Repository navigation
Add MLPipeline as a primitive #226
Description
Create the mlblocks.MLPipeline primitive to allow setting a whole pipeline as a single primitive.
The primitive arguments will be the mlblocks.MLPipeline arguments, which means that the pipeline will be specified as either:
- A pipeline name
- A json path
- A pipeline dictionary
- Individual arguments
Some usage examples:
Python pipeline specification - specify the sub-pipeline by its name
from mlblocks import MLPipeline
primitives = [
'mlblocks.MLPipeline',
'xgboost.XGBClassifier'
]
init_params = {
'mlblocks.MLPipeline#1': {
'pipeline': 'my-preprocessing-pipeline'
},
}
pipeline = MLPipeline(primitives, init_params=init_params)
Pipeline dictionary - specify the subpipelines by name, dict and json path
pipeline_dict = {
'primitives': [
'mlblocks.MLPipeline',
'mlblocks.MLPipeline',
'mlblocks.MLPipeline',
],
'init_params': {
'mlblocks.MLPipeline#1': {
'pipeline': 'a-preprocessing-pipeline',
},
'mlblocks.MLPipeline#2': {
'primitives': [
'mlprimitives.feature_extraction.CategoricalEncoder',
'sklearn.preprocessing.StandardScaler',
],
},
'mlblocks.MLPipeline#3': {
'primitives': [
'path/to/my/postprocessing/pipeline.json',
],
},
},
}
Activity