contour plots, heat plots, composed of several single plots

[felix...@googlemail.com]

Dear all,

because matplotlib.pyplot is slow, I want to use pyqtgraph to display plots of scalar-valued functions of two variables. The figures I want to achieve are similar to contour plots or heat plots, in general they are false color displays of scalar-valued functions of two variables. In my special case they are parts of spectrogram data, which are combined to the plot of a whole spectrogram.

My question is, if there is already a function or an object in pyqtgraph available, which is suitable to combine these parts of the spectrogram to a large spectrogram and show the result as an image? The axes should be suitably configurable and the position and the scale of the little spectrogram parts should also be choosable.

I found a way to combine my little spectrogram parts using a GraphicsLayoutWidget, an ImageItem and translate and scale methods of the ImageItem object. This works fine besides the "DeprecationWarning: Deprecated Qt API, will be removed in 0.13.0.", which is shown because of the usage of the translate and scale methods.

I wonder if there exists already another more simple way inside pqtgraph to show individual spectrogram parts as a whole spectrogram?

An example is attached to the e-mail, which shows the principle of my ideas. There are shown individual images of sin-functions in y-direction and x-direction instead of spectrogram data to make the example easy.

Do you have suggestions?

feli_x

[Patrick]

Hi,

There is nothing fundamentally wrong with what you are doing. You can just keep creating ImageItems and adding them to the plot with appropriate scale and translate transforms applied. I have no idea how many you'd need to add before you run into performance issues. Obviously, with a very large data set you'd want to slice and/or downsample your data depending on the view range of the plot. If your spectral slices are small (a few data points) you may want to merge slices together (say 256--1024 pixels wide).

Here's some code which demonstrates the "new" way to do the translate and scale:

# ...
# plot = pyqtgraph plotitem

# Keep list of image items
spec_images = []

# Loop through existing data slices
# (or add each slice as they are acquired)
for specslice in specslices:
    # ...
    # t = time axis coordinates of slice
    # f = frequency axis of slice(s)
    # Compute scale factors for each image dimension
    # (if identical for each slice, could compute outside the loop instead)
    f_scale = (f[-1] - f[0])/(f.shape[0]-1) if f.shape[0] > 1 else 1.0
    t_scale = (t[-1] - t[0])/(t.shape[0]-1) if t.shape[0] > 1 else 1.0
    #...
    tr = QtGui.QTransform()
    # Centre pixels on the data points
    tr.translate(t[0] - t_scale/2, f[0] - f_scale/2)
    tr.scale(t_scale, f_scale)
    image = pg.ImageItem(specslice)
    image.setTransform(tr)
    plot.addItem(image)
    spec_images.append(image)

Patrick

[felix...@googlemail.com]

Dear Patrick,

thanks for your nice answer and pointing me to the new way to do transformations and scaling!

Kind regards

feli_x