kanglcn / moraine Goto Github PK

Three plotting functions are needed currently:

• basic points plot (with rasterize, dynspread, inspect_points). The inspect_points fuction do not work well with geoviews, so directly calculating the coordinate in Web Mercator coordinate is needed. To make the plot clean, inspect_points hover should prevent showing the Web Mercator but I haven't find ways to do it.
• a bunch of plots showing every interferograms. To accelerating the rendering, the figsize should be smaller.
• a points plot to show the velocity with a click function to trigger the time series plotting of that point. This can be achieved with hv.streams.Tap. Here is an example to do it: https://holoviews.org/reference/streams/bokeh/Tap.html

the threads_per_worker have to be 1 in de_emi, otherwise it will crash. No reason.

Add set.diff, set.xor, set.union,set.intersect functions and their corresponding de_diff, de_xor, de_union, de_intersect command.

Phase linking for PTA requires cupy.optimize to be implented which is tracked at rapidsai/raft#923 and cupy/cupy#6112

This works:

arr = da.ones((100,100),chunks=(10,10),dtype=bool)
arr_delayed = arr.to_delayed()

num_true_delayed = np.empty_like(arr_delayed,dtype=object)
with np.nditer(arr_delayed,flags=['multi_index','refs_ok'], op_flags=['readwrite']) as arr_it:
    for arr_block in arr_it:
        idx = arr_it.multi_index
        num_true_delayed[idx] = delayed(np.count_nonzero)(arr_delayed[idx],keepdims=True)
        num_true_delayed[idx] = da.from_delayed(num_true_delayed[idx],shape=(1,1),meta=np.array((),dtype=int))

num_true = da.block(num_true_delayed.tolist())

num_true.compute()

but this doesn't work:

arr = da.ones((100,100),chunks=(10,10),dtype=bool)
arr_delayed = arr.to_delayed()

num_true_delayed = np.empty_like(arr_delayed,dtype=object)
with np.nditer(arr_delayed,flags=['multi_index','refs_ok'], op_flags=['readwrite']) as arr_it:
    for arr_block in arr_it:
        idx = arr_it.multi_index
        num_true_delayed[idx] = delayed(np.count_nonzero)(arr_block,keepdims=True)
        num_true_delayed[idx] = da.from_delayed(num_true_delayed[idx],shape=(1,1),meta=np.array((),dtype=int))

num_true = da.block(num_true_delayed.tolist())

num_true.compute()

I have no idea with this.

i.e., Available GPU, GPU number, available thread, default plot figsize, dpi, etc.

Hello @kanglcn,

I would be interested in running your package on, .rslc and .diff .lat .lon elevation.rdc dem.rslc produced by snap2stamps for PsINSAR analysis on the Stamps package.

I believe this package is similar to dolphin from Isce2 and TomoSAR.

Specifically I want to replace the .rslc and .diff files with updated phase linked .rslc files and .diff as to be used in the Stamps package. I believe this is something it does, if I am mistaken I apologise.

if it is though, is there a step to step document or any other sort to produce these files?

thank you.

I am worry about if int32 is enough for rslc, coh, cov index since there flattenned array are big.

Provide 3 kind of API:

• Array-based API
• File-based API
• CLI

.cell-output-stdout {
overflow-y: scroll;
max-height: 300px;
}

Like DS selection, the selected DS are chunked according to the az chunk size and thus, the chunk size is irregular. However, zarr do not support irregular chunk size. So the rechunk have to be done. However, it is really time consuming. I don't have any solutions.

nbdev_test seems do not free the GPU memory imediately even when setting the n_workers to 1.

And it looks it does not allow read and write files.

Considering using cupyx.scipy.sparse.linalg.eigsh as the EMI eig-decomposer for banded coherence matrix.

Use https://github.com/rapidsai/kvikio when it is mature.

Currently a gc_map_inversion approach is used.
This command is extremely slow when the image is big. Considering using pt2geo for it.