Scikit-learn working with example data sets

At the start of machine learning examples you will need some example data sets. Scikit-learn provides few good examples, here is more what it provides and how to use it.

Scikit-learn example data sets

Scikit-learn comes with number of existing example data sets you can use when starting with machine learning. This is how you can find them:

>>> from sklearn import datasets
>>> dir(datasets)

You can see three type of functions:

1. make_<SOME_DATA_SET> - Sample generators. Generate samples of synthetic data sets.

2. load_<SOME_DATA_SET> - Load local data sets which come with Scikit-learn. These are parts of python sklearn package.

   All available:

   • boston_house_prices.csv
   • breast_cancer.csv
   • diabetes_data.csv.gz
   • diabetes_target.csv.gz
   • digits.csv.gz
   • iris.csv
   • linnerud_exercise.csv
   • linnerud_physiological.csv
   • wine_data.csv

3. fetch_<SOME_DATA_SET> - Load external data set, download them from Internet. These are too bit to include them into Python package and some of them include multiple binary files.

   All available:

   • fetch_20newsgroups
   • fetch_20newsgroups_vectorized
   • fetch_lfw_pairs
   • fetch_lfw_people
   • fetch_mldata
   • fetch_olivetti_faces
   • fetch_species_distributions
   • fetch_california_housing
   • fetch_covtype
   • fetch_rcv1
   • fetch_kddcup99
   • fetch_openml

   And this is how you can fetch some data:

   >>> california_housing = datasets.fetch_california_housing()
   Downloading Cal. housing from https://ndownloader.figshare.com/files/5976036 to /home/radek/scikit_learn_data
   

Scikit-learn examples data format

This is how you can load the data:

   >>> bp = datasets.load_boston()
   >>> dir(bp)
   ['DESCR', 'data', 'feature_names', 'filename', 'target']
   >>> type(bp)
   <class 'sklearn.utils.Bunch'>
   

Where sklearn.utils.Bunch is a dictionary-like object that exposes its keys as attributes.