PySpark DataFrame 入門

PySpark 入門

Benjamin Schmidt

Data Engineer

關於 DataFrame

DataFrame：表格格式（列／欄）
支援類 SQL 操作
類似 pandas DataFrame 或 SQL TABLE
結構化資料

Dataframes

從檔案儲存建立 DataFrame

# Create a DataFrame from CSV
census_df = spark.read.csv('path/to/census.csv', header=True, inferSchema=True)

列印 DataFrame

# Show the first 5 rows of the DataFrame
census_df.show()

   age  education.num marital.status         occupation income
0   90              9        Widowed                  ?  <=50K
1   82              9        Widowed    Exec-managerial  <=50K
2   66             10        Widowed                  ?  <=50K
3   54              4       Divorced  Machine-op-inspct  <=50K
4   41             10      Separated     Prof-specialty  <=50K

列印 DataFrame 結構（Schema）

# Show the schema
census_df.printSchema()

Output:
root
 |-- age: integer (nullable = true)
 |-- education.num: integer (nullable = true)
 |-- marital.status: string (nullable = true)
 |-- occupation: string (nullable = true)
 |-- income: string (nullable = true)

在 PySpark DataFrame 上做基本分析

# .count() will return the total row numbers in the DataFrame
row_count = census_df.count()
print(f'Number of rows: {row_count}')

# groupby() allows the use of sql-like aggregations
census_df.groupBy('gender').agg({'salary_usd': 'avg'}).show()

Other aggregate functions are:

sum()
min()
max()

PySpark 分析常用函式

.select()：選取特定欄位
.filter()：依條件篩選列
.groupBy()：依一個或多個欄位分組
.agg()：對分組資料做彙總

關鍵函式範例

# Using filter and select, we can narrow down our DataFrame
filtered_census_df = census_df.filter(df['age'] > 50).select('age', 'occupation')
filtered_census_df.show()

Output
+---+------------------+
|age|       occupation |
+---+------------------+
| 90|                 ?|
| 82|   Exec-managerial|
| 66|                 ?|
| 54| Machine-op-inspct|
+---+------------------+

一起來練習吧！

PySpark 入門