Data exploration is an important initial step in any research project, where you might examine your data by creating many plots, select different regions in parameter space to investigate, and explore how different datasets relate to one another. The Glue visualization application (https://glueviz.org/) is an easy, but powerful, way to explore a variety of data types across multiple disciplines without any coding. In this workshop, I will provide an introduction to the basic functions of Glue through hands-on exercises.
Prior to the workshop please install Glue on your own computer; instructions can be found on this workshop's GitHub repository : https://github.com/ageller/IntroToGlue. I will provide a handful of datasets, though I also encourage you to bring your own data to explore with Glue. There are no other prerequisites for this workshop.
The easiest installation method is to download the standalone executable file from the Glue website. See here for instructions. This version lacks various useful plugins (by default), but you can attempt to install them afterward you install Glue. To do so, open the Glue application on your computer and click on the "Plugins" menu in the top bar and select "Plugin Manager". In this workshop I will demo the glue-geospatial plugin; I recommend that you install that on your Glue version as well. You may choose to install other plugins as you so desire.
I have sometimes noticed strange behavior on various machines when installing plugins using the method above, and not all plugins appear to be available on every OS through the standalone executable. An alternative is to install Glue using conda or using pip, preferably through miniforge. On my Windows laptop, I have had most luck using the pip installation method. I recommend creating a new virtual environment for Glue, which I will do with conda.
$ conda create -n glueviz-env python=3.9
$ conda activate glueviz-env
$ pip install glueviz[all,qt] glue-geospatial glue-plotly pyQt5
For Mac users, you may have to remove the [all,qt] portion of the pip command above.
If you encounter any trouble installing the plugins (and probably the issue would be with the geospatial plugin), that is OK! You can still work through nearly all of this workshop with the default Glue installation you get when downloaded the executable file.
If you downloaded the standalone executable for Glue, you can launch it by double clicking. If you installed Glue with conda on the command line, you can launch Glue by typing.
$ glue
With Glue open, you can drag any dataset (e.g., a .csv file) into the main Glue window to load the data into the app. Alternatively, you can click on the "Import Data" button at the top of the app. Once you have loaded the data into Glue, you should see it on the left-hand side of the app within the "Data" menu under "Data Collection". You can load multiple datasets into Glue, and have access to all of them within the "Data" menu. If you double click on a dataset, you can change the color, label and default point size.
Exercise: Start by loading in any of the .csv files from the data/ directory in this repo. Additionally, if you have installed the glue-geospatial addon (see above), you may want to load in the ChicagoGeoTIFF.tif map file.
Once your data is loaded into Glue, you can can create a plot by clicking on the data object within the "Data" menu and dragging it into the main window. This should bring up a dialogue box that asks what kind of plot you want to create (e.g., 2D scatter, histogram, etc.). After you select the type of plot, it will appear in the main window of the app. You can create multiple plots of the same dataset, and also create plots using different datasets, all within the same workspace.
At the top of each plot window, there are buttons to allow you to manipulate the plot. If you hover over these buttons, you will see what they do.
On the left panel of Glue, you will also see the the "Plot Layers" and "Plot Options" boxes have been populated. Use the "Plot Layers" section to control the color, opacity and other characteristics of the figure. Use the "Plot Options" section to change the data that is being plotting, the axes limits, etc.
Exercise: Create a variety of plots from the .csv file(s) that you loaded into Glue, and explore all the different plotting options.
At the top of each plot window there are various ways to select data on your plot (these buttons have red dots with blue regions). If you click on one of those buttons and then click+drag across your plot to define a region, you will create a new subset of your data. This subset will appear in the "Plot Layers" panel on the left side of the app. You can now create additional plots showing only this subset by dragging that dataset into the main window (like you did to create the initial plot).
At the top of the app, you will see different selection modes (red circles). Hover over these to see what each refers to, and try them out.
Exercise: Make a selection on one of your datasets to create a Subset. Then drag this Subset into the main portion of the app to create a new plot. Then change the selection mode to intersection, and make a new selection on your Subset.
If you have data with similar attributes, you can link them together. For example, multiple datasets in this repo have latitude, longitude and date attributes. If we link these together, we can plot multiple datasets (from different files) on top of each other on one figure within Glue. In order to link data together, click on the "Data Manager" menu and select "Link Data". This will bring up a new window where you can tell Glue which attribute from one dataset should be linked to an attribute in another dataset. Select a dataset for "Dataset 1" from the dropdown, and select a different dataset for "Dataset 2". Then within the list of components for each file, select the two attributes that you want to link, and click the "Glue attributes" button.
Now for any linked datasets, you can add an additional dataset onto an existing plot by dragging that dataset directly onto that plot (as opposed to onto the main window). One example where this works well is plotting geographical data (e.g., those with latitudes and longitudes) on top of a map.
Exercise: Create links between your datasets. Then create at least one plot that includes multiple datasets. If you installed the glue-geospatial plugin, plot some data on top of the map. Otherwise, link together two datasets from .csv files that have date columns and plot those on top of each other. Make selections on your new combined plot, and notice that this selection will be populated on other plots where these datasets are visible.
Plots: You can export plots from Glue by clicking on the disk symbol within a given plot window. This will give you the option to download a png ("Save plot to file") or to download a Python script that would reproduce the plot. (If you installed the glue-plotly, which should come with the standalone executable, you should also see an option to download as a Plotly HTML page; if this works you can open the resulting HTML file in your browser by double clicking. But... this is an experimental feature and does not always work :(
Data: You can export data to a file (e.g., .csv file) by clicking on the "Export Data/Subsets" button in the top bar of the Glue app. This will bring up a dialog box that has a dropdown for you to select the dataset, subset (or full dataset), and attributes to save. You also have a few options for file types to download to. This may be particularly useful for downloading a subset that you created by some complex manual selection on a dataset.
Session: You can also save your session so that you can start up where you left off after closing Glue. To do so, you can click on the "Export Session" button in the top bar of the app. Then you will be free to close your Glue session without losing your work. A word of warning: I have seen that these sessions can be version dependent. In other words, if you save a Glue session using on version of Glue, you may not be able to open it with another version of Glue (e.g., one that you may upgrade to in the future). I'm also not sure how well these saved sessions behave across different operating systems. Also, various plot customizations (e.g., axes labels) do not appear to save.
Exercise: Test out all these export options.