Immersion Analytics supports up to 18 dimensions simultaneously for each data point, providing comprehensive insight into multi-dimensional data. Mathematically, this means you’re folding up to 3,060 unique scatterplot combinations into a singal view, a massive cognitive upgrade.
This is accomplished by rendering numerous special effects onto each data point, where the intensity of each effect conveys an added numeric dimension. We call this patented technology the Dimensional Engine™. It’s made intelligible by incrementally layering dimensions as special effects using the eye icons on the legend, one at a time, in a process we call Stepwise Storytelling™.
First, consider this analogy for how your perspective expands even when expanding from just 2 to 3 dimensions, (play the 37sec video),
From the 2D perspective, that looks like the classic game Tic Tac Toe is being played and red seems to have won. Add just one more dimension and you see that reality is more complex — and blue is about to win. By oversimplifying data to fit the limit of plots on dashboards, this is analogous to why you’ve been forming inaccurate conclusions and missing key insight without even knowing.
How do people solve this today?
Even the simplest spreadsheet with more than four columns exceeds what you can plot on X-Y-size-color axes of a single plot. That’s why practitioners apply the following techniques trying to understand multidimensional data,
- Design BI-style dashboards based on a-priori assumption of which factor interactions matter and which can be ignored.
- Consider an 18 column dataset for which you’ve created a dashboard with 10 plots based on assumptions you’ve made.
- There are (18 Choose 4) = 3,060 unique, unordered combinations of X-Y-size-color plots. This math has been verified by MIT professor & MacArthur fellow Dr. Erik Demaine, advisor to Immersion Analytics. Any of these combinations may uniquely reveal insight not offered by the others.
- Said differently, your odds of revealing the insight on a 10 plot dashboard are 1 in (3,060 / 10) = 1 in 360.
- It’s actually worse because you’re also blind to interaction between more than four dimensions.
- Dashboards are great when you know what you want to look at, however, offer poor information coverage. It’s a pervasive root cause of unknown unknowns and mistakes.
- Apply AI or machine learning to the problem. We love AI & machine learning, however, this also has limits
- Inherently fails to consider factors not codified in the training dataset (i.e. lacks “commonsense” and domain experience)
- Ineffective for nonstationary problems where the “rules of the game” change over time (e.g. financial markets, cybersecurity)
- Hard for regular businesspeople to apply their subject-matter expertise to understand & govern the model
- Dimensionality Reduction techniques e.g. principal components analysis (PCA) or t-distributed stochastic neighbor embedding (t-SNE).
- These will generate your desired number of synthetic axes, say X, Y and Z from an arbitrary set of input dimensions
- However, X becomes a synthetic combination of a subset of the list, as does Y and Z.
- That’s why visualizing dimensionally-reduced outputs forfeits your ability to connect with each axis as a dimension of your data you intuitively understand.
- Said differently, it forefits your ability to apply commonsense to the data, so it misses the point entirely.
- Pair Plots, the number of combinations of which grows geometrically as columns are added to the dataset. These also require the analyst to recall across time to form connections between disparate plots — can you remember what you saw on the 5th plot when looking at the 2075th one?
- Rely on summary statistics like correlation, applied to multivariate data, which miss outliers & exception cases almost always worth considering.
Fortunately, there’s now a better way.
Learn more about the Dimensional Engine™.
