Introducing the CLI

Many changes here, largely related to introducing the Command Line Interface ported from DirtyDuck

Refactor functionality to ExperimentBase class [Resolves #400]
Feature testing functionality [Resolves #420]
Storage refactoring, experiment CLI [Resolves #424]
Timechop visualization CLI [Resolves #437]

Initial v3 beta

Merge pull request #401 from dssg/push_to_base

Refactor functionality to ExperimentBase class [Resolves #400]

Fix old version for benchmark

v1.0.3

Manual bump version

Component Integration, S3 support

• Triage components (Timechop, Collate, Architect, Catwalk, Results-Schema, Audition) ported from other repositories into src.triage.component
• Audition - rule generator + multiple output models [Resolved #355 #346 #343]
• Audition readme (#382)
• Fixing s3 support [Resolves #364]
• inserting NULLs for individual feature importance - #361

v1.0.2

Using the git version

Performance/Usability updates, More temporal configuration

Pin tables so it installs without HDF requirement
Thresholding unit test [Resolves #71]
Parallelize Training [Resolves #100]
change to UChicago license (#110)
Close session when done
Update README [Resolves #106] (#107)
Workflow and performance updates [Resolves #45, #93, #99, #111, #112, #113, #114]
Remove errant logging statements
Add more docstrings
Merge experiment rows [Resolves #103]
Incorporate timechop interval parsing change
Fix evaluation of sparsely-labeled test matrices [Resolves #116]
Control number of database-connecting processes [Resolves #104]
Update setup.py to look for md files
Remove n_jobs from parameters sent to model_hash [Resolves #126]
Log more split info [Resolves #124]
Fix thresholding ties bug [Resolves #133]
Add sort_seed for scoring
More training logging
Write train uuid to models table [Resolves #128]
Move explain statements to new validate method, which can be manually run [Resolves #130]
Remove n_jobs from model group config [Resolves #132]
Include traceback when child process errors [Resolves #140]
Fix log split # denominator [Resolves #142]
Add train and test example frequency [Resolves #148]
Save test_uuid in predictions tables [Resolves #144]
Separate train/test prediction windows [Resolves #152]
Allow custom model group keys [Resolves #154]
Handle multiple label window sin LabelGenerator#generate_all_labels [Resolves #156]

First-style ranking

Switch rankings to 'first', so even in the case of ties predictions don't get the same rank.

Feature Groups, Parallelization, Dynamic Categoricals

• Remove rankings per as_of_date if as_of_date is in index [Resolves #59]
• Instantiate pipeline components in constructor [Resolves #44]
• Take in test_durations and pass to timechop [Resolves #62]
• Pass matrix directory to Architect [Resolves #64]
• Support parallelism [Resolves #28]
• Add pipelines to packages
• Switch to QueuePool
• Fix exclusion of feature aggregation tables [Resolves #66]
• Dynamic Categorical Choices [Resolves #33]
• Support model_comment [Resolves #75]
• Bypass collate aggregate tables [Resolves #76]
• Add logging per group table
• Feature Groups [Resolves #50]
• Have BinaryLabelGenerator look for outcome_date [Resolves #84]
• Standardize pipeline components on factory pattern [Resolves #85]
• Parallelize Feature Generation [Resolves #83]
• Add leave-one-in feature group mixer strategy [Resolves #88]
• Schema updates [Resolves #94]

Modeling pipeline, inspections-style time splitting and evaluation

Merge pull request #57 from dssg/inspection_evaluations

Support inspections-style evaluation [Resolves #54]

Initial release

Merge pull request #14 from dssg/yield_trained_models

Allow yielding of trained models [Resolves #11]