pyspark reduce key being a tuple values nested lists

Question

My problem is the following: I am parsing users interactions, each time an interaction is detected I emit ((user1,user2),((date1,0),(0,1))). The zero's are here for the direction of the interaction.

I cannot figure out why I cannot reduce this output with the following reduce function:

def myFunc2(x1,x2):
    return (min(x1[0][0],x2[0][0]),max(x1[0][0],x2[0][0]),min(x1[0][1],x2[0][1]),max(x1[0][1],x2[0][1]),x1[1][0]+x2[1][0],x1[1][1]+x2[1][1])

The output of my mapper (flatmap(myFunc)) is correct:

((7401899, 5678002), ((1403185440.0, 0), (1, 0)))
((82628194, 22251869), ((0, 1403185452.0), (0, 1)))
((2162276, 98056200), ((1403185451.0, 0), (1, 0)))
((0509420, 4827510), ((1403185449.0, 0), (1, 0)))
((7974923, 9235930), ((1403185450.0, 0), (1, 0)))
((250259, 6876774), ((0, 1403185450.0), (0, 1)))
((642369, 6876774), ((0, 1403185450.0), (0, 1)))
((82628194, 22251869), ((0, 1403185452.0), (0, 1)))
((2162276, 98056200), ((1403185451.0, 0), (1, 0)))

But running

lines.flatMap(myFunc) \
              .map(lambda x: (x[0], x[1])) \
              .reduceByKey(myFunc2)

Gives me the error

return (min(x1[0][0],x2[0][0]),max(x1[0][0],x2[0][0]),min(x1[0][1],x2[0][1]),max(x1[0][1],x2[0][1]),x1[1][0]+x2[1][0],x1[1][1]+x2[1][1])

TypeError: 'int' object has no attribute 'getitem'

I guess I am messing something up in my keys but I don't know why (I tried to recast the key to tuple as said here but same error)

Some idea ? Thanks a lot

Katya Willard · Answer 1 · 2016-05-25T13:21:26.277

Okay, I think the problem here is that you are indexing too deep in items that don't go as deep as you think.

Let's examine myFunc2

def myFunc2(x1,x2):
    return (min(x1[0][0],x2[0][0]),max(x1[0][0],x2[0][0]),min(x1[0][1],x2[0][1]),max(x1[0][1],x2[0][1]),x1[1][0]+x2[1][0],x1[1][1]+x2[1][1])

Given your question above, the input data will look like this:

((467401899, 485678002), ((1403185440.0, 0), (1, 0)))

Let's go ahead and assign that data row equal to a variable.

x = ((467401899, 485678002), ((1403185440.0, 0), (1, 0)))

What happens when we run x[0]? We get (467401899, 485678002). When we run x[1]? We get ((1403185440.0, 0), (1, 0)). That's what your map statement is doing, I believe.

Okay. That's clear.

In your function myFunc2, you have two parameters, x1 and x2. Those correspond to the variables above: x1 = x[0] = (467401899, 485678002) and x2 = x[1] = ((1403185440.0, 0), (1, 0))

Now let's examine just the first part of your return statement in your function.

min(x1[0][0], x2[0][0])

So, x1 = (467401899, 485678002). Cool. Now, what's x1[0]? Well, that's 467401899. Obviously. But wait! What's x1[0][0]? You're tryinig to get the zeroth index of the item at x1[0], but the item at x1[0] isn't a list or a tuple, it's just an int. And objects of <type 'int'> don't have a method called getitem.

To summarize: you're digging too deep into objects that are not nested that deeply. Think carefully about what you are passing into myFunc2, and how deep your objects are.

I think the first part of the return statement for myFunc2 should look like:

return min(x1[0], x2[0][0]). You can index deeper on x2 because x2 has more deeply nested tuples!

When I run the following, it works just fine:

a = sc.parallelize([((7401899, 5678002), ((1403185440.0, 0), (1, 0))),
((82628194, 22251869), ((0, 1403185452.0), (0, 1))),
((2162276, 98056200), ((1403185451.0, 0), (1, 0))),
((1509420, 4827510), ((1403185449.0, 0), (1, 0))),
((7974923, 9235930), ((1403185450.0, 0), (1, 0))),
((250259, 6876774), ((0, 1403185450.0), (0, 1))),
((642369, 6876774), ((0, 1403185450.0), (0, 1))),
((82628194, 22251869), ((0, 1403185452.0), (0, 1))),
((2162276, 98056200), ((1403185451.0, 0), (1, 0)))])

b = a.map(lambda x: (x[0], x[1])).reduceByKey(myFunc2)

b.collect()

[((1509420, 4827510), ((1403185449.0, 0), (1, 0))),
 ((2162276, 98056200), (1403185451.0, 1403185451.0, 0, 0, 2, 0)),
 ((7974923, 9235930), ((1403185450.0, 0), (1, 0))), 
 ((7401899, 5678002), ((1403185440.0, 0), (1, 0))), 
 ((642369, 6876774), ((0, 1403185450.0), (0, 1))), 
 ((82628194, 22251869), (0, 0, 1403185452.0, 1403185452.0, 0, 2)),
 ((250259, 6876774), ((0, 1403185450.0), (0, 1)))]

I am not sure to follow you. myfunc2 is the function passed to the reducer ? `x1 = x[0]` being my key, being `(467401899, 485678002)`. So, I am probably missing something here but, myfun2 takes two inputs, being two values with same key. So x1 and x2 should have the format of `((1403185440.0, 0), (1, 0))`. I guess it's me missing a point here but I do not see why the key should be passed as input to the reducer (badly worded). — HorusH, May 10 '16 at 07:59
I think I may also be struggling to understand your example. Could you add more data like `((467401899, 485678002), ((1403185440.0, 0), (1, 0)))` in your question? That way I could actually run the code and try to replicate your error. — Katya Willard, May 10 '16 at 15:00
Thanks again for your answer. I thought that the issue might be from the (unnecessary) use of the `map` after the `flatmap` but I get the same output before and after the `map` and the same error. I think that there is an error in my code for grouping records with the key `(user_id_1,user_id_2)` but I cannot figure out what am I doing wrong — HorusH, May 12 '16 at 14:24

pyspark reduce key being a tuple values nested lists

1 Answers1