Basic Data Science with scikit-learn

Planetary Examples

VO and Planetary Mapping Workshop 1-3 July 2019

Data Science

Data science is a multi-disciplinary field that uses scientific methods, processes, algorithms and systems to extract knowledge and insights from structured and unstructured data.

https://en.wikipedia.org/wiki/Data_science

Focus on data wrt methods

Data Science approaches

Statistics
'Classical' Machine Learning
Deep Learning

Does still have sense?

Yes (particularly for vector data)

Data discovery tool
Better interpretability
Useful for pre-processing of DL applications
No mandatory GPU

Scikit-learn

https://scikit-learn.org/

Open source, commercially usable - BSD license
Python
Built on NumPy, SciPy, and matplotlib

Pandas

https://pandas.pydata.org/

A data manipulation library

Seaborn

https://seaborn.pydata.org/

A data visualisation library

Resources

Scikit-learn documentation
Scientific Python Introduction, scikit-learn chapter by Gael Varoquaux et al.
Introduction to Machine Learning with Python by Andreas Mueller
Hands-On Machine Learning with Scikit-Learn and TensorFlow by Aurélien Géron
Python Data Science Handbook by Jake Vanderplas
...

Learning Problems Categories

Supervised Learning
- Classification (discrete output values)
- Regression (continuous output values)
Unsupervised Learning
- Clustering (discover groups)
- Manifold Learning (discover the 'true' dimensionality of a distribution)
- ...

Data structure

Data as tables, catalogs ie vector information

Rows are catalog entries, granules, features (in GIS nomenclature), samples

Columns are observables, attributes (in GIS nomenclature), input parameters, features (in scikit-learn glossary)

Predictions are labels, real values, targets

X = input_data[n_samples,n_attributes]

					Y = target_data[n_samples,n_targets]

API structure

Estimators (algorithm classes) share common basic methods

fit: it takes some samples X, targets y if the model is supervised, validate the input data and estimate and store model attributes from the estimated parameters and provided data.
predict: It makes a prediction for each sample, taking X as input (in a classifier or regressor
transform: if the estimator is a transformertransforms the input, usually only X, into a transformed space.

pipeline: pipeline of transforms with a final estimator.