Tightened and more flexible testing

[kenalba]

This PR is kind of a mess, so consider this more of a review request than an actual pull request, but I spent some time honing our testing corpora and the way we treat it.

Broadly speaking, I removed the test_data folder because every document in there was already in sample_novels. Instead, now, I've created three .csv files - large_test_corpus, small_test_corpus, and tiny_test_corpus - which will only select out the texts we want to use to test from our sample_novels. This way of thinking about test corpora as metadata-first rather than file-first is, I think, more flexible in the long run. The addition of tiny_test_corpus also gives us a 4 document corpus to test our most computationally hungry functions on, which is what motivated this shift in the first place.

It did mean adding an ignore_warnings flag to the corpus generator, since we now expect that the generator won't load every text file in a directory.

After doing this, I looked to see what functions are taking a long time to test and, where possible, rewrote them to work on more compact corpora. Doing this cut the time it takes to run coverage on my machine in half.

[codecov-io]

Codecov Report

Merging #136 into master will decrease coverage by 1.00%. The diff coverage is 32.38%.

@@            Coverage Diff             @@
##           master     #136      +/-   ##
==========================================
- Coverage   44.54%   43.53%   -1.01%     
==========================================
  Files          12       12              
  Lines        1623     1656      +33     
  Branches      353      365      +12     
==========================================
- Hits          723      721       -2     
- Misses        847      882      +35     
  Partials       53       53              

Impacted Files	Coverage Δ
gender_analysis/analysis/dunning.py	30.29% <0.00%> (ø)
gender_analysis/document.py	82.14% <ø> (ø)
gender_analysis/analysis/gender_adjective.py	30.37% <19.23%> (-9.63%)	:arrow_down:
gender_analysis/analysis/gender_frequency.py	49.63% <50.00%> (ø)
gender_analysis/corpus.py	66.66% <50.00%> (ø)
gender_analysis/gender.py	96.15% <93.33%> (ø)
gender_analysis/analysis/instance_distance.py	34.58% <100.00%> (ø)

---

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 5d9882f...233ed1b. Read the comment docs.