Are you a scientist who programs for a living, but have never studied how to write software well?
Professional software developers have many tools at their disposal to program quickly, accurately, and in large groups. These tools are equally, if not more, important to scientists who most deliver correct results on dynamic projects, yet many scientists have never been introduced to tools to make their job easier.
This course is designed for practicing scientists who want quick techniques to make their jobs easier. Lectures are focused on practicality, not purity, but students will get the necessary background to consult professional resources available on the Internet to learn more.
Topics which will be covered:
These techniques will not be taught as part of one particular language or development model, but as techniques applicable to any project.
The target audience is new PhD students in computational (not computer) sciences and above, who have not been exposed to these topics before.
Students should already program in some language, and have serious preexisting projects as part of other work or studies. Most course examples will be in Python or C, but these languages are not required. Students will be expected to learn specifically how to apply these techniques to whatever language and projects they have.
This course will be to basic for someone who has previously used or has knowledge of most topics listed.
This is a proposed schedule. All classes on Tuesday at 15:00. Links go to previous versions of tutorials, presented to my group last year. Until the lecture, they won't be refined.
Internet relay chat (IRC) is a channel-based communication system. There is one channel, where everyone from the class can ask questions to anyone else.
Software is used everywhere in the modern world, especially in science. There are many "computational fields" where work is purely making calculations in code, but also almost all fields software is used for data processing and analysis. Software is the ultimate in reproduciability and verifiability. And most importantly, analysis is less often "off-the-shelf" and scientists must create new tools themselves.
Despite the importance of good software, scientists are rarely trained in its use. Imagine a physicist building a machine without a training in basic electronics: this is many scientists. There are many professionals whose life is managing software. This is not what most scientists want to do, but the fact remains that it is an important part of what we do.
The software of science is not just a tangential thing. Science must be correct. Science must be fast. Science must be reproducible. Science should be open and shared. In science, there are specific, time-tested techniques which people use to get their answers. People must plan before beginning, keep good records, ensure that results are reproducible, try to generalize, and so on. Likewise, there are specific time-tested techniques of software development. If people add a small amount of structure to their practical work (similar to an an experimentalist writing down parameters before running experiments, there can be a huge improvement in productivity.
But what is this structure? Here are a few examples. Version control tracks history of code. With this, you can retrieve the exact version at any time in the past. You can find old versions or determine exactly what results are affected by a bug. Most importantly, you can collaborate properly. No longer is there multiple versions for each student, or mailing paper revisions back and forth. Everyone does their work, and results can be merged automatically, even if two people edit the same file. Automated software testing can help you find bugs early, and keep them from appearing as your project drags on and on. Debuggers can help you spend less time debugging and more time running. A modular design can help you reuse, instead of recreate, code. And it goes on and on.
The purpose of this course is to help scientists to learn these concepts. It is not designed to be a rigorous software development course, since that would be far too abstract. Instead, it assumes the audience are experienced scientific programmers, and presents techniques one at a time, in independent packages which can be immediately applied to actual projects. The course also isn't designed to teach specific tools, since scientific workflows are incredibly diverse. Instead, it introduces important concepts, and the student's task is to adapt this to their specific workflow. To take a specific example, this course won't say "here is an code editor to use". There are just too many valid options. It will say "here is something your editor should be able to do".
Science is no longer just small codes and independent projects. Science is large, science is interdependent, and science is collaborative. This course will give researches the tools needed to work and contribute in this modern world.
This course is open-source, licensed under the GPLv3, and all materials are available on github and contributions are welcome.