overhaul inference framework

[knaaptime]

• expose counterfactual population generators (#172 )
• parallelize single value inference simulations (#35.)
• parallelize comparative inference
• add a randomization method based on individual-level permutation
• update inference plotting for new seaborn api
• properly handle generalized spatial indices in inference framework
• fix #178
• fix #179
• use more performant sklearn distance metrics in favor of libpysal DistanceBand weights
• fix boostrap inference for single value
• implement bootstrap inference for two value
• beef up docstrings

[knaaptime]

tagging @darribas and @ljwolf in case there is a better way to do the parallelization. Currently the threading backend is much faster for simulate_evenness and simulate_systematic_randomization, but multiprocessing seems to work much better for simulate_reallocation. The loky backend always crashes for any of them?

[knaaptime]

cc: @martinfleis

[knaaptime]

scratch that... the loky backend is fastest for all approaches, i just needed to update joblib and friends

[knaaptime]

@sjsrey @renanxcortes ok, this should be all set. We get a nice speedup in the inference classes thanks to the parallelization, the only side effect is that setting a random seed is a bit more complicated, so i opted to turn the test precision down to two decimals

[codecov-commenter]

Codecov Report

Merging #174 (dcae6bb) into master (403bcf6) will decrease coverage by 2.28%. The diff coverage is 61.57%.

@@            Coverage Diff             @@
##           master     #174      +/-   ##
==========================================
- Coverage   65.77%   63.49%   -2.29%     
==========================================
  Files         116      118       +2     
  Lines        4322     4421      +99     
==========================================
- Hits         2843     2807      -36     
- Misses       1479     1614     +135     

Impacted Files	Coverage Δ
segregation/network/network.py	25.58% <0.00%> (-1.70%)	:arrow_down:
segregation/tests/test_batch_compute.py	100.00% <ø> (ø)
segregation/tests/test_multiscalar_profile.py	100.00% <ø> (ø)
segregation/util/network.py	84.31% <0.00%> (ø)
segregation/inference/randomization.py	21.62% <21.62%> (ø)
segregation/singlegroup/spatial_dissim.py	87.17% <50.00%> (ø)
segregation/inference/comparative.py	61.64% <61.64%> (ø)
segregation/batch/batch_compute.py	90.36% <75.75%> (-9.64%)	:arrow_down:
segregation/inference/inference_wrappers.py	72.80% <78.43%> (-13.51%)	:arrow_down:
segregation/_base.py	82.47% <80.00%> (-0.69%)	:arrow_down:
... and 31 more

---

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 403bcf6...dcae6bb. Read the comment docs.

[knaaptime]

excellent, thanks James. I was pretty sure #177 only touched infrastructure, so figured it was easiest to continue focusing on this PR while touches more. I wasnt sure if there would be conflicts and since 177 is smaller it seemed easier to rebase. I'm open to suggestions if you think it would be easier some other way though

[jGaboardi]

excellent, thanks James. I was pretty sure #177 only touched infrastructure, so figured it was easiest to continue focusing on this PR while touches more. I wasnt sure if there would be conflicts and since 177 is smaller it seemed easier to rebase. I'm open to suggestions if you think it would be easier some other way though

Yep, probably better to do #177 after this PR merged.