Add option for limiting rows of retrieved of results

[bburky]

I found the pandas-gbq interface easy to use and wanted to also use it for creating tables in BigQuery, not just downloading all the results at once. The existing capabilities of read_gbq() is actually already sufficient to do this, because you can just set query.destinationTable in the job configuration. However, I would like to limit the number of retrieved rows to a small sample of the whole table that was created instead of downloading the many thousands of rows that were created.

I've already played with making the changes myself in a project I'm working on:

http://nbviewer.jupyter.org/github/bburky/subredditgenderratios/blob/master/Subreddit%20Gender%20Ratios.ipynb

In the current code for run_query(), you read all rows from the table by converting the iterator into a list. Instead, you could pass the iterator to itertools.islice() first to limit it to a configurable limit. You can look at my code to see how it could be done.

Also, if you're interested I could contribute the IPython %%bigquery cell magic I am using in that project. It should be a very simple wrapper around read_gbq().

[max-sixty]

Adding the %%bigquery magic would be interesting. How similar is that to the magic in google's datalab?

[bburky]

Oh, I didn't know about datalab. My %%bigquery magic is basically a thin wrapper around read_gbq(). I can't quite tell how datalab's %%bq magic works, it may do something similar around their own APIs. They also have a --name argument for reusing results, but I can't tell if it saves tables, or just lets you save the output to a variable for reuse.

Would work on an IPython magic be done in this repository, or a separate standalone one? It would introduce a dependency on IPython.

[max-sixty]

Would work on an IPython magic be done in this repository, or a separate standalone one? It would introduce a dependency on IPython.

I'm not sure actually (it could be an optional dependency though). @jreback @tswast ?

[jreback]

you can do it in this repo; it a separate piece so the dep is fine (and only impacts the magic piece)

[tswast]

Yeah, could be an optional dependency.

@bburky: My coworker @alixhami has started some work on making a %%bigquery magic, too. He's planning on sending a PR soon to https://github.com/GoogleCloudPlatform/google-cloud-python to add the magic command there. You may wish to coordinate.

[tswast]

Follow-up for magics, the google-cloud-bigquery library is adding a BigQuery magic in googleapis/google-cloud-python#4983

[max-sixty]

I think we can close this - it's open because of the %%bigquery, which is now upstream.

Re limiting rows - that's very easy to do with a LIMIT clause

Reopen if anyone disagrees

[tswast]

Re limiting rows - that's very easy to do with a LIMIT clause

The issue was opened with the thought that you could do a query with a lot of results and write to a destination table, but only want to sample the results.

However, I would like to limit the number of retrieved rows to a small sample of the whole table that was created instead of downloading the many thousands of rows that were created.

Limiting the maximum results via pandas-gbq probably doesn't make the most sense if you want a representative sample. For that you'd want to add a little randomness and select from the destination table.

Or @bburky did you not want a representative sample, more just a preview?

[max-sixty]

Ah, preview makes sense. Sorry for being overzealous.

[bburky]

Yes. I was running a query that returned many many results and saving the results with `query.destinationTable`. To sanity check the query I ran, I wanted to see a 100 row preview or so. A possible implementation is just `itertools.islice()` of the query results. I don't mind if this isn't completely built into pandas-gbq, but there wasn't any way for me to hook or modify anything except copy pasting and modifying `run_query()`.