For help using the new tools, check out the New Workspaces User Guide!
Downsampling is a method of reducing the amount of samples in timeseries data down to a more manageable size. When tuned accurately, it looks on par with the non-downsampled data especially at longer timescales while requiring significantly less datapoints.
Why do we show downsampled data?
Culture reactors frequently samples online data from its sensors – over the course of a 72 hour experiment, a single measurement could have over 100,000+ data points. This enables us to provide high resolution data even when analyzing a narrow window of data for a given process. However, this does come at a cost, particularly when trying to plot large datasets from multiple reactors. The benefits of high data resolution are also lost when viewing more than a few hours worth of data.
To maximize page loading performance with minimal losses in data, Culture downsamples the data coming out of the reactors. Our downsampling approach is designed to maintain data accuracy and consistency so most deviations are apparent when analyzing data. This allows us to load plots smoothly without missing out on critical process trends. See below for a side-by-side of downsampled and non-downsampled data.
The loss of data resolution from downsampling are not very pronounced on the hours scale, and even less so on the days scale. Below is an example for a 4 hour window showing dissolved oxygen data. The top graph shows downsampling enabled and the bottom one shoes the same data with downsampling turned off.
1a. 4 hour window of DO% with downsampling enabled
1b. 4 hour window of DO% with downsampling disabled
How is downsampling calculated?
Downsampled data is calculated using two minute windows of data. Using data in the two minute window, we calculate the median of the measured values and the median of the time points. These medians are displayed as the measurement value and timepoint in downsampled data.
Why is there a toggle for downsampling?
Downsampling greatly improves performance at a cost of some data resolution. However, we recognize that there can be a need to see every possible data point in a narrow time range. For instance, it is useful to see live data with the maximum resolution when trying to troubleshoot a process excursion. As such, we now allow downsampling to be toggled on and off.
Why is the downsampling toggle sometimes grayed out?
To prevent performance issues, downsampling can only be turned off for smaller datasets. The toggle is disabled if the time range is too large, there are too many topics to plot, and/or there are too many bays to plot.
2a. Change the x-axis to a more narrow time range
2b. Lower the number of runs selected
2c. Reduce the amount of topics plotted
