Filtering a spark dataframe based on date

问题:

I have a dataframe of

date, string, string

I want to select dates before a certain period. I have tried the following with no luck

 data.filter(data("date") < new java.sql.Date(format.parse("2015-03-14").getTime))

I'm getting an error stating the following

org.apache.spark.sql.AnalysisException: resolved attribute(s) date#75 missing from date#72,uid#73,iid#74 in operator !Filter (date#75 < 16508);

As far as I can guess the query is incorrect. Can anyone show me what way the query should be formatted?

I checked that all enteries in the dataframe have values - they do.

回答1:

The following solutions are applicable since spark 1.5 :

For lower than :

// filter data where the date is lesser than 2015-03-14
data.filter(data("date").lt(lit("2015-03-14")))      

For greater than :

// filter data where the date is greater than 2015-03-14
data.filter(data("date").gt(lit("2015-03-14"))) 

For equality, you can use either equalTo or === :

data.filter(data("date") === lit("2015-03-14"))

If your DataFrame date column is of type StringType, you can convert it using the to_date function :

// filter data where the date is greater than 2015-03-14
data.filter(to_date(data("date")).gt(lit("2015-03-14"))) 

You can also filter according to a year using the year function :

// filter data where year is greater or equal to 2016
data.filter(year($"date").geq(lit(2016))) 

回答2:

In PySpark(python) one of the option is to have the column in unix_timestamp format.We can convert string to unix_timestamp and specify the format as shown below. Note we need to import unix_timestamp and lit function

from pyspark.sql.functions import unix_timestamp, lit

df.withColumn("tx_date", to_date(unix_timestamp(df_cast["date"], "MM/dd/yyyy").cast("timestamp")))

Now we can apply the filters

df_cast.filter(df_cast["tx_date"] >= lit('2017-01-01')) \
       .filter(df_cast["tx_date"] <= lit('2017-01-31')).show()

回答3:

Don't use this as suggested in other answers

.filter(f.col("dateColumn") < f.lit('2017-11-01'))

But use this instead

.filter(f.col("dateColumn") < f.unix_timestamp(f.lit('2017-11-01 00:00:00')).cast('timestamp'))

This will use the TimestampType instead of the StringType, which will be more performant in some cases. For example Parquet predicate pushdown will only work with the latter.