The Platform Wasn't the Problem

Turns out the platform wasn't the cause of the wobbliness of the data. It has continued and arguably gotten worse.

It looks like there's almost a periodic quality to the wobbliness. 

Here's what I'm going to do:

• Poll the sensor a few times in rapid succession and average the results.
• Remove the big box fan right next to the setup.
• Hope really hard that one of those works.

Sharing Data

I'm sharing the data that went into this graph in the cleverly-named "data.zip" file found on the main project page. It contains two files: auto.csv and manual.csv. Each contains datetime iso strings (created using datetime.isoformat() in Python) and corresponding values in ounces.

Current problems with the data:

• The early manual weight data didn't capture the time, but I can tell you that it was usually the morning, around 9 am.
• There are large swaths of time in the automatically-recorded data where there's no data. I didn't realize the sensor was down (seems to go down regularly, need to figure out why). Now I have a script that checks whether the sensors are reporting properly and emails me if they aren't.
• Every time the automated sensor goes down, taring is the first step on boot. The next code revision will store one tare weight so this won't get regularly reset.
• Over the course of this time period, I pruned both the roots and the shoots. I'm not sure when that happened, but it should obviously affect the weight.