get_config('X_train_transformed')
Number of times pregnant Plasma glucose concentration a 2 hours in an oral glucose tolerance test Diastolic blood pressure (mm Hg) Triceps skin fold thickness (mm) 2-Hour serum insulin (mu U/ml) Body mass index (weight in kg/(height in m)^2) Diabetes pedigree function Age (years)
34 0.588235 0.613065 0.639344 0.313131 0.000000 0.411326 0.185312 0.400000
221 0.117647 0.793970 0.737705 0.000000 0.000000 0.470939 0.310418 0.750000
531 0.000000 0.537688 0.622951 0.000000 0.000000 0.675112 0.259607 0.050000
518 0.764706 0.381910 0.491803 0.000000 0.000000 0.488823 0.043553 0.333333
650 0.058824 0.457286 0.442623 0.252525 0.118203 0.375559 0.066610 0.033333
... ... ... ... ... ... ... ... ...
628 0.294118 0.643216 0.655738 0.000000 0.000000 0.515648 0.028181 0.400000
456 0.058824 0.678392 0.442623 0.000000 0.000000 0.397914 0.260034 0.683333
398 0.176471 0.412060 0.573770 0.000000 0.000000 0.314456 0.132792 0.066667
6 0.176471 0.391960 0.409836 0.323232 0.104019 0.461997 0.072588 0.083333
294 0.000000 0.809045 0.409836 0.000000 0.000000 0.326379 0.075149 0.733333

537 rows × 8 columns

绘制某列数据的柱状图:

s.get_config('X_train_transformed')['Number of times pregnant'].hist()
<AxesSubplot:>

png

当然也可以利用如下代码创建任务示例来初始化环境:

from pycaret.classification import setup
# s = setup(data, target = 'Class variable', session_id = 0, preprocess = True, train_size = 0.7, verbose = False)

模型训练与评估

PyCaret提供了compare_models函数,通过使用默认的10折交叉验证来训练和评估模型库中所有可用估计器的性能:

best = s.compare_models()
# 选择某些模型进行比较
# best = s.compare_models(include = ['dt', 'rf', 'et', 'gbc', 'lightgbm'])
# 按照召回率返回n_select性能最佳的模型
# best_recall_models_top3 = s.compare_models(sort = 'Recall', n_select = 3)
  Model Accuracy AUC Recall Prec. F1 Kappa MCC TT (Sec)
lr Logistic Regression 0.7633 0.8132 0.4968 0.7436 0.5939 0.4358 0.4549 0.2720
ridge Ridge Classifier 0.7633 0.8113 0.5178 0.7285 0.6017 0.4406 0.4560 0.0090
lda Linear Discriminant Analysis 0.7633 0.8110 0.5497 0.7069 0.6154 0.4489 0.4583 0.0080
ada Ada Boost Classifier 0.7465 0.7768 0.5655 0.6580 0.6051 0.4208 0.4255 0.0190
svm SVM - Linear Kernel 0.7408 0.8087 0.5921 0.6980 0.6020 0.4196 0.4480 0.0080
nb Naive Bayes 0.7391 0.7939 0.5442 0.6515 0.5857 0.3995 0.4081 0.0080
rf Random Forest Classifier 0.7337 0.8033 0.5406 0.6331 0.5778 0.3883 0.3929 0.0350
et Extra Trees Classifier 0.7298 0.7899 0.5181 0.6416 0.5677 0.3761 0.3840 0.0450
gbc Gradient Boosting Classifier 0.7281 0.8007 0.5567 0.6267 0.5858 0.3857 0.3896 0.0260
lightgbm Light Gradient Boosting Machine 0.7242 0.7811 0.5827 0.6096 0.5935 0.3859 0.3876 0.0860
qda Quadratic Discriminant Analysis 0.7150 0.7875 0.4962 0.6225 0.5447 0.3428 0.3524 0.0080
knn K Neighbors Classifier 0.7131 0.7425 0.5287 0.6005 0.5577 0.3480 0.3528 0.2200
dt Decision Tree Classifier 0.6685 0.6461 0.5722 0.5266 0.5459 0.2868 0.2889 0.0100
dummy Dummy Classifier 0.6518 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0120

返回当前设置中所有经过训练的模型中的最佳模型:

best_ml = s.automl()
# best_ml
# 打印效果最佳的模型
print(best)
LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,
                   intercept_scaling=1, l1_ratio=None, max_iter=1000,
                   multi_class='auto', n_jobs=None, penalty='l2',
                   random_state=0, solver='lbfgs', tol=0.0001, verbose=0,
                   warm_start=False)

数据可视化

PyCaret也提供了plot_model函数可视化模型的评估指标,plot_model函数中的plot用于设置评估指标类型。plot可用参数如下(注意并不是所有的模型都支持以下评估指标):

  • pipeline: Schematic drawing of the preprocessing pipeline
  • auc: Area Under the Curve
  • threshold: Discrimination Threshold
  • pr: Precision Recall Curve
  • confusion_matrix: Confusion Matrix
  • error: Class Prediction Error
  • class_report: Classification Report
  • boundary: Decision Boundary
  • rfe: Recursive Feature Selection
  • learning: Learning Curve
  • manifold: Manifold Learning
  • calibration: Calibration Curve
  • vc: Validation Curve
  • dimension: Dimension Learning
  • feature: Feature Importance
  • feature_all: Feature Importance (All)
  • parameter: Model Hyperparameter
  • lift: Lift Curve
  • gain: Gain Chart
  • tree: Decision Tree
  • ks: KS Statistic Plot
# 提取所有模型预测结果
models_results = s.pull()
models_results
Model Accuracy AUC Recall Prec. F1 Kappa MCC TT (Sec)
lr Logistic Regression 0.7633 0.8132 0.4968 0.7436 0.5939 0.4358 0.4549 0.272
ridge Ridge Classifier 0.7633 0.8113 0.5178 0.7285 0.6017 0.4406 0.4560 0.009
lda Linear Discriminant Analysis 0.7633 0.8110 0.5497 0.7069 0.6154 0.4489 0.4583 0.008
ada Ada Boost Classifier 0.7465 0.7768 0.5655 0.6580 0.6051 0.4208 0.4255 0.019
svm SVM - Linear Kernel 0.7408 0.8087 0.5921 0.6980 0.6020 0.4196 0.4480 0.008
nb Naive Bayes 0.7391 0.7939 0.5442 0.6515 0.5857 0.3995 0.4081 0.008
rf Random Forest Classifier 0.7337 0.8033 0.5406 0.6331 0.5778 0.3883 0.3929 0.035
et Extra Trees Classifier 0.7298 0.7899 0.5181 0.6416 0.5677 0.3761 0.3840 0.045
gbc Gradient Boosting Classifier 0.7281 0.8007 0.5567 0.6267 0.5858 0.3857 0.3896 0.026
lightgbm Light Gradient Boosting Machine 0.7242 0.7811 0.5827 0.6096 0.5935 0.3859 0.3876 0.086
qda Quadratic Discriminant Analysis 0.7150 0.7875 0.4962 0.6225 0.5447 0.3428 0.3524 0.008
knn K Neighbors Classifier 0.7131 0.7425 0.5287 0.6005 0.5577 0.3480 0.3528 0.220
dt Decision Tree Classifier 0.6685 0.6461 0.5722 0.5266 0.5459 0.2868 0.2889 0.010
dummy Dummy Classifier 0.6518 0.5000 0.0000 0.0000 0.0000 0.0000 0.0000 0.012
s.plot_model(best, plot = 'confusion_matrix')

png

如果在jupyter环境,可以通过evaluate_model函数来交互式展示模型的性能:

# s.evaluate_model(best)

模型预测

predict_model函数实现对数据进行预测,并返回包含预测标签prediction_label和分数prediction_score的Pandas表格。当data为None时,它预测测试集(在设置功能期间创建)上的标签和分数。

# 预测整个数据集
res = s.predict_model(best, data=data)
# 查看各行预测结果
# res
  Model Accuracy AUC Recall Prec. F1 Kappa MCC
0 Logistic Regression 0.7708 0.8312 0.5149 0.7500 0.6106 0.4561 0.4723
# 预测用于数据训练的测试集
res = s.predict_model(best)
  Model Accuracy AUC Recall Prec. F1 Kappa MCC
0 Logistic Regression 0.7576 0.8553 0.5062 0.7193 0.5942 0.4287 0.4422

模型保存与导入

# 保存模型到本地
_ = s.save_model(best, 'best_model', verbose = False)
# 导入模型
model = s.load_model( 'best_model')
# 查看模型结构
# model
Transformation Pipeline and Model Successfully Loaded
# 预测整个数据集
res = s.predict_model(model, data=data)
  Model Accuracy AUC Recall Prec. F1 Kappa MCC
0 Logistic Regression 0.7708 0.8312 0.5149 0.7500 0.6106 0.4561 0.4723

1.2 回归

PyCaret提供了regression模型实现回归任务,regression模块与classification模块使用方法一致。

# 加载保险费用示例数据集
from pycaret.datasets import get_data
data = get_data(dataset='./datasets/insurance', verbose=False)
# 从网络下载
# data = get_data(dataset='insurance', verbose=False)
data.head()
age sex bmi children smoker region charges
0 19 female 27.900 0 yes southwest 16884.92400
1 18 male 33.770 1 no southeast 1725.55230
2 28 male 33.000 3 no southeast 4449.46200
3 33 male 22.705 0 no northwest 21984.47061
4 32 male 28.880 0 no northwest 3866.85520
# 创建数据管道
from pycaret.regression import RegressionExperiment
s = RegressionExperiment()
# 预测charges列
s.setup(data, target = 'charges', session_id = 0)
# 另一种数据管道创建方式
# from pycaret.regression import *
# s = setup(data, target = 'charges', session_id = 0)
  Description Value
0 Session id 0
1 Target charges
2 Target type Regression
3 Original data shape (1338, 7)
4 Transformed data shape (1338, 10)
5 Transformed train set shape (936, 10)
6 Transformed test set shape (402, 10)
7 Numeric features 3
8 Categorical features 3
9 Preprocess True
10 Imputation type simple
11 Numeric imputation mean
12 Categorical imputation mode
13 Maximum one-hot encoding 25
14 Encoding method None
15 Fold Generator KFold
16 Fold Number 10
17 CPU Jobs -1
18 Use GPU False
19 Log Experiment False
20 Experiment Name reg-default-name
21 USI eb9d
<pycaret.regression.oop.RegressionExperiment at 0x200dedc2d30>
# 评估各类模型
best = s.compare_models()
  Model MAE MSE RMSE R2 RMSLE MAPE TT (Sec)
gbr Gradient Boosting Regressor 2723.2453 23787529.5872 4832.4785 0.8254 0.4427 0.3140 0.0550
lightgbm Light Gradient Boosting Machine 2998.1311 25738691.2181 5012.2404 0.8106 0.5525 0.3709 0.1140
rf Random Forest Regressor 2915.7018 26780127.0016 5109.5098 0.8031 0.4855 0.3520 0.0670
et Extra Trees Regressor 2841.8257 28559316.9533 5243.5828 0.7931 0.4671 0.3218 0.0670
ada AdaBoost Regressor 4180.2669 28289551.0048 5297.6817 0.7886 0.5935 0.6545 0.0210
ridge Ridge Regression 4304.2640 38786967.4768 6188.6966 0.7152 0.5794 0.4283 0.0230
lar Least Angle Regression 4293.9886 38781666.5991 6188.3301 0.7151 0.5893 0.4263 0.0210
llar Lasso Least Angle Regression 4294.2135 38780221.0039 6188.1906 0.7151 0.5891 0.4264 0.0200
br Bayesian Ridge 4299.8532 38785479.0984 6188.6026 0.7151 0.5784 0.4274 0.0200
lasso Lasso Regression 4294.2186 38780210.5665 6188.1898 0.7151 0.5892 0.4264 0.0370
lr Linear Regression 4293.9886 38781666.5991 6188.3301 0.7151 0.5893 0.4263 0.0350
dt Decision Tree Regressor 3550.6534 51149204.9032 7095.9170 0.6127 0.5839 0.4537 0.0230
huber Huber Regressor 3769.3076 53638697.2337 7254.7108 0.6095 0.4528 0.2187 0.0250
par Passive Aggressive Regressor 4144.7180 62949698.1775 7862.7604 0.5433 0.4634 0.2465 0.0210
en Elastic Net 7248.9376 89841235.9517 9405.5846 0.3534 0.7346 0.9238 0.0210
omp Orthogonal Matching Pursuit 8916.1927 130904492.3067 11356.4120 0.0561 0.8781 1.1598 0.0180
knn K Neighbors Regressor 8161.8875 137982796.8000 11676.3735 -0.0011 0.8744 0.9742 0.0250
dummy Dummy Regressor 8892.4478 141597492.8000 11823.4271 -0.0221 0.9868 1.4909 0.0210
print(best)
GradientBoostingRegressor(alpha=0.9, ccp_alpha=0.0, criterion='friedman_mse',
                          init=None, learning_rate=0.1, loss='squared_error',
                          max_depth=3, max_features=None, max_leaf_nodes=None,
                          min_impurity_decrease=0.0, min_samples_leaf=1,
                          min_samples_split=2, min_weight_fraction_leaf=0.0,
                          n_estimators=100, n_iter_no_change=None,
                          random_state=0, subsample=1.0, tol=0.0001,
                          validation_fraction=0.1, verbose=0, warm_start=False)

1.3 聚类

ParCaret提供了clustering模块实现无监督聚类。

数据准备

# 导入珠宝数据集
from pycaret.datasets import get_data
# 根据数据集特征进行聚类
data = get_data('./datasets/jewellery')
# data = get_data('jewellery')
Age Income SpendingScore Savings
0 58 77769 0.791329 6559.829923
1 59 81799 0.791082 5417.661426
2 62 74751 0.702657 9258.992965
3 59 74373 0.765680 7346.334504
4 87 17760 0.348778 16869.507130
# 创建数据管道
from pycaret.clustering import ClusteringExperiment
s = ClusteringExperiment()
# normalize归一化数据
s.setup(data, normalize = True, verbose = False)
# 另一种数据管道创建方式
# from pycaret.clustering import *
# s = setup(data, normalize = True)
<pycaret.clustering.oop.ClusteringExperiment at 0x200dec86340>

模型创建

PyCaret在聚类任务中提供create_model选择合适的方法来构建聚类模型,而不是全部比较。

kmeans = s.create_model('kmeans')
  Silhouette Calinski-Harabasz Davies-Bouldin Homogeneity Rand Index Completeness
0 0.7581 1611.2647 0.3743 0 0 0

create_model函数支持的聚类方法如下:

s.models()
Name Reference
ID
kmeans K-Means Clustering sklearn.cluster._kmeans.KMeans
ap Affinity Propagation sklearn.cluster._affinity_propagation.Affinity...
meanshift Mean Shift Clustering sklearn.cluster._mean_shift.MeanShift
sc Spectral Clustering sklearn.cluster._spectral.SpectralClustering
hclust Agglomerative Clustering sklearn.cluster._agglomerative.AgglomerativeCl...
dbscan Density-Based Spatial Clustering sklearn.cluster._dbscan.DBSCAN
optics OPTICS Clustering sklearn.cluster._optics.OPTICS
birch Birch Clustering sklearn.cluster._birch.Birch
print(kmeans)
# 查看聚类数
print(kmeans.n_clusters)
KMeans(algorithm='lloyd', copy_x=True, init='k-means++', max_iter=300,
       n_clusters=4, n_init='auto', random_state=1459, tol=0.0001, verbose=0)
4

数据展示

# jupyter环境下交互可视化展示
# s.evaluate_model(kmeans)
# 结果可视化 
# 'cluster' - Cluster PCA Plot (2d)
# 'tsne' - Cluster t-SNE (3d)
# 'elbow' - Elbow Plot
# 'silhouette' - Silhouette Plot
# 'distance' - Distance Plot
# 'distribution' - Distribution Plot
s.plot_model(kmeans, plot = 'elbow')

png

标签分配与数据预测

为训练数据分配聚类标签:

result = s.assign_model(kmeans)
result.head()
Age Income SpendingScore Savings Cluster
0 58 77769 0.791329 6559.830078 Cluster 2
1 59 81799 0.791082 5417.661621 Cluster 2
2 62 74751 0.702657 9258.993164 Cluster 2
3 59 74373 0.765680 7346.334473 Cluster 2
4 87 17760 0.348778 16869.507812 Cluster 1

为新的数据进行标签分配:

predictions = s.predict_model(kmeans, data = data)
predictions.head()
Age Income SpendingScore Savings Cluster
0 -0.042287 0.062733 1.103593 -1.072467 Cluster 2
1 -0.000821 0.174811 1.102641 -1.303473 Cluster 2
2 0.123577 -0.021200 0.761727 -0.526556 Cluster 2
3 -0.000821 -0.031712 1.004705 -0.913395 Cluster 2
4 1.160228 -1.606165 -0.602619 1.012686 Cluster 1

1.4 异常检测

PyCaret的anomaly detection模块是一个无监督的机器学习模块,用于识别与大多数数据存在显著差异的罕见项目、事件或观测值。通常,这些异常项目会转化为某种问题,如银行欺诈、结构缺陷、医疗问题或错误。anomaly detection模块的使用类似于cluster模块。

数据准备

from pycaret.datasets import get_data
data = get_data('./datasets/anomaly')
# data = get_data('anomaly')
Col1 Col2 Col3 Col4 Col5 Col6 Col7 Col8 Col9 Col10
0 0.263995 0.764929 0.138424 0.935242 0.605867 0.518790 0.912225 0.608234 0.723782 0.733591
1 0.546092 0.653975 0.065575 0.227772 0.845269 0.837066 0.272379 0.331679 0.429297 0.367422
2 0.336714 0.538842 0.192801 0.553563 0.074515 0.332993 0.365792 0.861309 0.899017 0.088600
3 0.092108 0.995017 0.014465 0.176371 0.241530 0.514724 0.562208 0.158963 0.073715 0.208463
4 0.325261 0.805968 0.957033 0.331665 0.307923 0.355315 0.501899 0.558449 0.885169 0.182754
from pycaret.anomaly import AnomalyExperiment
s = AnomalyExperiment()
s.setup(data, session_id = 0)
# 另一种加载方式
# from pycaret.anomaly import *
# s = setup(data, session_id = 0)
  Description Value
0 Session id 0
1 Original data shape (1000, 10)
2 Transformed data shape (1000, 10)
3 Numeric features 10
4 Preprocess True
5 Imputation type simple
6 Numeric imputation mean
7 Categorical imputation mode
8 CPU Jobs -1
9 Use GPU False
10 Log Experiment False
11 Experiment Name anomaly-default-name
12 USI 54db
<pycaret.anomaly.oop.AnomalyExperiment at 0x200e14f5250>

模型创建

iforest = s.create_model('iforest')
print(iforest)
IForest(behaviour='new', bootstrap=False, contamination=0.05,
    max_features=1.0, max_samples='auto', n_estimators=100, n_jobs=-1,
    random_state=0, verbose=0)

anomaly detection模块所支持的模型列表如下:

s.models()
Name Reference
ID
abod Angle-base Outlier Detection pyod.models.abod.ABOD
cluster Clustering-Based Local Outlier pycaret.internal.patches.pyod.CBLOFForceToDouble
cof Connectivity-Based Local Outlier pyod.models.cof.COF
iforest Isolation Forest pyod.models.iforest.IForest
histogram Histogram-based Outlier Detection pyod.models.hbos.HBOS
knn K-Nearest Neighbors Detector pyod.models.knn.KNN
lof Local Outlier Factor pyod.models.lof.LOF
svm One-class SVM detector pyod.models.ocsvm.OCSVM
pca Principal Component Analysis pyod.models.pca.PCA
mcd Minimum Covariance Determinant pyod.models.mcd.MCD
sod Subspace Outlier Detection pyod.models.sod.SOD
sos Stochastic Outlier Selection pyod.models.sos.SOS

标签分配与数据预测

为训练数据分配聚类标签:

result = s.assign_model(iforest)
result.head()
Col1 Col2 Col3 Col4 Col5 Col6 Col7 Col8 Col9 Col10 Anomaly Anomaly_Score
0 0.263995 0.764929 0.138424 0.935242 0.605867 0.518790 0.912225 0.608234 0.723782 0.733591 0 -0.016205
1 0.546092 0.653975 0.065575 0.227772 0.845269 0.837066 0.272379 0.331679 0.429297 0.367422 0 -0.068052
2 0.336714 0.538842 0.192801 0.553563 0.074515 0.332993 0.365792 0.861309 0.899017 0.088600 1 0.009221
3 0.092108 0.995017 0.014465 0.176371 0.241530 0.514724 0.562208 0.158963 0.073715 0.208463 1 0.056690
4 0.325261 0.805968 0.957033 0.331665 0.307923 0.355315 0.501899 0.558449 0.885169 0.182754 0 -0.012945

为新的数据进行标签分配:

predictions = s.predict_model(iforest, data = data)
predictions.head()
Col1 Col2 Col3 Col4 Col5 Col6 Col7 Col8 Col9 Col10 Anomaly Anomaly_Score
0 0.263995 0.764929 0.138424 0.935242 0.605867 0.518790 0.912225 0.608234 0.723782 0.733591 0 -0.016205
1 0.546092 0.653975 0.065575 0.227772 0.845269 0.837066 0.272379 0.331679 0.429297 0.367422 0 -0.068052
2 0.336714 0.538842 0.192801 0.553563 0.074515 0.332993 0.365792 0.861309 0.899017 0.088600 1 0.009221
3 0.092108 0.995017 0.014465 0.176371 0.241530 0.514724 0.562208 0.158963 0.073715 0.208463 1 0.056690
4 0.325261 0.805968 0.957033 0.331665 0.307923 0.355315 0.501899 0.558449 0.885169 0.182754 0 -0.012945

1.5 时序预测

PyCaret时间序列预测Time Series模块支持多种预测方法,如ARIMA、Prophet和LSTM。它还提供了各种功能来处理缺失值、时间序列分解和数据可视化。

数据准备

# 乘客时序数据
from pycaret.datasets import get_data
# 下载路径:https://raw.githubusercontent.com/sktime/sktime/main/sktime/datasets/data/Airline/Airline.csv
data = get_data('./datasets/airline')
# data = get_data('airline')
Date Passengers
0 1949-01 112
1 1949-02 118
2 1949-03 132
3 1949-04 129
4 1949-05 121
import pandas as pd
data['Date'] = pd.to_datetime(data['Date'])
# 并将Date设置为列号
data.set_index('Date', inplace=True)
from pycaret.time_series import TSForecastingExperiment
s = TSForecastingExperiment()
# fh: 用于预测的预测范围。默认值设置为1,即预测前方一点。,fold: 交叉验证中折数
s.setup(data, fh = 3, fold = 5, session_id = 0, verbose = False)
# from pycaret.time_series import *
# s = setup(data, fh = 3, fold = 5, session_id = 0)
<pycaret.time_series.forecasting.oop.TSForecastingExperiment at 0x200dee26910>

模型训练与评估

best = s.compare_models()
  Model MASE RMSSE MAE RMSE MAPE SMAPE R2 TT (Sec)
stlf STLF 0.4240 0.4429 12.8002 15.1933 0.0266 0.0268 0.4296 0.0300
exp_smooth Exponential Smoothing 0.5063 0.5378 15.2900 18.4455 0.0334 0.0335 -0.0521 0.0500
ets ETS 0.5520 0.5801 16.6164 19.8391 0.0354 0.0357 -0.0740 0.0680
arima ARIMA 0.6480 0.6501 19.5728 22.3027 0.0412 0.0420 -0.0796 0.0420
auto_arima Auto ARIMA 0.6526 0.6300 19.7405 21.6202 0.0414 0.0421 -0.0560 10.5220
theta Theta Forecaster 0.8458 0.8223 25.7024 28.3332 0.0524 0.0541 -0.7710 0.0220
huber_cds_dt Huber w/ Cond. Deseasonalize & Detrending 0.9002 0.8900 27.2568 30.5782 0.0550 0.0572 -0.0309 0.0680
knn_cds_dt K Neighbors w/ Cond. Deseasonalize & Detrending 0.9381 0.8830 28.5678 30.5007 0.0555 0.0575 0.0908 0.0920
lr_cds_dt Linear w/ Cond. Deseasonalize & Detrending 0.9469 0.9297 28.6337 31.9163 0.0581 0.0605 -0.1620 0.0820
ridge_cds_dt Ridge w/ Cond. Deseasonalize & Detrending 0.9469 0.9297 28.6340 31.9164 0.0581 0.0605 -0.1620 0.0680
en_cds_dt Elastic Net w/ Cond. Deseasonalize & Detrending 0.9499 0.9320 28.7271 31.9952 0.0582 0.0606 -0.1579 0.0700
llar_cds_dt Lasso Least Angular Regressor w/ Cond. Deseasonalize & Detrending 0.9520 0.9336 28.7917 32.0528 0.0583 0.0607 -0.1559 0.0560
lasso_cds_dt Lasso w/ Cond. Deseasonalize & Detrending 0.9521 0.9337 28.7941 32.0557 0.0583 0.0607 -0.1560 0.0720
br_cds_dt Bayesian Ridge w/ Cond. Deseasonalize & Detrending 0.9551 0.9347 28.9018 32.1013 0.0582 0.0606 -0.1377 0.0580
et_cds_dt Extra Trees w/ Cond. Deseasonalize & Detrending 1.0322 0.9942 31.4048 34.3054 0.0607 0.0633 -0.1660 0.1280
rf_cds_dt Random Forest w/ Cond. Deseasonalize & Detrending 1.0851 1.0286 32.9791 35.4666 0.0641 0.0670 -0.3545 0.1400
lightgbm_cds_dt Light Gradient Boosting w/ Cond. Deseasonalize & Detrending 1.1409 1.1040 34.5999 37.9918 0.0670 0.0701 -0.3994 0.0900
ada_cds_dt AdaBoost w/ Cond. Deseasonalize & Detrending 1.1441 1.0843 34.7451 37.3681 0.0664 0.0697 -0.3004 0.0920
gbr_cds_dt Gradient Boosting w/ Cond. Deseasonalize & Detrending 1.1697 1.1094 35.4408 38.1373 0.0697 0.0729 -0.4163 0.0900
omp_cds_dt Orthogonal Matching Pursuit w/ Cond. Deseasonalize & Detrending 1.1793 1.1250 35.7348 38.6755 0.0706 0.0732 -0.5095 0.0620
dt_cds_dt Decision Tree w/ Cond. Deseasonalize & Detrending 1.2704 1.2371 38.4976 42.4846 0.0773 0.0814 -1.0382 0.0860
snaive Seasonal Naive Forecaster 1.7700 1.5999 53.5333 54.9143 0.1136 0.1211 -4.1630 0.1580
naive Naive Forecaster 1.8145 1.7444 54.8667 59.8160 0.1135 0.1151 -3.7710 0.1460
polytrend Polynomial Trend Forecaster 2.3154 2.2507 70.1138 77.3400 0.1363 0.1468 -4.6202 0.1080
croston Croston 2.6211 2.4985 79.3645 85.8439 0.1515 0.1684 -5.2294 0.0140
grand_means Grand Means Forecaster 7.1261 6.3506 216.0214 218.4259 0.4377 0.5682 -59.2684 0.1400

数据展示

# jupyter环境下交互可视化展示
# plot参数支持:
# - 'ts' - Time Series Plot
# - 'train_test_split' - Train Test Split
# - 'cv' - Cross Validation
# - 'acf' - Auto Correlation (ACF)
# - 'pacf' - Partial Auto Correlation (PACF)
# - 'decomp' - Classical Decomposition
# - 'decomp_stl' - STL Decomposition
# - 'diagnostics' - Diagnostics Plot
# - 'diff' - Difference Plot
# - 'periodogram' - Frequency Components (Periodogram)
# - 'fft' - Frequency Components (FFT)
# - 'ccf' - Cross Correlation (CCF)
# - 'forecast' - "Out-of-Sample" Forecast Plot
# - 'insample' - "In-Sample" Forecast Plot
# - 'residuals' - Residuals Plot
# s.plot_model(best, plot = 'forecast', data_kwargs = {'fh' : 24})

数据预测

# 使模型拟合包括测试样本在内的完整数据集
final_best = s.finalize_model(best)
s.predict_model(best, fh = 24)
s.save_model(final_best, 'final_best_model')
Transformation Pipeline and Model Successfully Saved





(ForecastingPipeline(steps=[('forecaster',
                             TransformedTargetForecaster(steps=[('model',
                                                                 ForecastingPipeline(steps=[('forecaster',
                                                                                             TransformedTargetForecaster(steps=[('model',
                                                                                                                                 STLForecaster(sp=12))]))]))]))]),
 'final_best_model.pkl')

2 数据处理与清洗

2.1 缺失值处理

各种数据集可能由于多种原因存在缺失值或空记录。移除具有缺失值的样本是一种常见策略,但这会导致丢失可能有价值的数据。一种可替代的策略是对缺失值进行插值填充。在setup函数中可以指定如下参数,实现缺失值处理:

  • imputation_type:取值可以是 'simple' 或 'iterative'或 None。当imputation_type设置为 'simple' 时,PyCaret 将使用简单的方式(numeric_imputation和categorical_imputation)对缺失值进行填充。而当设置为 'iterative' 时,则会使用模型估计的方式(numeric_iterative_imputer,categorical_iterative_imputer)进行填充处理。如果设置为 None,则不会执行任何缺失值填充操作
  • numeric_imputation: 设置数值类型的缺失值,方式如下:
    • mean: 用列的平均值填充,默认
    • drop: 删除包含缺失值的行
    • median: 用列的中值填充
    • mode: 用列最常见值填充
    • knn: 使用K-最近邻方法拟合
    • int or float: 用指定值替代
  • categorical_imputation:
    • mode: 用列最常见值填充,默认
    • drop: 删除包含缺失值的行
    • str: 用指定字符替代
  • numeric_iterative_imputer: 使用估计模型拟合值,可输入str或sklearn模型, 默认使用lightgbm
  • categorical_iterative_imputer: 使用估计模型差值,可输入str或sklearn模型, 默认使用lightgbm

加载数据

# load dataset
from pycaret.datasets import get_data
# 从本地加载数据,注意dataset是数据的文件名
data = get_data(dataset='./datasets/hepatitis', verbose=False)
# data = get_data('hepatitis',verbose=False)
# 可以看到第三行STEROID列出现NaN值
data.head()
Class AGE SEX STEROID ANTIVIRALS FATIGUE MALAISE ANOREXIA LIVER BIG LIVER FIRM SPLEEN PALPABLE SPIDERS ASCITES VARICES BILIRUBIN ALK PHOSPHATE SGOT ALBUMIN PROTIME HISTOLOGY
0 0 30 2 1.0 2 2 2 2 1.0 2.0 2.0 2.0 2.0 2.0 1.0 85.0 18.0 4.0 NaN 1
1 0 50 1 1.0 2 1 2 2 1.0 2.0 2.0 2.0 2.0 2.0 0.9 135.0 42.0 3.5 NaN 1
2 0 78 1 2.0 2 1 2 2 2.0 2.0 2.0 2.0 2.0 2.0 0.7 96.0 32.0 4.0 NaN 1
3 0 31 1 NaN 1 2 2 2 2.0 2.0 2.0 2.0 2.0 2.0 0.7 46.0 52.0 4.0 80.0 1
4 0 34 1 2.0 2 2 2 2 2.0 2.0 2.0 2.0 2.0 2.0 1.0 NaN 200.0 4.0 NaN 1
# 使用均值填充数据
from pycaret.classification import ClassificationExperiment
s = ClassificationExperiment()
# 均值
# s.data['STEROID'].mean()
s.setup(data = data, session_id=0, target = 'Class',verbose=False, 
        # 设置data_split_shuffle和data_split_stratify为False不打乱数据
        data_split_shuffle = False, data_split_stratify = False,
        imputation_type='simple', numeric_iterative_imputer='drop')
# 查看转换后的数据
s.get_config('dataset_transformed').head()
AGE SEX STEROID ANTIVIRALS FATIGUE MALAISE ANOREXIA LIVER BIG LIVER FIRM SPLEEN PALPABLE SPIDERS ASCITES VARICES BILIRUBIN ALK PHOSPHATE SGOT ALBUMIN PROTIME HISTOLOGY Class
0 30.0 2.0 1.000000 2.0 2.0 2.0 2.0 1.0 2.0 2.0 2.0 2.0 2.0 1.0 85.000000 18.0 4.0 66.53968 1.0 0
1 50.0 1.0 1.000000 2.0 1.0 2.0 2.0 1.0 2.0 2.0 2.0 2.0 2.0 0.9 135.000000 42.0 3.5 66.53968 1.0 0
2 78.0 1.0 2.000000 2.0 1.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 0.7 96.000000 32.0 4.0 66.53968 1.0 0
3 31.0 1.0 1.509434 1.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 0.7 46.000000 52.0 4.0 80.00000 1.0 0
4 34.0 1.0 2.000000 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 1.0 99.659088 200.0 4.0 66.53968 1.0 0
# 使用knn拟合数据
from pycaret.classification import ClassificationExperiment
s = ClassificationExperiment()
s.setup(data = data, session_id=0, target = 'Class',verbose=False, 
        # 设置data_split_shuffle和data_split_stratify为False不打乱数据
        data_split_shuffle = False, data_split_stratify = False,
        imputation_type='simple', numeric_imputation = 'knn')
# 查看转换后的数据
s.get_config('dataset_transformed').head()
AGE SEX STEROID ANTIVIRALS FATIGUE MALAISE ANOREXIA LIVER BIG LIVER FIRM SPLEEN PALPABLE SPIDERS ASCITES VARICES BILIRUBIN ALK PHOSPHATE SGOT ALBUMIN PROTIME HISTOLOGY Class
0 30.0 2.0 1.0 2.0 2.0 2.0 2.0 1.0 2.0 2.0 2.0 2.0 2.0 1.0 85.000000 18.0 4.0 91.800003 1.0 0
1 50.0 1.0 1.0 2.0 1.0 2.0 2.0 1.0 2.0 2.0 2.0 2.0 2.0 0.9 135.000000 42.0 3.5 61.599998 1.0 0
2 78.0 1.0 2.0 2.0 1.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 0.7 96.000000 32.0 4.0 75.800003 1.0 0
3 31.0 1.0 1.8 1.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 0.7 46.000000 52.0 4.0 80.000000 1.0 0
4 34.0 1.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 1.0 108.400002 200.0 4.0 62.799999 1.0 0
# 使用lightgbmn拟合数据
# from pycaret.classification import ClassificationExperiment
# s = ClassificationExperiment()
# s.setup(data = data, session_id=0, target = 'Class',verbose=False, 
#         # 设置data_split_shuffle和data_split_stratify为False不打乱数据
#         data_split_shuffle = False, data_split_stratify = False,
#         imputation_type='iterative', numeric_iterative_imputer = 'lightgbm')
# 查看转换后的数据
# s.get_config('dataset_transformed').head()

2.2 类型转换

虽然 PyCaret具有自动识别特征类型的功能,但PyCaret提供了数据类型自定义参数,用户可以对数据集进行更精细的控制和指导,以确保模型训练和特征工程的效果更加符合用户的预期和需求。这些自定义参数如下:

  • numeric_features:用于指定数据集中的数值特征列的参数。这些特征将被视为连续型变量进行处理
  • categorical_features:用于指定数据集中的分类特征列的参数。这些特征将被视为离散型变量进行处理
  • date_features:用于指定数据集中的日期特征列的参数。这些特征将被视为日期型变量进行处理
  • create_date_columns:用于指定是否从日期特征中创建新的日期相关列的参数
  • text_features:用于指定数据集中的文本特征列的参数。这些特征将被视为文本型变量进行处理
  • text_features_method:用于指定对文本特征进行处理的方法的参数
  • ignore_features:用于指定在建模过程中需要忽略的特征列的参数
  • keep_features:用于指定在建模过程中需要保留的特征列的参数
# 转换变量类型
from pycaret.datasets import get_data
data = get_data(dataset='./datasets/hepatitis', verbose=False)

from pycaret.classification import *
s = setup(data = data, target = 'Class', ignore_features  = ['SEX','AGE'], categorical_features=['STEROID'],verbose = False,
         data_split_shuffle = False, data_split_stratify = False)
# 查看转换后的数据,前两列消失,STEROID变为分类变量
s.get_config('dataset_transformed').head()
STEROID ANTIVIRALS FATIGUE MALAISE ANOREXIA LIVER BIG LIVER FIRM SPLEEN PALPABLE SPIDERS ASCITES VARICES BILIRUBIN ALK PHOSPHATE SGOT ALBUMIN PROTIME HISTOLOGY Class
0 0.0 2.0 2.0 2.0 2.0 1.0 2.0 2.0 2.0 2.0 2.0 1.0 85.000000 18.0 4.0 66.53968 1.0 0
1 0.0 2.0 1.0 2.0 2.0 1.0 2.0 2.0 2.0 2.0 2.0 0.9 135.000000 42.0 3.5 66.53968 1.0 0
2 1.0 2.0 1.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 0.7 96.000000 32.0 4.0 66.53968 1.0 0
3 1.0 1.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 0.7 46.000000 52.0 4.0 80.00000 1.0 0
4 1.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 1.0 99.659088 200.0 4.0 66.53968 1.0 0

2.3 独热编码

当数据集中包含分类变量时,这些变量通常需要转换为模型可以理解的数值形式。独热编码是一种常用的方法,它将每个分类变量转换为一组二进制变量,其中每个变量对应一个可能的分类值,并且只有一个变量在任何给定时间点上取值为 1,其余变量均为 0。可以通过传递参数categorical_features来指定要进行独热编码的列。例如:

# load dataset
from pycaret.datasets import get_data
data = get_data(dataset='./datasets/pokemon', verbose=False)
# data = get_data('pokemon')
data.head()
# Name Type 1 Type 2 Total HP Attack Defense Sp. Atk Sp. Def Speed Generation Legendary
0 1 Bulbasaur Grass Poison 318 45 49 49 65 65 45 1 False
1 2 Ivysaur Grass Poison 405 60 62 63 80 80 60 1 False
2 3 Venusaur Grass Poison 525 80 82 83 100 100 80 1 False
3 3 VenusaurMega Venusaur Grass Poison 625 80 100 123 122 120 80 1 False
4 4 Charmander Fire NaN 309 39 52 43 60 50 65 1 False
# 对Type 1实现独热编码
len(set(data['Type 1']))
18
from pycaret.classification import *
s = setup(data = data, categorical_features =["Type 1"],target = 'Legendary', verbose=False)
# 查看转换后的数据Type 1变为独热编码
s.get_config('dataset_transformed').head()
# Name Type 1_Grass Type 1_Ghost Type 1_Water Type 1_Steel Type 1_Psychic Type 1_Fire Type 1_Poison Type 1_Fairy ... Type 2 Total HP Attack Defense Sp. Atk Sp. Def Speed Generation Legendary
202 187.0 Hoppip 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... Flying 250.0 35.0 35.0 40.0 35.0 55.0 50.0 2.0 False
477 429.0 Mismagius 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 ... NaN 495.0 60.0 60.0 60.0 105.0 105.0 105.0 4.0 False
349 319.0 SharpedoMega Sharpedo 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 ... Dark 560.0 70.0 140.0 70.0 110.0 65.0 105.0 3.0 False
777 707.0 Klefki 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 ... Fairy 470.0 57.0 80.0 91.0 80.0 87.0 75.0 6.0 False
50 45.0 Vileplume 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... Poison 490.0 75.0 80.0 85.0 110.0 90.0 50.0 1.0 False

5 rows × 30 columns

2.4 数据平衡

在 PyCaret 中,fix_imbalance 和 fix_imbalance_method 是用于处理不平衡数据集的两个参数。这些参数通常用于在训练模型之前对数据集进行预处理,以解决类别不平衡问题。

  • fix_imbalance 参数:这是一个布尔值参数,用于指示是否对不平衡数据集进行处理。当设置为 True 时,PyCaret 将自动检测数据集中的类别不平衡问题,并尝试通过采样方法来解决。当设置为 False 时,PyCaret 将使用原始的不平衡数据集进行模型训练
  • fix_imbalance_method 参数:这是一个字符串参数,用于指定处理不平衡数据集的方法。可选的值包括:
    • 使用 SMOTE(Synthetic Minority Over-sampling Technique)来生成人工合成样本,从而平衡类别(默认参数smote)
    • 使用imbalanced-learn提供的估算模型
# 加载数据
from pycaret.datasets import get_data
data = get_data(dataset='./datasets/credit', verbose=False)
# data = get_data('credit')
data.head()
LIMIT_BAL SEX EDUCATION MARRIAGE AGE PAY_1 PAY_2 PAY_3 PAY_4 PAY_5 ... BILL_AMT4 BILL_AMT5 BILL_AMT6 PAY_AMT1 PAY_AMT2 PAY_AMT3 PAY_AMT4 PAY_AMT5 PAY_AMT6 default
0 20000 2 2 1 24 2 2 -1 -1 -2 ... 0.0 0.0 0.0 0.0 689.0 0.0 0.0 0.0 0.0 1
1 90000 2 2 2 34 0 0 0 0 0 ... 14331.0 14948.0 15549.0 1518.0 1500.0 1000.0 1000.0 1000.0 5000.0 0
2 50000 2 2 1 37 0 0 0 0 0 ... 28314.0 28959.0 29547.0 2000.0 2019.0 1200.0 1100.0 1069.0 1000.0 0
3 50000 1 2 1 57 -1 0 -1 0 0 ... 20940.0 19146.0 19131.0 2000.0 36681.0 10000.0 9000.0 689.0 679.0 0
4 50000 1 1 2 37 0 0 0 0 0 ... 19394.0 19619.0 20024.0 2500.0 1815.0 657.0 1000.0 1000.0 800.0 0

5 rows × 24 columns

# 查看数据各类别数
category_counts = data['default'].value_counts()
category_counts
default
0    18694
1     5306
Name: count, dtype: int64
from pycaret.classification import *
s = setup(data = data, target = 'default', fix_imbalance = True, verbose = False)
# 可以看到类1数据量变多了
s.get_config('dataset_transformed')['default'].value_counts()
default
0    18694
1    14678
Name: count, dtype: int64

2.5 异常值处理

PyCaret的remove_outliers函数可以在训练模型之前识别和删除数据集中的异常值。它使用奇异值分解技术进行PCA线性降维来识别异常值,并可以通过setup中的outliers_threshold参数控制异常值的比例(默认0.05)。

from pycaret.datasets import get_data

data = get_data(dataset='./datasets/insurance', verbose=False)
# insurance = get_data('insurance')
# 数据维度
data.shape
(1338, 7)
from pycaret.regression import *
s = setup(data = data, target = 'charges', remove_outliers = True ,verbose = False, outliers_threshold = 0.02)
# 移除异常数据后,数据量变少
s.get_config('dataset_transformed').shape
(1319, 10)

2.6 特征重要性

特征重要性是一种用于选择数据集中对预测目标变量最有贡献的特征的过程。与使用所有特征相比,仅使用选定的特征可以减少过拟合的风险,提高准确性,并缩短训练时间。在PyCaret中,可以通过使用feature_selection参数来实现这一目的。对于PyCaret中几个与特征选择相关参数的解释如下:

  • feature_selection:用于指定是否在模型训练过程中进行特征选择。可以设置为 True 或 False。
  • feature_selection_method:特征选择方法:
    • 'univariate': 使用sklearn的SelectKBest,基于统计测试来选择与目标变量最相关的特征。
    • 'classic(默认)': 使用sklearn的SelectFromModel,利用监督学习模型的特征重要性或系数来选择最重要的特征。
    • 'sequential': 使用sklearn的SequentialFeatureSelector,该类根据指定的算法(如前向选择、后向选择等)以及性能指标(如交叉验证得分)逐步选择特征。
  • n_features_to_select:特征选择的最大特征数量或比例。如果<1,则为起始特征的比例。默认为0.2。该参数在计数时不考虑 ignore_features 和 keep_features 中的特征。
from pycaret.datasets import get_data
data = get_data('./datasets/diabetes')
Number of times pregnant Plasma glucose concentration a 2 hours in an oral glucose tolerance test Diastolic blood pressure (mm Hg) Triceps skin fold thickness (mm) 2-Hour serum insulin (mu U/ml) Body mass index (weight in kg/(height in m)^2) Diabetes pedigree function Age (years) Class variable
0 6 148 72 35 0 33.6 0.627 50 1
1 1 85 66 29 0 26.6 0.351 31 0
2 8 183 64 0 0 23.3 0.672 32 1
3 1 89 66 23 94 28.1 0.167 21 0
4 0 137 40 35 168 43.1 2.288 33 1
from pycaret.regression import *
# feature_selection选择特征, n_features_to_select选择特征比例
s = setup(data = data, target = 'Class variable', feature_selection = True, feature_selection_method = 'univariate',
          n_features_to_select = 0.3, verbose = False)
# 查看哪些特征保留下来
s.get_config('X_transformed').columns
s.get_config('X_transformed').head()
Plasma glucose concentration a 2 hours in an oral glucose tolerance test Body mass index (weight in kg/(height in m)^2)
56 187.0 37.700001
541 128.0 32.400002
269 146.0 27.500000
304 150.0 21.000000
32 88.0 24.799999

2.7 归一化

数据归一化

在 PyCaret 中,normalize 和 normalize_method 参数用于数据预处理中的特征缩放操作。特征缩放是指将数据的特征值按比例缩放,使之落入一个小的特定范围,这样可以消除特征之间的量纲影响,使模型训练更加稳定和准确。下面是关于这两个参数的说明:

  • normalize: 这是一个布尔值参数,用于指定是否对特征进行缩放。默认情况下,它的取值为 False,表示不进行特征缩放。如果将其设置为 True,则会启用特征缩放功能。
  • normalize_method: 这是一个字符串参数,用于指定特征缩放的方法。可选的值有:
    • zscore(默认): 使用 Z 分数标准化方法,也称为标准化或 Z 标准化。该方法将特征的值转换为其 Z 分数,即将特征值减去其均值,然后除以其标准差,从而使得特征的均值为 0,标准差为 1。
    • minmax: 使用 Min-Max 标准化方法,也称为归一化。该方法将特征的值线性转换到指定的最小值和最大值之间,默认情况下是 [0, 1] 范围。
    • maxabs: 使用 MaxAbs 标准化方法。该方法将特征的值除以特征的最大绝对值,将特征的值缩放到 [-1, 1] 范围内。
    • robust: 使用 RobustScaler 标准化方法。该方法对数据的每个特征进行中心化和缩放,使用特征的中位数和四分位数范围来缩放特征。
from pycaret.datasets import get_data
data = get_data('./datasets/pokemon')
data.head()
# Name Type 1 Type 2 Total HP Attack Defense Sp. Atk Sp. Def Speed Generation Legendary
0 1 Bulbasaur Grass Poison 318 45 49 49 65 65 45 1 False
1 2 Ivysaur Grass Poison 405 60 62 63 80 80 60 1 False
2 3 Venusaur Grass Poison 525 80 82 83 100 100 80 1 False
3 3 VenusaurMega Venusaur Grass Poison 625 80 100 123 122 120 80 1 False
4 4 Charmander Fire NaN 309 39 52 43 60 50 65 1 False
# Name Type 1 Type 2 Total HP Attack Defense Sp. Atk Sp. Def Speed Generation Legendary
0 1 Bulbasaur Grass Poison 318 45 49 49 65 65 45 1 False
1 2 Ivysaur Grass Poison 405 60 62 63 80 80 60 1 False
2 3 Venusaur Grass Poison 525 80 82 83 100 100 80 1 False
3 3 VenusaurMega Venusaur Grass Poison 625 80 100 123 122 120 80 1 False
4 4 Charmander Fire NaN 309 39 52 43 60 50 65 1 False
# 归一化
from pycaret.classification import *
s = setup(data, target='Legendary', normalize=True, normalize_method='robust', verbose=False)

数据归一化结果:

s.get_config('X_transformed').head()
# Name Type 1_Water Type 1_Normal Type 1_Ice Type 1_Psychic Type 1_Fire Type 1_Rock Type 1_Fighting Type 1_Grass ... Type 2_Electric Type 2_Normal Total HP Attack Defense Sp. Atk Sp. Def Speed Generation
403 -0.021629 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0 0.195387 -0.333333 0.200000 0.875 1.088889 0.125 -0.288889 0.000000
471 0.139870 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0 0.179104 0.333333 0.555556 -0.100 -0.111111 -0.100 1.111111 0.333333
238 -0.448450 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0 -1.080054 -0.500000 -0.555556 -0.750 -0.777778 -1.000 -0.333333 -0.333333
646 0.604182 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 ... 0.0 0.0 -0.618725 -0.166667 -0.333333 -0.500 -0.555556 -0.500 0.222222 0.666667
69 -0.898342 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 ... 0.0 0.0 -0.265943 -0.833333 -0.888889 -1.000 1.222222 0.000 0.888889 -0.666667

5 rows × 46 columns

特征变换

归一化会重新调整数据,使其在新的范围内,以减少方差中幅度的影响。特征变换是一种更彻底的技术。通过转换改变数据的分布形状,使得转换后的数据可以被表示为正态分布或近似正态分布。PyCaret中通过transformation参数开启特征转换,transformation_method设置转换方法:yeo-johnson(默认)和分位数。此外除了特征变换,还有目标变换。目标变换它将改变目标变量而不是特征的分布形状。此功能仅在pycarte.regression模块中可用。使用transform_target开启目标变换,transformation_method设置转换方法。

from pycaret.classification import *
s = setup(data = data, target = 'Legendary', transformation = True, verbose = False)
# 特征变换结果
s.get_config('X_transformed').head()
# Name Type 1_Psychic Type 1_Water Type 1_Rock Type 1_Grass Type 1_Dragon Type 1_Ghost Type 1_Bug Type 1_Fairy ... Type 2_Electric Type 2_Bug Total HP Attack Defense Sp. Atk Sp. Def Speed Generation
165 52.899003 0.009216 0.043322 -0.000000 -0.000000 -0.000000 -0.000000 -0.0 -0.0 -0.0 ... -0.0 -0.0 93.118403 12.336844 23.649090 13.573010 10.692443 8.081703 26.134255 0.900773
625 140.730289 0.009216 -0.000000 0.095739 -0.000000 -0.000000 -0.000000 -0.0 -0.0 -0.0 ... -0.0 -0.0 66.091344 9.286671 20.259153 13.764668 8.160482 6.056644 9.552506 3.679456
628 141.283084 0.009216 -0.000000 -0.000000 0.043322 -0.000000 -0.000000 -0.0 -0.0 -0.0 ... -0.0 -0.0 89.747939 10.823299 29.105379 11.029571 11.203335 6.942091 27.793080 3.679456
606 137.396878 0.009216 -0.000000 -0.000000 -0.000000 0.061897 -0.000000 -0.0 -0.0 -0.0 ... -0.0 -0.0 56.560577 8.043018 10.276208 10.604937 6.949265 6.302465 19.943809 3.679456
672 149.303914 0.009216 -0.000000 -0.000000 -0.000000 -0.000000 0.029706 -0.0 -0.0 -0.0 ... -0.0 -0.0 72.626190 10.202245 26.061259 11.435493 7.199607 6.302465 20.141156 3.679456

5 rows × 46 columns

3 参考

与[机器学习] 低代码机器学习工具PyCaret库使用指北相似的内容:

[机器学习] 低代码机器学习工具PyCaret库使用指北

PyCaret是一个开源、低代码Python机器学习库,能够自动化机器学习工作流程。它是一个端到端的机器学习和模型管理工具,极大地加快了实验周期,提高了工作效率。PyCaret本质上是围绕几个机器学习库和框架(如scikit-learn、XGBoost、LightGBM、CatBoost、spaCy

牛逼!50.3K Star!一个自动将屏幕截图转换为代码的开源工具

1、背景 在当今快节奏的软件开发环境中,设计师与开发者之间的协同工作显得尤为重要。然而,理解并准确实现设计稿的意图常常需要耗费大量的时间和沟通成本。为此,开源社区中出现了一个引人注目的项目——screenshot-to-code,它利用AI人工智能技术(机器学习算法和视觉分析技术),将屏幕截图自动转

给大家分享一套非常棒的python机器学习课程

给大家分享一套非常棒的python机器学习课程——《AI小天才:让小学生轻松掌握机器学习》,2024年5月完结新课,提供配套的代码+笔记+软件包下载!学完本课程,可以轻松掌握机器学习的全面应用,复杂特征工程,数据回归,分类,算法的项目实战应用,以小学生的视角和知识储备即可学会。课程名字:AI小天才:

如何查看服务器的Raid缓存等配置的情况

摘要 最近总遇到同一批机器的IO不一样的情况. 感觉可能跟硬件设备和Raid卡的设置不一样有关系. 所以今天学习研究了下storcli的命令. 希望能够进行一些数据的收集. Storcli简介 storcli 是LSI公司官方提供的Raid卡管理工具, storcli已经基本代替了megacli,是

机器学习即代码的时代已经到来

译者注: 到底是 AI 会吃掉软件还是软件会吃掉 AI?为了 job security 工程师应该把宝押在哪儿?这篇 2021 年的文章提供的一些视角似乎印证了它现在的流行,有点“运筹于帷幄之中,决胜于数年之后”的意思,颇值得软件架构师和产品经理们内省一番。 2021 版的 《人工智能现状报告》 于

29.4K star! 仅需几行代码快速构建机器学习 Web 应用项目,无需前端技能!

大家好,我是狂师! 今天给大家推荐一款开源的Python库:Gradio! Gradio是一个开源的Python库,用于创建机器学习和数据科学的交互式应用和演示。 项目地址: https://github.com/gradio-app/gradio 1、项目介绍 Gradio旨在简化展示和测试机器学

机器学习算法(一):1. numpy从零实现线性回归

系列文章目录 机器学习算法(一):1. numpy从零实现线性回归 机器学习算法(一):2. 线性回归之多项式回归(特征选取) @目录系列文章目录前言一、理论介绍二、代码实现1、导入库2、准备数据集3、定义预测函数(predict)4 代价(损失)函数5 计算参数梯度6 批量梯度下降7 训练8 可视

SMOTE与SMOGN算法R语言代码

本文介绍基于R语言中的UBL包,读取.csv格式的Excel表格文件,实现SMOTE算法与SMOGN算法,对机器学习、深度学习回归中,训练数据集不平衡的情况加以解决的具体方法~

Java中可以用的大数据推荐算法

在Java中实现大数据推荐算法时,通常会使用一些开源的机器学习库,如Apache Mahout、Weka、DL4J(DeepLearning4j,用于深度学习)或者Spark MLlib(用于在Spark集群上运行)。由于完整实现一个大数据推荐算法的代码量可能非常大,并且需要配合具体的数据集和环境进

Jupyter Notebook入门指南

Jupyter Notebook是一套基于web的交互式开发环境。用户可以在线开发和分享包含代码和输出的交互式文档,支持实时代码,数学方程,可视化和 markdown等。用途包括:数据清理和转换,数值模拟,统计建模,机器学习等等。