Python comes equipped with a large library of packages for machine learning tasks.
The packages we will look at in this chapter are as follows:
- The IPython console
- NumPy, which is an extension that adds support for multi-dimensional arrays, matrices, and high-level mathematical functions
- SciPy, which is a library of scientific formulae, constants, and mathematical functions
- Matplotlib, which is for creating plots
- Scikit-learn, which is a library for machine learning tasks such as classification, regression, and clustering
There is only enough space to give you a flavor of these huge libraries, and an important skill is being able to find and understand the reference material for the various packages. It is impossible to present all the different functionality in a tutorial style documentation, and it is important to be able to find your way around the sometimes dense API references. A thing to remember is that the majority of these packages are put together by the open source community. They are not monolithic structures like you would expect from a commercial product, and therefore, understanding the various package taxonomies can be confusing. However, the diversity of approaches of open source software, and the fact that ideas are being contributed continually, give it an important advantage.
However, the evolving quality of open source software has its down side, especially for ML applications. For example, there was considerable reluctance on behalf of the Python machine learning user community to move from Python 2 to 3. Because Python 3 broke backwards compatibility; importantly, in terms of its numerical handling, it was not a trivial process to update the relevant packages. At the time of writing, all of the important (well important for me!) packages, and all those used in this book, were working with Python 2.7 or 3x. The major distributions of Python have Python 3 versions with a slightly different package set.
Python is a versatile general purpose programming language. It is an interpreted language and can run interactively from a console. It does not require a compiler like C++ or Java, so the development time tends to be shorter. It is available for free download and can be installed on many different operating systems including UNIX, Windows, and Macintosh. It is especially popular for scientific and mathematical applications. Python is relatively easy to learn compared to languages such as C++ and Java, with similar tasks using fewer lines of code.
Python is not the only platform for machine learning, but it is certainly one of the most used. One of its major alternatives is R. Like Python, it is open source, and while it is popular for applied machine learning, it lacks the large development community of Python. R is a specialized tool for machine learning and statistical analysis. Python is a general-purpose, widely-used programming language that also has excellent libraries for machine learning applications.
Another alternative is Matlab. Unlike R and Python, it is a commercial product. As would be expected, it contains a polished user interface and exhaustive documentation. Like R, however, it lacks the versatility of Python. Python is such an incredibly useful language that your effort to learn it, compared to the other platforms, will provide far greater pay-offs. It also has excellent libraries for network, web development, and microcontroller programming. These applications can complement or enhance your work in machine learning, all without the pain of clumsy integrations and the learning or remembering of the specifics of different languages.