Save trainned model

• commonly used name model.bin

• f_out to tell that this is the output file

• pickle.dump → save

• we use pipeline beacause

  • if we save dv, model separately we need to state it when importing → keep the whole process in pipeline

  from sklearn.pipeline import make_pipeline
  
  pipeline = make_pipeline(
      DictVectorizer(),
      LogisticRegression(solver='liblinear')
  )
  

with open('model.bin', 'wb') as f_out:
    pickle.dump(pipeline, f_out)

Loaded the model back

with open('model.bin', 'rb') as f_in:
    pipeline = pickle.load(f_in)

Convert ipynb to python script

## it should work in github code space

jupyter nbconvert --to=script notebook_name.ipynb 

• split script to → train.py, predict.py

Fast API