armanmaesumi / torchrbf Goto Github PK

I've setup an interpolator with survey data, to interpolate signal strength of specific wifi APs. The survey data covers the indoor space.

Here is the code setting up scipy interpolator:

    xs = [grid_point.x for grid_point in grid_points]
    ys = [grid_point.y for grid_point in grid_points]
    rssis = [scan['RSSI'] for scan in ap_survey_scans]
    interpolator = Rbf(xs, ys, rssis, function='linear', smooth=20)

Here it is for torchtbf. It runs many orders of magnitude faster with torchrbf 🔥:

    grid_points = torch.tensor([([grid_point.x, grid_point.y]) for grid_point in grid_points])
    rssis = torch.tensor([scan['RSSI'] for scan in ap_survey_scans]).unsqueeze(1)
    interpolator = RBFInterpolator(grid_points, rssis, smoothing=20.0, kernel='linear')

This is the survey data provided, each black dot representing a data point:

This is the results of interpolated signal from an individual AP. I've visualised it where larger dots == stronger interpolated signal:

This is the results (visualised in red this time) from torchrbf:

What's happening here is that outside of the survey area, torchrbf has a harder time interpolating the way the signal should decrease.

Let me know if there's more data I can provide to help resolve this issue (can provide data privately if it's helpful).