intendednull/code-jam-5

Python Discord's 5th code jam

code-jam-5's Issues

Handle cases when input is given without using autocomplete

The current implementation relies on autocomplete to provide coordinates. A user can type something without using autocomplete, but a search in such case would currently result in a 400.

One solution is to fall back to using the given name to search for a city when coordinates are not present. Another is to possibly force the top result in autocomplete to be picked, but this may end up making the wrong choice.

Caching system to lighten load on Azavea API

Substitute for Google location search bar

Graphing

Presentation

Graphs should be interactive, so matplotlib is out of the question I guess since it only creates static images. Each indicator could have its own scatter plot graph with a best fit line. I was considering doing a curve instead, but that may not align well with the whole % changed thing we have going on which is based on a linear regression.

I'm not sure how to present all the indicators; the current approach will end up taking a lot of vertical space with the graphs added. Maybe have a a tab for each indicator?

Implementation

I was looking at plotly.js. It seems really simple to set up, as can be seen in this example. I'm not sure what kind of format it wants the data in. It supports things like pandas dataframes but all we currently use are two numpym arrays, one for each axis.

A better way to create the azavea client session

I'm not sure the lazy load is avoidable. Unless we make create_app async, lazy loading seems like the best way to do it.

Originally posted by @Zer0897 in https://github.com/_render_node/MDIzOlB1bGxSZXF1ZXN0UmV2aWV3VGhyZWFkMTg0MTk1MjE1OnYy/pull_request_review_threads/discussion

Error: Resize must be passed a displayed plot div element.

Getting this from plotly.js. Not sure what it is yet. Basic research reveals it may be related to the graphs being responsive. They do still seem be responsive so maybe this error can just be ignored.

Is there an inaccuracy in how the top indicators are determined?

The top indicators are determined by doing linear regressions on the indicators and comparing their slopes. But the indicators use different data in different units. In one unit a change of 0.1 could be significant but that wont compare well to something that has a change of 1 (slope of 0.1 vs slope of 1).

Also, here are plots of heat_wave_incidents and total_precipitation respectively:

They show an ~8.9% increase and ~0.39% decrease respectively.