a thoughtful web.
Share good ideas and conversation.   Login or Take a Tour!
comment by kantos
kantos  ·  41 days ago  ·  link  ·    ·  parent  ·  post: Hubski Craft Fair v1.50 - January 16, 2020

Truth be told, I merely know a little about a lot, so I'm not mega-fluent in lingo. I'm assuming ML means machine learning. This is the first time I've seen Keras, and a google search into it looks hella neat. Looking beyond my scope, currently.

What I'm using is solely Python libraries. For API access ('requests' package to retrieve, and 'pandas' for data manipulation) to both that dataset and Google API ('bokeh' for visualization, but 100% planning on something else) to make a nicer map. Those parts are complete thus far. Also, some dynamic querying (meaning you can look at all fireballs within a specific year or month or recurring month/year/decade) rather than just 'from this date backward/forward'.

The following checklist is sorta me cleaning out my brain space, but also goal setting to look back on here.

What I'd like to present in the thread later is a cleaner version:

- not using Bokeh

- show generalized shockwave impact areas per fireball, see: Chelyabinsk:

- be able to share my code without sharing my Google Maps API key (hehe)

- clean up symbology with proper visual scaling per impact energy

Stretch Goals:

- Better hover tool-tips

- Create GUI to dynamic query, rather than going into the code to query

- Efficiently import CSV data into a PostgreSQL database, and pull from there rather than API since the CSVs give velocity components.

I really just wanted to make the map on the site more interactive, and less static.... Also gives me something to present to employers in an application to NASA. O:^)





veen  ·  41 days ago  ·  link  ·  

I might have some tips for ya as I do quite a bit of Python and PostGIS these days. PostGIS is a Postgresql extention that allows you to work with geodata and is the bread and butter of most of my geo algorithems these days.

Look into Seaborn and Plotly for visualizations. Seaborn is easier, but Plotly extends to Plotly Dash which I think is phenomenal and has been on my to-learn list for over a year now. It does require you to understand Python decorators, which I haven't figured out yet. But it means you can load dynamic data into a customizable dashboard without any front end dev, and no matter what job you end up with that's a damned usefull piece of wizardry to have.

My north star in visual, clear design is the Vignelli Canon. Look it up, absorb it, you won't regret it I think.

For the key thing: learn how to use github, save the key in a file, use .gitignore for the file, open(file, r).

kingmudsy  ·  40 days ago  ·  link  ·  

+1 for plotly! It was my bread and butter for data science in college :)

kantos  ·  40 days ago  ·  link  ·  

    I might have some tips for ya as I do quite a bit of Python and PostGIS these days. PostGIS is a Postgresql extention that allows you to work with geodata and is the bread and butter of most of my geo algorithems these days.

You're one of the inspirations for the project.

And, I can confidently speak well to what the heck this means now too:

    Plotly Dash which I think is phenomenal and has been on my to-learn list for over a year now. It does require you to understand Python decorators, which I haven't figured out yet. But it means you can load dynamic data into a customizable dashboard without any front end dev, and no matter what job you end up with that's a damned usefull piece of wizardry to have.

    My north star in visual, clear design is the Vignelli Canon. Look it up, absorb it, you won't regret it I think.

I have to say that using something other than Seaborn and Plotly were explicit design constraints last semester. So, I'm ready to hulk out using those this time around.

    For the key thing: learn how to use github, save the key in a file, use .gitignore for the file, open(file, r).

Sweet lord, thank you. That's been bugging me. Using Github makes it more simple.

Also, found a free version of Vignelli Canon. Woot!

veen  ·  40 days ago  ·  link  ·  

That's wonderful to hear - glad I could inspire you a bit. :)

I taught myself matplotlib as part of my master's thesis, as I had to automate generating almost a thousand graphs to then pick the interesting ones. The good thing is that once you learn some graphing library, the others are usually in a very similar vein. It's almost always structured as "here's a graph object, update the object attributes to change things in the graph". Most of them work really well with Pandas dataframes which is dope. But I recommended the other two over mplib because they're easier to learn, as far as I can tell.

    Also, found a free version of Vignelli Canon. Woot!

I always google that one PDF. Maybe I should get around to actually buying a copy since I like it so much...I do have some giftcards lyin' around somewhere.