spark-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Michael Armbrust <mich...@databricks.com>
Subject Re: column expression in left outer join for DataFrame
Date Wed, 25 Mar 2015 19:10:35 GMT
Thats a good question.  In this particular example, it is really only
internal implementation details that make it ambiguous.  However, fixing
this was a very large change so we have defered it to Spark 1.4 and instead
print a warning now when you construct trivially equal expressions.  I can
try to explain a little bit about why solving this generally is (mostly)
impossible.  Consider the following:

val df = sqlContext.load(...)

val df1 = df
val df2 = df

df1.join(df2, df1("a") === df2("a"))

Compared with

"SELECT * FROM df df1 JOIN df df2 WHERE df1.a = df2.a"

In the first example, the assigning of df to df1 and df2 is completely
transparent to the catalyst optimizer as it is happening in Scala code.
This means that df1("a") and df2("a") are completely indistinguishable to
us (at least without crazy macro magic).  In contrast, the aliasing is
visible to the optimizer when are doing it in SQL instead of Scala and thus
we can differentiate.

In your case you are doing transformations, and we could assign new unique
ids each time a transformation is done.  However, we don't do this today,
and its a pretty big change.  There is a JIRA for this: SPARK-6231
<https://issues.apache.org/jira/browse/SPARK-6231>

On Wed, Mar 25, 2015 at 11:47 AM, S Krishna <skrishna.id@gmail.com> wrote:

> Hi,
>
> Thanks for your response.  I am not clear about why the query is ambiguous.
>
> val both = df_2.join(df_1, df_2("country")===df_1("country"),
> "left_outer")
>
> I thought df_2("country")===df_1("country") indicates that the country
> field in the 2 dataframes should match and df_2("country") is the
> equivalent of df_2.country in SQL, while  df_1("country") is the
> equivalent of df_1.country in SQL. So I am not sure why it is ambiguous. In
> Spark 1.2.0 I have used the same logic using SparkSQL  and Tables ( e.g.
>  "WHERE tab1.country = tab2.country")  and had no problems getting the
> correct result.
>
> thanks
>
>
>
>
>
> On Wed, Mar 25, 2015 at 11:05 AM, Michael Armbrust <michael@databricks.com
> > wrote:
>
>> Unfortunately you are now hitting a bug (that is fixed in master and will
>> be released in 1.3.1 hopefully next week).  However, even with that your
>> query is still ambiguous and you will need to use aliases:
>>
>> val df_1 = df.filter( df("event") === 0)
>>                   . select("country", "cnt").as("a")
>> val df_2 = df.filter( df("event") === 3)
>>                   . select("country", "cnt").as("b")
>> val both = df_2.join(df_1, $"a.country" === $"b.country"), "left_outer")
>>
>>
>>
>> On Tue, Mar 24, 2015 at 11:57 PM, S Krishna <skrishna.id@gmail.com>
>> wrote:
>>
>>> Hi,
>>>
>>> Thanks for your response. I modified my code as per your suggestion, but
>>> now I am getting a runtime error. Here's my code:
>>>
>>> val df_1 = df.filter( df("event") === 0)
>>>                   . select("country", "cnt")
>>>
>>> val df_2 = df.filter( df("event") === 3)
>>>                   . select("country", "cnt")
>>>
>>> df_1.show()
>>> //produces the following output :
>>> // country    cnt
>>> //   tw           3000
>>> //   uk           2000
>>> //   us           1000
>>>
>>> df_2.show()
>>> //produces the following output :
>>> // country    cnt
>>> //   tw           25
>>> //   uk           200
>>> //   us           95
>>>
>>> val both = df_2.join(df_1, df_2("country")===df_1("country"),
>>> "left_outer")
>>>
>>> I am getting the following error when executing the join statement:
>>>
>>> java.util.NoSuchElementException: next on empty iterator.
>>>
>>> This error seems to be originating at DataFrame.join (line 133 in
>>> DataFrame.scala).
>>>
>>> The show() results show that both dataframes do have columns named
>>> "country" and that they are non-empty. I also tried the simpler join ( i.e.
>>> df_2.join(df_1) ) and got the same error stated above.
>>>
>>> I would like to know what is wrong with the join statement above.
>>>
>>> thanks
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> On Tue, Mar 24, 2015 at 6:08 PM, Michael Armbrust <
>>> michael@databricks.com> wrote:
>>>
>>>> You need to use `===`, so that you are constructing a column expression
>>>> instead of evaluating the standard scala equality method.  Calling methods
>>>> to access columns (i.e. df.county is only supported in python).
>>>>
>>>> val join_df =  df1.join( df2, df1("country") === df2("country"),
>>>> "left_outer")
>>>>
>>>> On Tue, Mar 24, 2015 at 5:50 PM, SK <skrishna.id@gmail.com> wrote:
>>>>
>>>>> Hi,
>>>>>
>>>>> I am trying to port some code that was working in Spark 1.2.0 on the
>>>>> latest
>>>>> version, Spark 1.3.0. This code involves a left outer join between two
>>>>> SchemaRDDs which I am now trying to change to a left outer join
>>>>> between 2
>>>>> DataFrames. I followed the example  for left outer join of DataFrame
at
>>>>>
>>>>> https://databricks.com/blog/2015/02/17/introducing-dataframes-in-spark-for-large-scale-data-science.html
>>>>>
>>>>> Here's my code, where df1 and df2 are the 2 dataframes I am joining on
>>>>> the
>>>>> "country" field:
>>>>>
>>>>>  val join_df =  df1.join( df2,  df1.country == df2.country,
>>>>> "left_outer")
>>>>>
>>>>> But I got a compilation error that value  country is not a member of
>>>>> sql.DataFrame
>>>>>
>>>>> I  also tried the following:
>>>>>  val join_df =  df1.join( df2, df1("country") == df2("country"),
>>>>> "left_outer")
>>>>>
>>>>> I got a compilation error that it is a Boolean whereas a Column is
>>>>> required.
>>>>>
>>>>> So what is the correct Column expression I need to provide for joining
>>>>> the 2
>>>>> dataframes on a specific field ?
>>>>>
>>>>> thanks
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> View this message in context:
>>>>> http://apache-spark-user-list.1001560.n3.nabble.com/column-expression-in-left-outer-join-for-DataFrame-tp22209.html
>>>>> Sent from the Apache Spark User List mailing list archive at
>>>>> Nabble.com.
>>>>>
>>>>> ---------------------------------------------------------------------
>>>>> To unsubscribe, e-mail: user-unsubscribe@spark.apache.org
>>>>> For additional commands, e-mail: user-help@spark.apache.org
>>>>>
>>>>>
>>>>
>>>
>>
>

Mime
View raw message