release.rst

Release Notes

• v1.7.0 (in development; not yet released)
• v1.6.0
  • Add nmr_transactions
  • Replace is_controlling_capital with is_stakeable
  • upload now reports if submission is stakeable
  • Add simple unit testing of functions used by the Report class
  • Bugfix report.user_nmr corner cases such when no nmr awarded
  • Bugfix report.group_confidence corner cases such as when cumsum < $6000
  • Update readme results by using a recent Numerai dataset
• v1.5.0
  • Add report.user_summary
  • Add report.user_nmr
  • Add report.user_nmr_tax
  • Add report.single_stake_payout
  • Rename report.new_users to report.headcount and add columns
• v1.4.0
  • Add LoocvSplitter
  • Add report.reputation
  • Add report.group_confidence
  • Add report.group_consistency
  • Add report.group_burn
  • Add confidence and s/c to Leaderboard
  • Bugfix when sorting prediction.performance by consistency
• v1.3.0
  • Add report.burn
  • Add report.participation
  • Add report.big_staker
  • Add report.new_user
  • Add report.user_participation
  • Add report.all
  • Add history
  • Improve report.earn
  • Leaderboard now includes stake amount s
  • Add utility functions isint and isstring
• v1.2.0
  • Add Report class
  • Add Leaderboard class
  • Remove top_consistency, ten99, top_stakers, top_earners; use Report instead
  • Remove download_leaderboard; use Leaderboard instead
  • Add get_current_round_number
  • report.consistency now uses ln(2) as benchmark for rounds before R102; ties are now broken with number of rounds entered
• v1.1.0
  • Add load_example_predictions
  • Add prediction.hash
  • prediction.check now checks for correlation to example predictions, maximum absolution Z score, minimum and maximum values
  • example_predictions model now rounds to 5 decimal places
  • SciPy is now a primary dependency of numerox (not just through sklearn)
• v1.0.0
  • Add support for tournament number; defaults to current tournament #1
  • NumerAPI 0.9.0 is required
  • is_controlling_capital 58% consistency now required
• v0.9.0
  • Change log loss benchmark to 0.693 from ln(2)
  • Add the model used by Numerai to generate example_predictions.csv
  • Add top_consistency
  • Add live logloss to download_leaderboard
  • top_stakers now returns a dataframe
  • Use Python's decimal.Decimal to avoid staking confidence rounding errors
  • Add 'logloss_pass' and 'length' to prediction.metrics_per_era
  • Add requirements.txt (thanks dhj-io)
• v0.8.0
  • show_stakes: refactor, reformat output, add more options
  • get_stakes: refactor, reformat output, add more options
  • Rename download_earnings to download_leaderboard
  • Add load_prediction_csv
  • tournament_number input variable renamed round_number
  • Can now handle nmr winnings in the staking tournament: ten99, top_stakers, top_earners, download_leaderboard
• v0.7.0
  • Add top_earners
  • Add round_resolution_date
  • Add nmr_resolution_price and get prices from coinmarketcap
  • Add year_to_tournament_range
  • Add more options to top_stakers
  • Remove nmr_price which hard coded prices
  • round_number input variable renamed tournament_number
• v0.6.0
  • Add ten99
  • Add top_stakers
  • Add download_earnings
  • Add nmr_at_addr
  • Add token_price_data
  • Add historical_price
• v0.5.0
  • Add compare_data
  • Add show_stakes example
  • prediction.save now has an append mode
  • prediction.performance now returns a dataframe
  • prediction.dominance now returns a dataframe
  • prediction.summary now returns a dataframe
  • Remove prediction.performance_df
  • Remove prediction.dominance_df
  • Remove prediction.summary_df
  • prediction.performance keyword changed from cols to columns
  • More prediction methods can now handle empty predictions
• v0.4.0
  • Numerox requires NumerAPI 0.8.1
  • Add merge_predictions
  • Add prediction.check
  • Add more columns to prediction.originality
  • Add optional choice of columns in prediction.performance_df
  • Column name changed to 'concord' in concordance function
  • Add example comparing performance of a single change across models
  • Add data.y_to_nan
  • Bugfix: prediction.__repr__ sometimes showed wrong fraction of missing y
  • Bugfix: merging predictions with same name but different ids may fail
  • More unit tests
• v0.3.1
  • Unit test coverage 89%, up from 65%
  • Improve reporting of upload status
  • Add testing.HiddenPrints
  • Bug fix: checking for equality of empty Predictions crashes
  • get_stakes informative error message when round_number < 61
  • More unit tests
• v0.3.0
  • Add upload to make submissions
  • download is the new name for download_dataset
  • Add prediction.compare
  • Add prediction.loc for indexing by Numerai row ids
  • Add prediction.rename
  • Add prediction.drop
  • prediction.concordance 3x faster when prediction contains 10 names
  • prediction.concordance now sorts by concordance
  • prediction.merge is no longer an inplace operation
  • prediction.merge_arrays is no longer an inplace operation
  • No longer take name as input: prediction.to_csv, prediction.summary, prediction.summary_df
  • Remove model.hash
  • Examples can now be run after installation: nx.examples.run_all_examples()
  • Redo compare_models example
  • Make more use of numerapi
  • Python package requests is no longer a dependency
  • Rewrite testing.micro_prediction for better unit testing
  • More unit tests
• v0.2.0
  • This release makes a large change to the numerox API
  • There are now 3 main classes instead of 4
  • The Report class has been merged into the Prediction class
  • The Prediction class can now hold the predictions from multiple models
  • New features have been added to the Prediction class
• v0.1.2
  • Numerox now uses (and requires) NumerAPI
  • Add file overwrite protection option to download_dataset
  • Beware: download_dataset will now raise by default if file exists
  • Add report.__setitem__
  • Add report.__contains__
  • Add data.loc for indexing by Numerai row ids
  • Add report.originality
  • Add report indexing (report.__getitem__)
  • More unit tests
• v0.1.1
  • Complete rewrite of all performance metrics
  • Add metrics_per_model
  • Add report.dominance
  • Add report.dominance_df
  • Add prediction.performance_df
  • Add mlpc model
  • Remove xgboost model to remove optional xgboost dependency
  • Rewrite examples of comparing performance of multiple models
  • More unit tests
• v0.1.0
  • Add report.correlation
  • Add prediction.consistency
  • Add prediction.metrics_per_era
  • Can now specify which metrics to calculate in metrics_per_era
  • Add sort_by to show_stakes
  • Add prediction.yhatnew
  • Add xgboost model
  • Add randomforest model
  • Add logisticPCA model
  • Models at top level: nx.extatrees instead of nx.model.extratrees, etc
  • logistic model now uses less regularization by default
  • Bugfix: display model name correctly when parameter dictionary is empty
  • More unit tests
• v0.0.9
  • Add ability to work with new (round 85) Numerai datasets
  • Update play_data with new numerai dataset
  • run now hides from your model the y you are trying to predict
  • Cumsum in show_stakes and get_stakes now dollars above you
  • model.hash combined hash of data, model name, and model parameters
  • Gentle refactor of splitters to reuse code
  • Bugfix: crash when balancing already balanced data
  • More unit tests
• v0.0.8
  • Add show_stakes
  • Add get_stakes
  • data.xnew is 3 times faster
  • data.column_list(x_only=False) replaces _column_list and _x_names
  • Example of Numerai's cross validation warning (hold out eras not rows)
  • Bugfix: data.xnew output didn't use contiguous memory
• v0.0.7
  • Add data.balance
  • Add data.subsample
  • Add data.hash
  • Add IgnoreEraCVSplitter
  • Add dataset_url function
  • All splitters now use a single base class
  • Add download_data_object to avoid hard coding path in examples
  • play_data is now data.y balanced
  • Rewrote update_play_data
  • More unit tests
• v0.0.6
  • Add concordance
  • New Runner class can run multiple models through a single data splitter
  • Update download_dataset for recent Numerai API change
  • Add RollSplitter roll forward fit-predict splits from consecutive eras
  • Add another verbosity level to run (verbosity=3)
  • Use play_data instead of numerai server or hard coding my local path
  • Bugfix: in v0.0.5 CVSplitter ran only a single cross validation fold
  • More unit tests
• v0.0.5
  • Data splitters can now be reused to run more than one model
  • To reuse a splitter, reset it: splitter.reset()
  • All splitters renamed; e.g. cheat_splitter is now CheatSplitter
  • Splitters are now iterator classes instead of generator functions
  • data.ids returns numpy string array copy instead of object array view
  • More unit tests
• v0.0.4
  • Add data.pca
  • Add examples of transforming features
  • You can now change the number of features with data.xnew
  • data.xnew is the new name of data.replace_x
  • shares_memory can now check datas with different number of x columns
  • More unit tests
• v0.0.3
  • Add examples
  • Add iterator data.era_iter
  • Add iterator data.region_iter
  • prediction.ids and prediction.yhat are now views instead of copies
  • Remove appveyor so that unit tests can use Python's tempfile
  • Bugfix: prediction.copy was not copying the index
  • Bugfix: mistakes in two unit tests meant they could never fail
  • More unit tests
• v0.0.2
  • data.x and data.y now return fast views instead of slow copies
  • era and region stored internally as floats
  • HDF5 datasets created with v0.0.1 cannot be loaded with v0.0.2
• v0.0.1
  • Preview release of numerox