智能家居>> customer-segmentation-toolkit>> 返回
项目作者： atemate

项目描述 ：
Data transformations toolkit made from jupyter notebook: https://www.kaggle.com/fabiendaniel/customer-segmentation
高级语言： Jupyter Notebook
项目地址: git://github.com/atemate/customer-segmentation-toolkit.git
创建时间： 2021-05-11T18:41:46Z
项目社区：https://github.com/atemate/customer-segmentation-toolkit
开源协议：Apache License 2.0
下载


Customer segmentation toolkit

Data transformations toolkit made by Team #2 for the MLOps Engineering Lab #2 “Feature Store for ML”.

Installation

pip install -U customer-segmentation-toolkit

Usage

  1. import pandas as pd
  2. from pathlib import Path

01. Load and split dataset

  1. import datetime
  2. from customer_segmentation_toolkit.load_split import load_data_csv, split_by_invoice_date
  4. ONLINEOFFLINE_DATE_SPLIT = datetime.date(2011,10,1)
  6. # Loading original dataset
  7. df = load_data_csv('../data/data.csv')
  8. print(f'Loaded dataset, shape: {df.shape}')
  10. # Splitting dataset to offline and online parts
  11. df_offline, df_online = split_by_invoice_date(df, ONLINEOFFLINE_DATE_SPLIT)
  13. print(f'Offline dataset shape: {df_offline.shape}')
  14. print(f'Offline invoices: from {df_offline["InvoiceDate"].min()} to {df_offline["InvoiceDate"].max()}')
  16. print(f'Online dataset shape: {df_online.shape}')
  17. print(f'Online invoices: from {df_online["InvoiceDate"].min()} to {df_online["InvoiceDate"].max()}')
  1. Loaded dataset, shape: (541909, 8)
  2. Offline dataset shape: (370931, 8)
  3. Offline invoices: from 2010-12-01 08:26:00 to 2011-09-30 17:22:00
  4. Online dataset shape: (170978, 8)
  5. Online invoices: from 2011-10-02 10:32:00 to 2011-12-09 12:50:00
  1. # Saving processed data
  2. OUTPUT = Path(f'../data/output/01_data_split_offline_online')
  3. OUTPUT.mkdir(exist_ok=True, parents=True)
  5. df_offline.to_csv(f'{OUTPUT}/no_live_data.csv', index=False)
  6. df_online.to_csv(f'{OUTPUT}/raw_live_data.csv', index=False)
  7. Path(f'{OUTPUT}/onlineoffline_date_split.txt').write_text(str(ONLINEOFFLINE_DATE_SPLIT))
  9. print(f'Output data saved to {OUTPUT}: {[p.name for p in Path(OUTPUT).iterdir()]}')
  1. Output data saved to ../data/output/01_data_split_offline_online: ['onlineoffline_date_split.txt', 'no_live_data.csv', 'raw_live_data.csv']

02. Clean dataset rows

  1. from customer_segmentation_toolkit.load_split import load_data_csv
  2. from customer_segmentation_toolkit.clean_rows import clean_data_rows
  4. # Loading raw offline dataset'
  5. df = load_data_csv('../data/output/01_data_split_offline_online/no_live_data.csv')
  6. print(f'Loaded raw offline dataset, shape: {df.shape}')
  8. # Cleaning the dataset
  9. df_cleaned = clean_data_rows(df)
  10. print(f'Cleaned offline dataset shape: {df.shape}')
  1. Loaded raw offline dataset, shape: (370931, 8)
  2. Cleaned offline dataset shape: (370931, 8)
  1. # Saving processed data
  2. OUTPUT = Path(f'../data/output/02_data_clean_rows')
  3. OUTPUT.mkdir(exist_ok=True, parents=True)
  5. df_cleaned.to_csv(f'{OUTPUT}/no_live_data__cleaned.csv', index=False)
  7. print(f'Output data saved to {OUTPUT}: {[p.name for p in Path(OUTPUT).iterdir()]}')
  1. Output data saved to ../data/output/02_data_clean_rows: ['no_live_data__cleaned.csv']

03. Analyse purchases

  1. import datetime
  2. from customer_segmentation_toolkit.load_split import load_data_csv
  3. from customer_segmentation_toolkit.analyse_purchases import build_product_list
  5. N_PURCHASE_CLUSTERS = 5
  6. TRAINTEST_DATE_SPLIT = datetime.date(2011,8,1)
  8. # Loading cleaned dataset
  9. df_cleaned = load_data_csv('../data/output/02_data_clean_rows/no_live_data__cleaned.csv')
  10. print(f'Loaded cleaned offline dataset, shape: {df_cleaned.shape}')
  12. list_products = build_product_list(df_cleaned)
  13. print(f'Built list of products:')
  14. print(pd.DataFrame(list_products).head())
  15. print('...')
  1. Loaded cleaned offline dataset, shape: (263815, 10)
  2. Built list of products:
  3.          0    1
  4. 0    heart  251
  5. 1  vintage  195
  6. 2      set  194
  7. 3      bag  158
  8. 4      box  147
  9. ...
  1. from customer_segmentation_toolkit.analyse_purchases import build_keywords_matrix
  3. # Building keywords count matrix
  4. THRESHOLD = [0, 1, 2, 3, 5, 10]
  5. matrix = build_keywords_matrix(df_cleaned, list_products, THRESHOLD)
  6. print(f'Built keywords count matrix (shape: {matrix.shape}):')
  7. print(matrix.head())
  1. Built keywords count matrix (shape: (3662, 188)):
  2.    heart  vintage  set  bag  box  glass  christmas  design  candle  flower  \
  3. 0      1        0    0    0    0      0          0       0       0       0   
  4. 1      0        0    0    0    0      0          0       0       0       0   
  5. 2      1        0    0    0    0      0          0       0       0       0   
  6. 3      0        0    0    0    0      0          0       0       0       0   
  7. 4      1        0    0    0    0      0          0       0       0       0   
  9.    ...  medium  hen  wallet  point  0<.<1  1<.<2  2<.<3  3<.<5  5<.<10  .>10  
  10. 0  ...       0    0       0      0      0      0      1      0       0     0  
  11. 1  ...       0    0       0      0      0      0      0      1       0     0  
  12. 2  ...       0    0       0      0      0      0      0      1       0     0  
  13. 3  ...       0    0       0      0      0      0      0      1       0     0  
  14. 4  ...       0    0       0      0      0      0      0      1       0     0  
  16. [5 rows x 188 columns]
  1. from customer_segmentation_toolkit.analyse_purchases import compute_purchase_clusters
  3. # Computing purchases clusters via Kmeans
  4. clusters = compute_purchase_clusters(matrix, N_PURCHASE_CLUSTERS)
  5. print(f'Built purchase clusters:')
  6. print(pd.Series(clusters).value_counts())
  1. Built purchase clusters:
  2. 1    1117
  3. 4     911
  4. 0     638
  5. 2     566
  6. 3     430
  7. dtype: int64
  1. from sklearn.metrics import silhouette_samples, silhouette_score
  2. from customer_segmentation_toolkit.analyse_purchases import plot_silhouette
  4. silhouette_avg = silhouette_score(matrix, clusters)
  5. sample_silhouette_values = silhouette_samples(matrix, clusters)
  6. # Plotting silhouette values
  7. plot_silhouette(N_PURCHASE_CLUSTERS, [-0.07, 0.33], len(matrix), sample_silhouette_values, clusters)

svg

  1. from customer_segmentation_toolkit.analyse_purchases import add_purchase_clusters_info
  3. # Constructing the result DataFrame
  4. df_with_clusters = add_purchase_clusters_info(df_cleaned, clusters, N_PURCHASE_CLUSTERS)
  5. print(f'Added purchase clusters info to the offline cleaned dataset:')
  6. print(f'Shape: {df_with_clusters.shape}')
  7. print(f'Columns: {list(df_with_clusters.columns)}')
  1. Added purchase clusters info to the offline cleaned dataset:
  2. Shape: (13081, 9)
  3. Columns: ['CustomerID', 'InvoiceNo', 'Basket Price', 'categ_0', 'categ_1', 'categ_2', 'categ_3', 'categ_4', 'InvoiceDate']
  1. from customer_segmentation_toolkit.load_split import split_by_invoice_date
  3. # Splitting the new dataset (offline + cluster info) to train+test
  4. df_offline_train, df_offline_test = split_by_invoice_date(df_with_clusters, TRAINTEST_DATE_SPLIT)
  5. print(f'Splitted: train of shape {df_offline_train.shape} + test of shape {df_offline_test.shape}')
  1. Splitted: train of shape (10054, 9) + test of shape (3027, 9)
  1. # Saving processed data
  3. OUTPUT = Path(f'../data/output/03_data_compute_description_keywords')
  4. OUTPUT.mkdir(exist_ok=True, parents=True)
  6. matrix.to_csv(f'{OUTPUT}/no_live_data__cleaned__keywords.csv', index=False)
  7. df_offline_train.to_csv(f'{OUTPUT}/no_live_data__cleaned__purchase_clusters__train.csv', index=False)
  8. df_offline_test.to_csv(f'{OUTPUT}/no_live_data__cleaned__purchase_clusters__test.csv', index=False)
  10. pd.DataFrame(THRESHOLD, columns=['threshold']).to_csv(f'{OUTPUT}/threshold.csv', index=False)
  12. Path(f'{OUTPUT}/n_purchase_clusters.txt').write_text(str(N_PURCHASE_CLUSTERS))
  13. Path(f'{OUTPUT}/traintest_date_split.txt').write_text(str(TRAINTEST_DATE_SPLIT))
  15. print(f'Output data saved to {OUTPUT}: {[p.name for p in Path(OUTPUT).iterdir()]}')
  1. Output data saved to ../data/output/03_data_compute_description_keywords: ['no_live_data__cleaned__keywords.csv', 'no_live_data__cleaned__purchase_clusters__test.csv', 'threshold.csv', 'n_purchase_clusters.txt', 'traintest_date_split.txt', 'no_live_data__cleaned__purchase_clusters__train.csv']

04. Analyse customer categories

  1. from customer_segmentation_toolkit.load_split import load_data_csv
  3. N_CUSTOMER_CLUSTERS = 11
  4. SELECTED_CUSTOMERS_CATEG_THRESHOLD = 40
  6. # Loading cleaned offline train dataset
  7. DATA = '../data/output/03_data_compute_description_keywords'
  8. N_PURCHASE_CLUSTERS = int(Path(f'{DATA}/n_purchase_clusters.txt').read_text().strip())
  10. basket_price = load_data_csv(f'{DATA}/no_live_data__cleaned__purchase_clusters__train.csv')
  11. print(f'Loaded purchase clusters data of shape: {basket_price.shape}')
  12. print(basket_price.head())
  13. print('...')
  1. Loaded purchase clusters data of shape: (10054, 9)
  2.   CustomerID  InvoiceNo  Basket Price  categ_0  categ_1  categ_2  categ_3  \
  3. 0      12347     537626        711.79    83.40   187.20   293.35   124.44   
  4. 1      12347     542237        475.39    53.10   168.75   169.20     0.00   
  5. 2      12347     549222        636.25    71.10   369.15   115.00     0.00   
  6. 3      12347     556201        382.52    78.06    74.40   168.76    19.90   
  7. 4      12348     539318        892.80     0.00   414.00     0.00     0.00   
  9.    categ_4                   InvoiceDate  
  10. 0    23.40 2010-12-07 14:57:00.000001024  
  11. 1    84.34 2011-01-26 14:29:59.999999744  
  12. 2    81.00 2011-04-07 10:42:59.999999232  
  13. 3    41.40 2011-06-09 13:01:00.000000256  
  14. 4   478.80 2010-12-16 19:09:00.000000000  
  15. ...
  1. from customer_segmentation_toolkit.analyse_customers import build_transactions_per_user
  3. # Building transactions per user
  4. transactions_per_user = build_transactions_per_user(basket_price, n_purchase_clusters=N_PURCHASE_CLUSTERS)
  5. print(f'Built transactions per user, shape: {transactions_per_user.shape}')
  6. print(transactions_per_user.head())
  7. print('...')
  1. Built transactions per user, shape: (3143, 13)
  2.   CustomerID  count     min     max        mean      sum    categ_0  \
  3. 0      12347      4  382.52  711.79  551.487500  2205.95  12.949523   
  4. 1      12348      3  227.44  892.80  495.746667  1487.24   0.000000   
  5. 2      12350      1  334.40  334.40  334.400000   334.40  27.900718   
  6. 3      12352      4  144.35  840.30  360.370000  1441.48   3.683714   
  7. 4      12353      1   89.00   89.00   89.000000    89.00  19.887640   
  9.      categ_1    categ_2    categ_3    categ_4  LastPurchase  FirstPurchase  
  10. 0  36.242889  33.831682   6.543213  10.432693            52            236  
  11. 1  54.059869   0.000000   0.000000  45.940131           117            227  
  12. 2  60.406699   0.000000   0.000000  11.692584           179            179  
  13. 3  77.977495   5.771846  11.859339   0.707606           131            165  
  14. 4  13.033708   0.000000  67.078652   0.000000            73             73  
  15. ...
  1. from customer_segmentation_toolkit.analyse_customers import (
  2.     plot_customers_pca,
  3.     convert_customers_df_to_np,
  4.     analyse_customers_pca,
  5. )
  7. # Analysing customers distribution via PCA
  8. matrix = convert_customers_df_to_np(transactions_per_user, N_PURCHASE_CLUSTERS)
  9. scaled_matrix, pca = analyse_customers_pca(matrix)
  11. plot_customers_pca(matrix, pca)

svg

  1. from customer_segmentation_toolkit.analyse_customers import compute_customer_clusters
  3. # Computing customers clusters via Kmeans
  4. clusters_clients = compute_customer_clusters(scaled_matrix, N_CUSTOMER_CLUSTERS)
  5. print('Computed customers clusters via Kmeans:')
  6. display(pd.Series(clusters_clients).value_counts())
  1. Computed customers clusters via Kmeans:
  5. 7     1186
  6. 6      475
  7. 0      305
  8. 3      276
  9. 8      239
  10. 9      235
  11. 1      226
  12. 4      152
  13. 2       32
  14. 5       10
  15. 10       7
  16. dtype: int64
  1. from sklearn.metrics import silhouette_samples, silhouette_score
  2. from customer_segmentation_toolkit.analyse_purchases import plot_silhouette
  4. silhouette_avg = silhouette_score(scaled_matrix, clusters_clients)
  5. sample_silhouette_values = silhouette_samples(scaled_matrix, clusters_clients)
  7. # Plotting silhouette values
  8. plot_silhouette(N_CUSTOMER_CLUSTERS, [-0.15, 0.55], len(scaled_matrix), sample_silhouette_values, clusters_clients)

svg

  1. from customer_segmentation_toolkit.analyse_customers import plot_customer_categories
  3. # Plotting customers categories
  4. plot_customer_categories(scaled_matrix, clusters_clients, N_CUSTOMER_CLUSTERS)

svg

  1. from customer_segmentation_toolkit.analyse_customers import add_customer_clusters_info
  3. # Constructing the result dataset
  4. merged_df = add_customer_clusters_info(transactions_per_user, clusters_clients)
  5. print(f'Constructed the result dataset:')
  6. print(f'Shape: {merged_df.shape}')
  7. print(f'Columns: {list(merged_df.columns)}')
  1. Constructed the result dataset:
  2. Shape: (3143, 14)
  3. Columns: ['CustomerID', 'count', 'min', 'max', 'mean', 'sum', 'categ_0', 'categ_1', 'categ_2', 'categ_3', 'categ_4', 'LastPurchase', 'FirstPurchase', 'cluster']
  1. from customer_segmentation_toolkit.analyse_customers import compute_aggregated_customer_clusters_info
  3. # Constructing the aggregated cluster info dataset
  4. selected_customers_df = compute_aggregated_customer_clusters_info(merged_df, N_PURCHASE_CLUSTERS, N_CUSTOMER_CLUSTERS,
  5.                                                                   categ_threshold=SELECTED_CUSTOMERS_CATEG_THRESHOLD)
  6. print('Constructed the aggregated cluster info:')
  7. print(f'Shape: {selected_customers_df.shape}')
  8. print(f'Columns: {list(selected_customers_df.columns)}')
  1. Constructed the aggregated cluster info:
  2. Shape: (11, 14)
  3. Columns: ['cluster', 'count', 'min', 'max', 'mean', 'sum', 'categ_0', 'categ_1', 'categ_2', 'categ_3', 'categ_4', 'LastPurchase', 'FirstPurchase', 'size']
  1. # Saving processed data
  2. OUTPUT = Path(f'../data/output/04_data_analyse_customers')
  3. OUTPUT.mkdir(exist_ok=True, parents=True)
  5. selected_customers_df.to_csv(f'{OUTPUT}/no_live_data__cleaned__purchase_clusters__train__selected_customers_aggregated.csv', index=False)
  6. merged_df.to_csv(f'{OUTPUT}/no_live_data__cleaned__purchase_clusters__train__customer_clusters.csv', index=False)
  8. Path(f'{OUTPUT}/n_customer_clusters.txt').write_text(str(N_CUSTOMER_CLUSTERS))
  10. print(f'Output data saved to {OUTPUT}: {[p.name for p in Path(OUTPUT).iterdir()]}')
  1. Output data saved to ../data/output/04_data_analyse_customers: ['n_customer_clusters.txt', 'no_live_data__cleaned__purchase_clusters__train__selected_customers_aggregated.csv', 'no_live_data__cleaned__purchase_clusters__train__customer_clusters.csv']

05. Download dataset and use it for training

  1. from customer_segmentation_toolkit.data_zoo import download_data_csv
  2. from sklearn.model_selection import train_test_split
  4. # Download dataset from the data_zoo:
  5. csv = 'no_live_data__cleaned__purchase_clusters__train__customer_clusters.csv'
  6. selected_customers: pd.DataFrame = download_data_csv(f'data/output/04_data_analyse_customers/{csv}')
  8. X = selected_customers[['mean', 'categ_0', 'categ_1', 'categ_2', 'categ_3', 'categ_4' ]]
  9. Y = selected_customers['cluster']
  11. X_train, X_test, Y_train, Y_test = train_test_split(X, Y, train_size = 0.8)
  13. X_train.shape, X_test.shape, Y_train.shape, Y_test.shape
  1. ((2514, 6), (629, 6), (2514,), (629,))