#49 Your technical skills are obsolete: now what?

Python Bytes - A podcast by Michael Kennedy and Brian Okken - Lunedì

Categorie:

Tecnica

Sponsored by DigitalOcean. They just launched Spaces, get started today with a free 2 month trial of Spaces by going to do.co/python

Brian #1: Conference videos for DjangoCon 2017 and PyGotham 2017

Michael #2: Python 3.6.3 released on Tue. All machines at FB are already running it (3 days)

  • Tweet: Did you hear that 3.6.3 was released on Tue? How about that all machines at FB are already running it? Over 36.3% of our Python apps are 3.6 via @llanga
  • See Jason Fried’s presentation on culture: Rules for Radicals: Changing the Culture of Python at Facebook
  • More Python 3 news
    • Ubuntu 17.10: “Python 2 is no longer installed by default. Python 3 has been updated to 3.6.”
    • PSA: Python 3.3 is end-of-life in 2 days. Are you prepared?

Brian #3: Your technical skills are obsolete: now what?

  • by Itamar Turner-Trauring
  • We’re big proponents of keeping your skills current, of learning new techniques and technologies. But how does that fit in with life and work.
  • This article is an opinion of how to work on new skills while at work, do it quickly, and look good to your manager.
  • It starts with a good discussion of real business reasons why some projects use older technology. Basically, cost vs benefit of change.
  • Steps to be part of the solution:
    • Identify obsolete and problematic technologies.
    • Identify potential replacements.
    • Get management buy in to get resources (you) to do a pilot project exploring new technology.
  • This process will help you be better at identifying problems, even if you don’t get approval to fix it.
  • He ends with a comment that if you don’t get approval, all is not lost, you have skills to apply to a new job.
  • I’d like to make sue you do a few more steps before giving up and looking for a new job. Before you consider a move to a new team or company, I think…
    • You should give your manager the benefit of the doubt and use this to start a conversation. Make sure you understand their reasons for saying no.
    • Make sure you are not proposing too much time taken away from your primary role in the company.
    • State that you want to improve your skills by providing value for the team and the company.
    • Is the “no” due to just bad timing? Is there a higher priority problem to work on?
    • You’ve just shown that you are someone interested in keeping your skills sharp and helping the company by expanding your role. If you’re still stuck at this point, then consider a move but also, …
  • Read this:
    • Team Geek: A Software Developer's Guide to Working Well with Others - Brian Fitzpatrick
    • Especially: - pg 117 : “Offensive vs Defensive work”. 50-70% of your time at work needs to be focused on creating new value for your company or your customers. No more than 30-50% on repaying technical debt. (Okken: Limit your process improvement / new technology exploration to no more than 10-20%, but try to never drop it below 5% of your time) - pg 113 : “It’s easier to ask for forgiveness than permission.” This is a fine line between doing the right thing and doing something you can get reprimanded for. Use good judgement. - Forgotten page number: A big part of your job is making your boss’s job easier and making your boss and your team look good.

Michael #4: Visualizing Garbage Collection Algorithms

  • By Ken Fox
  • Follow up from the excellent deep dive article in GC from Brian
  • Most developers take automatic garbage collection for granted.
  • It’s very difficult to see how GCs actually work.
  • GCs visualized (click on each image):
    • Cleanup At The End: aka No GC (e.g. Apache web server creates a small pool of memory per request and throws the entire pool away when the request completes)
    • Reference Counting Collector (e.g. Python’s first pass GC, Microsoft COM, C++ Smart Pointers. Memory fragmentation is interesting)
      • The red flashes indicate reference counting activity. A very useful property of reference counting is that garbage is detected as soon as possible — you can sometimes see a flash of red immediately followed by the area turning black.
    • Mark-Sweep Collector (e.g. is this Python’s secondary collector? Probably is my guess)
      • Mark-sweep eliminates some of the problems of reference count. It can easily handle cyclic structures and it has lower overhead since it doesn’t need to maintain counts.
    • Mark-Compact Collector (Oracle’s Hotspot JVM’s tenured object space uses mark-compact)
      • Mark-compact disposes of memory, not by just marking it free, but by moving objects down into the free space
      • The crazy idea of moving objects means that new objects can always just be created at the end of used memory. This is called a “bump” allocator and is as cheap as stack allocation, but without the limitations of stack size.
    • Copying Collector, aka Generational GC
      • The foundation of most high-performance garbage collection systems

Brian #5: pathlib — Filesystem Paths as Objects

  • from Doug Hellman’s PyMOTW-3
  • pathlib was introduced with Python 3.4
  • If you need to work with the file system in Python, you should be using pathlib.
  • Doug’s article is a really good overview.
  • Features
    • Building paths with overloaded / operator
    • Parsing paths with .parts, .parents, .suffix, .stem
    • Concrete paths such as Path.home(), Path.cwd()
    • Getting directory contents with .iterdir()
    • Pattern matching with .glob() and .rglob()
    • Reading and writing files with path objects.
    • Working with directories and symbolic links
    • File properties, permissions
    • Deleting files and directories
  • see also

Michael #6: LUMINOTH: Open source Computer Vision toolkit

  • Deep Learning toolkit for Computer Vision
  • Supports object detection and image classification, but are aiming for much more.
  • It is built in Python, using TensorFlow and Sonnet (Google’s Deep Learning framework and DeepMind’s graph library)
  • Easily train neural networks to detect and classify objects with custom data.
  • Use state of the art models such as Faster R-CNN (Region-based Convolutional Neural Networks)
  • Comes with GPGPU support
  • Simple training
    • Train your model by just typing lumi train. Do it locally or using Luminoth's built-in Google Cloud Platform support to train in the cloud.
    • Once training is done, you can use our Tensorboard integration to visualize progress and intermediate results.
  • Are also working on providing pre-trained checkpoints on popular datasets such as Pascal VOC2012

Bonus article:

The Cleaning Hand of Pytest - My experiences with different approaches to testing

  • by Wiktor Żurawik
  • Two case studies of having to use unittest after using pytest
  • Be sure to check out the “useful links” at the end of the article.

Our news

Visit the podcast's native language site