Pipelines is an open source automated machine learning training code generation utility. It's intended to help data scientists get a fast start on building new models by providing boilerplate sklearn pipeline training code for a given dataset.
The generated code includes:
- Data preparation and feature engineering
- Missing value imputation
- Standardization
- Cataegorical feature prep
- Text parsing; e.g., TF-IDF
- K-fold cross validation
- Grid search
- Model training
- Model comparison
Training code is generated as follows:
from pypelines.sklearn_pypeline import SupervisedPipeline
my_pypeline = SupervisedPipeline(data = your_dataframe,
target = 'dependent_variable_name',
model_type = 'classification' # or 'regression',
nfolds = 5 # default is 5
)
my_pypeline.get_code()
SupervisedPipeline() can also be passed a list of model names through an optional models parameter. If models is not specified, all available models will be included.
To see a listing of all available models, utilize pypelines.classification_model_list() and pypelines.regression_model_list()
my_pypeline.model_list()returns a list of models to be included in the training code.my_pypeline.grid_search()returns a dictionary containing the grid search parameters for every model to be included in the training code.my_pypeline.grid_search_for_model('modelName')returns a dictionary containing the grid search parameters for every model to be included in the training code.my_pypeline.set_grid_search_for_model('modelName', grid_search_dictionary)returns a dictionary containing the grid search parameters for every model to be included in the training code.my_pypeline.get_code()returns the generated training code.my_pypeline.code_to_clipboard()copies the training code to the clipboard.my_pypeline.code_to_file('/path/to/filename.py')saves the training code to a file.
See currently available models: Regression and Classification