View_corr_mat() - make new || leave old about scona

[KirstieJane]

Thanks @wingedRuslan - can you add an example of the additional functions that seaborn heatmap gives us? I personally prefer the current version because I like the black line around the outside of the plot, and I can't think of options that heatmap will give us that are useful in this case. Happy to be wrong though!

[wingedRuslan]

@KirstieJane, if the black line around the outside of the plot is the only reason, I could add this feature to the new function as well :)

you did this with the following line of code

    # Put a box around your data
    ax.add_patch(
     mpatches.Rectangle(
        (ax.get_xlim()[0], ax.get_ylim()[1]),
        ax.get_xlim()[1],
        ax.get_ylim()[0],
        fill=False,      # remove background
        color='k',
        linewidth=1) )

[wingedRuslan]

@KirstieJane, the only difference (that we care about) could be concerning the improved colorbar:

• in the existed version, the colorbar range is hardcoded to be [-1,0,1]. That's why only the diverging colormap can be used to plot corr_mat; You can see the output above if we use, for instance, sequential colormap.
• in the seaborn version, we can adjust colormap with the vmin, vmax parameters
  vmin, vmax - Values to anchor the colormap, otherwise they are inferred from the data and other keyword arguments.
  cbar - Whether to draw a colorbar (although, not a problem to add this param to the existed version)
  cbar_kws - Keyword arguments for fig.colorbar
  Adjust colorbar properties
  This param I used to set custom labels to the colorbar (-1,0,1) to create the same plot as the existed version.

[wingedRuslan]

Link to the existed function - view_corr_mat - here

[KirstieJane]

Thanks for the link @wingedRuslan! I like that function actually 😄

This is the code I was thinking of that's at the top of the tutorials:

# Read in sample data from the NSPN WhitakerVertes PNAS 2016 paper.
df, names, covars, centroids = datasets.NSPN_WhitakerVertes_PNAS2016.import_data()

# calculate residuals of the matrix df for the columns of names
df_res = scn.create_residuals_df(df, names, covars)

# create a correlation matrix over the columns of df_res
M = scn.create_corrmat(df_res, method='pearson')

# Initialise a weighted graph G from the correlation matrix M
G = scn.BrainNetwork(network=M, parcellation=names, centroids=centroids)

# Threshold G at cost 10 to create a binary graph with 10% as many edges as the complete graph G.
G10 = G.threshold(10)

M is the correlation matrix.

The correct way to threshold the graph is using the G.threshold() command because it makes a minimum spanning tree to make sure that every node is connected for any threshold (cost). BUT for visualization purposes that isn't super important, so we can just threshold using the np.percentile function as in the current code.

To be clear - we do not need to pass a graph G to the function. We can stick with giving it M or a dataframe etc.

The bin option is so that you can visualize the thresholded network as a (usually) black and white correlation matrix. It just sets all the values below the threshold to 0 and all those above to 1. This is what we do when we calculate the global measures.

So the current code looks great. Lets add vmin and vmax in whatever way you think is best (seaborn or adding those options to the current code), and maybe add a symmetric option (or whatever you used for the other colormaps to force vmin to be -vmax or vmax to be -vmin depending on which gives the largest range 😄

Thanks for checking!