a couple of questions about ttk usage

[Janus Gowda]

Hello,

I am new using ttk and I have a few newcomer questions.

I have managed to install the software and try a few examples from the videos. In the examples, data is loaded in paraview format that seems to be in some sort of mesh format. I wonder how to use the software from data collected from sensors. For example, how can one do typical persistence diagram analysis like in gudhi? I did load some one my data in csv format but I did not know how to proceed from there. Is this software especially designed for data in mesh format? Are there examples for normal sensor data or data in csv, numpy, etc?

The problem is that I do not understand how the data from the examples is formatted or constructed in the first place, so I cannot yet imagine how to use this software for my data analysis applications.

Thank you,

Janus

[Julien Tierny]

Hi Janus,

thanks for your email.

> I wonder how to use the software from data
> collected from sensors.

> I did load some one my data in csv format
> but I did not know how to proceed from there.

I understand that your sensor data comes in the form of a point cloud.
In your CSV file, each line represents a sample and each column represents a dimension. Is that right?

If so, the default strategy would be to construct a Rips complex out of your data. See these two examples:
https://topology-tool-kit.github.io/examples/persistentGenerators_householdAnalysis/
https://topology-tool-kit.github.io/examples/persistentGenerators_periodicPicture/

This construction has a couple of parameters (the maximum edge length and the maximum simplex dimension).
If you know that a specific dimension of your data (i.e. a specific column of your CSV file) is particularly important, you can then consider this dimension as a scalar field and run any TTK algorithm on it (e.g. persistence diagram). If you are more interested in the spatial distribution of your samples in their high-dimensional space, you can consider the function "Diameter" provided by the Rips complex or the Gaussian-based density estimation (checkbox in the Rips complex TTK module).

If your data lives in low dimensions (i.e. the number of columns in your CSV file is lower or equal to 3), then you have more options (for instance, you can compute a Delaunay triangulation or estimate the density directly on a regular grid). If your data lives in high dimensions, you also have the possibility to reduce the dimensionality of your data (see the DimensionReduction TTK module) and then run the topological analysis in low dimensions. Sometimes it provides interesting results.

I hope this helps.

Best,
--
Dr Julien Tierny
CNRS Researcher
Sorbonne Universite
https://julien-tierny.github.io/