Github user MLnick commented on the issue:
https://github.com/apache/spark/pull/18748
Ok, so I did some larger-scale test on a cluster (3x workers, each with 48
cores / 100GB allocated RAM with 1 executor).
On same `movielens-latest` datasets (~250,000 users and ~33,000 movies),
using a **30% sample** of user ids:
```
scala> // all users
scala> spark.time { model.recommendForAllUsers(k).foreach(_ => Unit) }
Time taken: 25104 ms
scala> // user sample
scala> spark.time { model.recommendForUserSubset(userSample, k).foreach(_
=> Unit) }
Time taken: 8963 ms
scala> 8963 / 25104.0
res16: Double = 0.35703473550031867
```
On a much larger dataset - Amazon books ratings data (8 million users, 2.3
million items) also using a **30% user sample**:
```
scala> // all users
scala> spark.time { model.recommendForAllUsers(k).foreach(_ => Unit) }
Time taken: 32985936 ms
=> 9.16 hours
scala> // user sample
scala> spark.time { model.recommendForUserSubset(userSample, k).foreach(_
=> Unit) }
Time taken: 8164421 ms
=> 2.26 hours
scala> 8164421 / 32985936.0
res7: Double = 0.24751218216151272
```
So it's a reasonably consistent range *25-35%* of time for a *30%* user
sample (I found broadly similar results with a 70% user sample, taking about
60% of the recommend-for-all time).
@mpjlu could you double check your results? What I find is consistent with
my expectations that computing for a subset should take time roughly
proportional to the ratio of the ids in the subset to the total. It appears to
me the extra `distinct` and `join` don't have too much impact on overall
runtime.
However your results are very different so we should understand why.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
