Load data with user defined schema

[aktech]

The to_gbp function fails when there is a user-defined schema which is slightly different from the schema of DataFrame.

Traceback:

Traceback (most recent call last):
  File "<stdin>", line 4, in <module>
  File "pandas_gbq/gbq.py", line 979, in to_gbq
    schema=table_schema)
  File "pandas_gbq/gbq.py", line 569, in load_data
    self.process_http_error(ex)
  File "pandas_gbq/gbq.py", line 450, in process_http_error
    raise GenericGBQException("Reason: {0}".format(ex))
pandas_gbq.gbq.GenericGBQException: Reason: 400 POST https://www.googleapis.com/upload/bigquery/v2/projects/aktech-labs/jobs?uploadType=resumable: Provided Schema does not match Table aktech-labs:pandas_test.gamma. Field A has changed type from FLOAT to STRING

The primary reason for this is that the load_data function ignores the schema argument.

Solution: The schema argument is now passed to the load_chunks function inside load_data function. A suitable integration test has been added.

[codecov-io]

Codecov Report

Merging #150 into master will decrease coverage by 46.5%.
The diff coverage is 10%.

@@            Coverage Diff             @@
##           master    #150       +/-   ##
==========================================
- Coverage    75.8%   29.3%   -46.51%     
==========================================
  Files           8       8               
  Lines        1546    1556       +10     
==========================================
- Hits         1172     456      -716     
- Misses        374    1100      +726

Impacted Files	Coverage Δ
pandas_gbq/gbq.py	20.56% <ø> (-56.02%)	⬇️
pandas_gbq/tests/test_gbq.py	22.93% <10%> (-61.67%)	⬇️
pandas_gbq/_load.py	62.5% <0%> (-35%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ed17886...c563fc3. Read the comment docs.

[max-sixty]

Supersedes #131

[max-sixty]

This looks great. Any feedback @tswast ?

@aktech did you run the tests? Otherwise I will. (Travis integration tests don't work on PRs because of the auth keys)

@tswast more generally, do you think it's OK to merge these small PRs that look good, so the tests run on master, and then correct in the 5% of cases where it fails? Or we should run the tests on an 'auth-enabled' travis first?

[tswast]

LGTM.

@maxim-lian So long as whomever merges keeps an eye on the master build and fixes / reverts based on the outcome, I'm okay merging without requiring an auth-enabled Travis.

[aktech]

Thanks for the quick reply and merging.

@aktech did you run the tests? Otherwise I will. (Travis integration tests don't work on PRs because of the auth keys)

Yes, I did ran the tests locally as well as on travis:
https://travis-ci.org/aktech/pandas-gbq/builds/353407949