如何在GPU DataFramecuDF中使用条件过滤数据框架？

Question

我想基于列值过滤cuDF数据帧，然后根据指定的条件创建一个新列。基本上，如何在cuDF中应用以下内容？

df.loc[df.column_name condition, 'new column name'] = 'value if condition is met'

Answer 1

在cuDF中赋予熊猫

# value to be replaced in series 
value = 'value if condition is met'
# condition to qualify for replacement
mask = df.column_name condition

# https://docs.rapids.ai/api/cudf/stable/
df['new column name'] = df.masked_assign(value, mask)

应用示例

"""explanation: 
  >> if there is no pool, pool_sqft should be 0
"""

# value to be replaced in series 
value = 0
# condition to qualify for replacement
mask = df_train['pool_count']==0

# https://docs.rapids.ai/api/cudf/stable/
df['pool_sqft'] = df.masked_assign(value, mask)

Answer 2

虽然masked_assign在某些条件下有效，但是applymap是syntactically better and functionally similar to the Pandas API。

此外，@ ashwin-srinath提到__setitem()__即将发布0.9版，因此您只能执行df[condition] = value。 masked_assign不是熊猫API函数，因此__setitem()__可能只赞成masked_assign。

Answer 3

您也可以使用.query()

示例：

expr = "(a == 2) or (b == 3)"
filtered_df = df.query(expr)

其中a和b是数据框中列的名称。