pyrasgo 0.2.5

Alpha version of the Rasgo Python interface.

pyRasgo is a python SDK to interact with the Rasgo API. Rasgo accelerates feature engineering for Data Scientists.

Visit us at https://www.rasgoml.com/ to turn your data into Features in minutes!

Documentation is available at: https://docs.rasgoml.com/rasgo-docs/pyrasgo/

Package Dependencies

• idna>=2.5,<3
• more-itertools
• pandas
• pyarrow>=3.0
• pydantic
• pyyaml
• requests
• snowflake-connector-python>=2.4.0
• tqdm

Release Notes

• v0.2.5 (Aug 18, 2021)

  • adds handling and user notification for highly null dataframes which would otherwise not function well with evaluate.profile or evaluate.feature_importance

• v0.2.4 (Aug 4, 2021)

  • supports tables named as restricted Snowflake keywords (e.g. ACCOUNT, CASE, ORDER) to be registered as Rasgo Sources

• v0.2.3 (July 30, 2021)

  • introduces publish.features_from_source_code() function. This function allows users to pass a custom SQL string to create a view in Snowflake using their own code. This function will: register a child source based off the parent source provided as input, register features from the new child source table.
  • introduces new workflow to publish.source_data() function. Pass in source_type="sql", sql_definition="<valid sql select string>" to create a new Rasgo DataSource as a view in Snowflake using custom SQL.
  • makes the features parameter optional in publish.features_from_source() function. If param is not passed, all columns in the underlying table that are not in the dimensions list will be registered as features
  • adds trigger_stats parameter to all publish method. When set to False, statistical profiling will not run for the data objects being published. Default = True
  • adds verbose parameter to all publish methods. When set to True, prints status messages to stdout to update users on progress of the function. Default = False.
  • introduces .sourceCode property on Rasgo DataSource and FeatureSet classes to display the SQL or python source code used to build the underlying table
  • introduces .render_sql_definition() method on Collection class to display the SQL used to create the underlying collection view
  • introduces .dimensions property on Rasgo Collection class to display all unique dimension columns in a Collection
  • introduces trigger_stats parameter in collection.generate_training_data() method to allow users to generate a sql view without kicking off correlation and join stats. Set to False to suppress stats jobs. Default=True.
  • Add support for optional catboost parameter train_dir in evaluate.feature_importance() function, which allows users to dictate where temporary training files are generated

• v0.2.2(July 14, 2021)

  • Allow for consistency in evaluate.feature_importance() evaluation metrics for unchanged dataframes
  • Allow users to control certain CatBoost parameters when running evaluate.feature_importance()

• v0.2.1(July 01, 2021)

  • expand evaluate.feature_importance() to support calculating importance for collections

• v0.2.0(June 24, 2021)

  • introduce publish.experiment() method to fast track dataframes to Rasgo objects
  • fix register bug

• v0.1.14(June 17, 2021)

  • improve new user signup experience in register() method
  • fix dataframe bug when experiment wasn't set

• v0.1.13(June 16, 2021)

  • intelligently run Regressor or Classifier model in evaluate.feature_importance()
  • improve model performance statistics in evaluate.feature_importance(): include AUC, Logloss, precision, recall for classification

• v0.1.12(June 11, 2021)

  • support fqtn in publish.source_data(table) parameter
  • trim timestamps in dataframe profiles to second grain

• v0.1.11(June 9, 2021)

  • hotfix for unexpected histogram output

• v0.1.10(June 8, 2021)

  • pin pyarrow dependency to < version 4.0 to prevent segmentation fault errors

• v0.1.9(June 8, 2021)

  • improve model performance in evaluate.feature_importance() by adding test set to catboost eval

• v0.1.8(June 7, 2021)

  • evaluate.train_test_split() function supports non-timeseries dataframes
  • evaluate.feature_importance() function now runs on an 80% training set
  • adds timeseries_index parameter to evaluate.feature_importance() & prune.features() functions

• v0.1.7(June 2, 2021)

  • expands dataframe series type recognition for profiling

• v0.1.6(June 2, 2021)

  • cleans up dataframe profiles to enhance stats and visualization for non-numeric data

• v0.1.5(June 2, 2021)

  • introduces pip install "pyrasgo[df]" option which will install: shap, catboost, & scikit-learn

• v0.1.4(June 2, 2021)

  • various improvements to dataframe profiles & feature_importance

• v0.1.3(May 27, 2021)

  • introduces experiment tracking on dataframes
  • fixes errors when running feature_importance on dataframes with NaN values

• v0.1.2(May 26, 2021)

  • generates column profile automatically when running feature_importance

• v0.1.1(May 24, 2021)

  • supports sharing public dataframe profiles
  • enforces assignment of granularity to dimensions in Publish methods based on list ordering

• v0.1.0(May 17, 2021)

  • introduces dataframe methods: evaluate, prune, transform
  • supports free pyrago trial registration

• v0.0.79(April 19, 2021)

  • support additional datetime data types on Features
  • resolve import errors

• v0.0.78(April 5, 2021)

  • adds include_shared param to get_collections() method

• v0.0.77(April 5, 2021)

  • adds convenience method to rename a Feature’s displayName
  • adds convenience method to promote a Feature from Sandbox to Production status
  • fixes permissions bug when trying to read Community data sources from a public org

• v0.0.76(April 5, 2021)

  • adds columns to DataSource primitive
  • adds verbose error message to inform users when a Feature name conflict is preventing creation

• v0.0.75(April 5, 2021)

  • introduce interactive Rasgo primitives

• v0.0.74(March 25, 2021)

  • upgrade Snowflake python connector dependency to 2.4.0
  • upgrade pyarrow dependency to 3.0