Skip to content

Latest commit

 

History

History
218 lines (204 loc) · 7.62 KB

CHANGES.md

File metadata and controls

218 lines (204 loc) · 7.62 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

v2.0.0 - 2020-03-13

Added

1.1 - 2019-05-10

  • Bump version to 1.1.
  • Pin version of scikit learn to 0.20 since we are still using python 2.7
  • Example vessel scoring
  • give shell friendly name
  • add figure source for document:
  • Update README.md
  • 61 add measures offset
  • Offset zero by default
  • Added PASSWORD option

multi-model-1.0 - 2016-11-20

  • Bugfix for AddMeasures
  • Support offsets for add_measures (#62) *Supportoffsetsforadd_measures *Somedocumentation *Bugfix *debug;add50%offset;checkimprovement *makeasclosetopre-shiftedversionaspossiblefortesting *centerranges *pulloutis_daylightasapredictor *usesegmentsindedupping
  • use segments in dedupping
  • pull out is_daylight as a predictor
  • center ranges
  • dedup messages
  • make as close to pre-shifted version as possible for testing
  • debug; add 50% offset; check improvement
  • Bugfix
  • Some documentation
  • Support offsets for add_measures
  • add PDF documentation of ML fishing score
  • More notebook cleanup
  • Moved more code to the modules and made the training use the same system used for production
  • Some notebook cleanup
  • Consolidate Notebooks
  • Edit text of Compare-Models
  • update Compare-Models text
  • remove models not in notebbooks
  • Make text in compare-with-dal more generic and rename to compare-models
  • move Dal Comparison
  • Accuracy vs ais degredation
  • Add model descriptions
  • move and rename model-description notebook
  • fix mispellings
  • fixed typo in AIS notebook
  • create model sensitivity notebook
  • add plots using binned data as comparison
  • Look at vessels with 100-200 PPD
  • add title for improved section
  • Improve fishing hours versus pts_per_day plot
  • add plot fishing hours vs pts-per-day at different max_gaps
  • remove redundant graph
  • combine plots
  • add fishing hours plots for dropped out data as well
  • prettier plots
  • Use fishing hours in point density comparison; compare effect of max gap
  • fix info leak and and add results using non-even training
  • fix notebook bug; add measure_count field to measures
  • add comparison of fishing hours
  • improve database munging (add_features / add_all_features); update notebook
  • rough draft - creation of reduced measures is messy
  • rerun all models
  • create model sensitivity notebook
  • Fix typos; clarify text
  • Bugfix and attribution
  • Added source links and description of the gear type specific models
  • Compare ps with dal
  • Add model comparisons to Model Descriptions notebook
  • add alignment to mulitline eqn in the hopes github renders correctly
  • Move markdown back to notebook
  • initial model descriptions
  • add missing retry
  • some description of model history
  • Rename notebook to 'Compare with Dal.ipynb'
  • Compare all results, not just PS
  • Add notebook comparing Dalhousie and GFW PS results
  • Switch RF model to use colspec
  • fix to use utcfromtimestamp
  • 11 daylight
  • Updated README with new data location
  • Bugfixed notebooks
  • Got time-of-day stuff to work
  • Bugfixes to day/night
  • First attempt at using daylight
  • Script to predict manually
  • Merge branch 'multi-model' into evaluate-non-model-results
  • Refactor so we can evalutae results from external models

release-1.0 - 2016-06-18

  • Remove unnecessary dependencies
  • Bugfix for missing dependency
  • Delete Broken model.ipynb
  • Install libatlas-dev first
    • coveralls
    • Building at Travis. We can at least test the install process.
    • .idea/, venv/, etc.
  • Remove unecessary dependencies and import optional depdencies in functions at runtime.
  • Added a script to anonymize mmsis
  • Multi model stuff
  • API bugfix
  • Bugfixes and new models
  • Add measures port rh
  • Bugfixes and final test
  • Generate all columns
  • Bugfix
  • Track down pipeline problem.
  • Added build files to gitignore
  • fix(?) new add measures
  • Broken model
  • Broken model?
  • Screen out points with speed==None
  • Removed ipython checkpoints from git
    • Improve way Kristina's data is fixed
    • Add method to turn CSV transit data
    • test how model is performing on new transit data and on
  • previously dumped models.
  • Bugfix
  • Generalized the numpy2message stuff
  • Began converting to new stream based API
  • Ported the rolling measures / add measures code from benthos-pipeline
  • Update README.md
  • Create LICENSE
  • Update README.md
  • Find and change add_measures to fix unscaled data
  • Added trained models to git
  • Merge branch 'installation'
  • Load models again
  • harmonize interface between LogisticModel and LogisticScorer
  • Basic pip install code
  • Basic pip install code
  • Cleanup
  • Bugfixed imports
  • Updated import paths
  • More cleanup
  • Added datasets and virtualenv to gitignore
  • More cleanup
  • Removed old stuff and reorganized the data a bit (classified-filtered.npz -> data/alex)
  • Bugfix
  • Better graphing
  • Rerun some notebook cells with better output
  • Added right-hand-side scale
    • Tweak Model interface so that we can use GridSearchCV from sklearn.
    • Setup Optimize_model.ipynb to use grid search
    • Add bad tracks that Egil created to Compare_single_model.ipynb
    • Added LogisticScorer class for possible use with Pipeline
    • Incorporated Egil's changes to the evaluate_model.py, plus more.
  • Added precision/recall to graphs
  • Refactor models
  • Slight generalizations to work with new dataset
  • Continue refactoring
    • Fold make_features options into models instead.
    • Setup notebooks to loop over models rather than doing by hand
    • Clean up and add docstrings.
  • Add multi-mindow logistic model
  • Generalized scripts a bit
  • add 12 hour heuristic
  • Modify score_development to use load_dataset_by_vessel
  • Start branch to refactor models to make them easier to compare
    • Made sklearn-like models for Logistic, Random Forest and Legacy models
    • Wrote preliminary evaluation code.
  • -> interface needs work
  • 486 Classifier gives fishing score for transits
  • Link to data
  • New scoring of all of Alex and Kristinas data
  • Scoring using Kristinas data
  • Score development bugfixing so it works for purse seiners too
  • Bugfixed parsing of Kristinas data
  • Oups
  • Merge branch 'master' of github.com:GlobalFishingWatch/vessel-scoring
  • Measurs with different windows, classified-filtered with lat/lon
  • Create README.md
  • Scored trawlers too
  • New logistic regression, plus some of Kristinas data
  • Some tools to deal with Kristinas data
  • Score reimplementation in numpy
  • Precision vs recall
  • Precision vs recall
  • Precision vs recall
  • Better score again
  • Better score
  • Better scoring
  • Used all different window sizes in score development
  • Added data for different window sizes
  • Stuff
  • Split up evaluation and development
  • Extracted some logic into modules
  • Some new data and more tools
  • Bugfix to exclude near-shore scores
  • Created a new optimal score
  • More work towards a better scoring fit
  • Matched polynomials to fishing probabilities for all variables
  • More data, and converted the notebook to use structured arrays
  • Better graphing
  • Requirements for ipython
  • Initial commit