Lesson 5: Switching cockpits (example: Correlations)

Visplore offers more cockpits than "Trends and Distributions". Each cockpit is a ready-to-use tool for a particular type of information, or particular task.

You can change cockpits at any time. Any currently defined focus, filters, and all other objects you made are kept and are available in the other cockpits as well.

If it is not shown already, click on the gray vertical "Choose cockpit" bar on the left edge of Visplore to open the list of available cockpits. Depending on your version of Visplore, there may be more cockpits:

Double-click on the cockpit "Correlations". In case a dialog "Correlations - Role assignment" is shown, just press OK.

The cockpit opens and looks like the following image. If you just switched over from another cockpit, and you have a focus or filter defined, clear them for now (press the small "x" symbols next to the words "Filter" and "Focus" in the top of Visplore.)

The "Correlations" cockpit is about finding correlating pairs of time series, as well as finding time series that correlate most with a specific target variable of interest. See the "Cockpits" section of the documentation to learn all details about this cockpit. Here, you learn some basic interactions, and using the cockpit in combination with other cockpits.

The "Overview" section in this cockpit initially shows a matrix of "Pairwise Correlation". Here, the first 25 time series are correlated with each other. Each pair of two time series is shown as a small plot in the matrix. Per cell, the names of the paired time series are stated above the cell, and in the right of the cell. The background color of a pair shows the correlation between the two time series: red means positive correlation, blue means negative (=inverse) correlation, white means they are independent. ">P" means that a pair does not pass a significance test, see the cockpit's detailed description

.

Hover some of the cells to get a detailed description of that pair as a tooltip window.

Click on cells to see the scatter plot of the clicked pair enlarged in the "Scatter Plot" view in the upper right.

Filter the matrix of displayed time series to temperature time series only, by typing "temp" in the field next to "Filter:" above the matrix.

Then, click the pair "Temperature_Outdoor_BrightCounty_Weather" and "Temperature_Indoor_BrightCounty_Weather". Visplore then looks like this:

The "Scatter Plot" shows that this indoor and outdoor temperature sensor at the location BrightCounty are highly correlated, which is not surprising. However, there are several points in the middle of the visualization, that do not lie on the generally correlated point cloud. Let's inspect them in detail.

To select them properly, switch selection mode by clicking the view title "Scatter Plot", then "Selection mode", and then select the Lasso option (rightmost orange symbol).

With the Lasso tool, circle the anomalous points by dragging the left mouse button, approximately like this:

As a result of selecting these records, the correlation matrix immediately updates to consider only the points in focus. More importantly in this case, the "Time Series" visualization highlights the selected points as well using colors, which helps us to localize them in a temporal context.

Zoom in to the first two occurrences in 09.2014 by dragging a rectangle around them with the right mouse button:

This view is interesting: the indoor temperature sensor seems to have some kind of oscillation, while the outdoor sensor recorded values regularly.

Zoom out again by clicking the button in the lower left corner of the "Time Series" view (see image above). Then, zoom in to another occurrence, to discover that the same pattern occurs several times.

To mark this finding for later, we can label the currently selected data records as a named condition, which we can use later on, even when switching to other cockpits.

Make sure your Lasso selection is still in place as your focus, then press the "name" button in the focus bar (see image below). Choose "Condition", type the name "Oscillation" for the condition, and press OK to label it.

Labeling data by creating named conditions is an important use case of Visplore. You can use the labels for further analysis in Visplore, or export them for downstream tasks in other tools, like Excel or Python.

Hover the mouse over the named condition "Oscillation" to highlight these records in all views.

You can also click the orange "Oscillation" area in the "Conditions" bar, to do many other things with this condition. For example, you can try "Add as variable" to make the condition available as a 1/0 column that can be analyzed like any other numerical data attribute - e.g. for visulization.

Switching back to another cockpit

Now let's switch back to the "Trends and Distributions" cockpit, to investigate further if other sensors from the location "Bright County" have similar anomalies.

Click the vertical gray bar "Choose Cockpit" at the left border of Visplore. Then, double-click the cockpit "Trends and Distributions". If a dialog named "Trends and Distributions - Role assignment" appears, just press OK.

The cockpit opens, possibly looking the same way as you left it before. But this time, note that we have taken along the named conditions, and possibly also a focus and filter from the Correlations cockpit.

Now clear any focus and filter that you may have (press "x" near Focus and Filter, if any).

Click the "Horizon graph" view in the "Overview of variables" section, and type "Bright county weather" in the "Filter variables by name" field to see only time series of the Bright County location. Visplore should look like this:

To see if oscillation events happen in more sensors at once, we need to zoom in a bit.

Click the "Oscillation" condition, then choose "Put in focus" to focus on the records of the condition.

Make sure, the time series "Temperature_Indoor_BrightCounty_Weather" with the oscillations is selected, so that it's shown in the "Time Series" view, and the oscillations are highlighted.

Zoom in to the first bigger oscillation spike by dragging a rectangle with the right mouse button:

Now we want to select the period around this oscillation event. Click the view title "Time Series", then "Selection mode", then select the horizontal interval tool (2nd from left:).

Drag a line in the "Time Series" from left to right, using the left mouse button, making sure to select some time before and after the oscillation as well (you may need to zoom out a bit first to achieve this, using CTRL + mousewheel). Note, how the horizon graph shows all Bright County weather time series of your selected period immediately:

The image shows that the indoor sensor of Bright County is the only one showing an oscillation at that exact time - however, note how the series "Evapotranspiration_Bright_County_Weather" also seems to oscillate, however at a slightly later time. You could now look at other oscillation events in the same way, to find out if this delayed pattern repeats.

Switching cockpits at any time allows to effectively combine the interaction and visualization tools, like we did in this example: from correlation anomalies to an overview of other possibly affected sensors.

Well done! You have mastered the concept of combining cockpits to address your complex tasks effectively! :)




>> Continue with Lesson 6: Compute new data attributes




License Statement for the Photovoltaic and Weather dataset used for Screenshots:
"Contains public sector information licensed under the Open Government Licence v3.0."
Source of Dataset (in its original form): https://data.london.gov.uk/dataset/photovoltaic--pv--solar-panel-energy-generation-data
License: UK Open Government Licence OGL 3: http://www.nationalarchives.gov.uk/doc/open-government-licence/version/3/
Dataset was modified (e.g. columns renamed) for easier communication of Visplore USPs.