Hello guys,
After training a custom deep-sphere model (based on ResidualLayer+HealpyChebyshev), I decided to save it in keras as usual, via callbacks:
callback_model= tf.keras.callbacks.ModelCheckpoint(
filepath=filepath+"cp-{epoch:04d}.ckpt",
monitor="val_loss",
verbose=1,
save_best_only=True,
save_weights_only=False,
mode="min")
However, I've got this error:

~/anaconda3/lib/python3.8/site-packages/keras/saving/saved_model/save_impl.py in call_and_return_conditional_lossess(*args, **kwargs)
631 def call_and_return_conditional_losses(*args, **kwargs):
632 """Returns layer (call_output, conditional losses) tuple."""
--> 633 call_output = layer_call(*args, **kwargs)
634 if version_utils.is_v1_layer_or_model(layer):
635 conditional_losses = layer.get_losses_for(

TypeError: call() got multiple values for argument 'training'

Then, I tried to save the model located in the examples folder: "quick_start.ipynb", and also I obtained the same error!! Models cannot be saved with model.save_weights(checkpoint_path) either. I wonder if this is a problem from my environment, or this is due to the custom layers defined in the repository.
I really appreciate your comments.
Best,
Javier

Hi,

This might not be the right place, but didn't find other contact options so posting it here.
It is regarding the autoencoder model in deepshphere tf2 tutorial. Using this tutorial, with a good amount of sample data, I am able to reconstruct a input healpy map with reasonably recovered features. However, instead of passing the training data in both x and y, if I try to pass a label dataset against the training dataset in y, in an attempt to denoising the map, I find that the model doesn't recover the desired features. I am wondering if there is any similar tutorial for denoising or if you have come across similar situation.

Many Thanks
Wasim

I try to run quick_start.ipynb in examples directory and I got the error

TypeError Traceback (most recent call last)
in
1 tf.keras.backend.clear_session()
----> 2 model = HealpyGCNN(nside=nside, indices=indices, layers=layers, n_neighbors=20)
3 batch_size = 16
4 model.build(input_shape=(None, len(indices), 1))
5 #model.summary(110)

~/deepsphere-cosmo-tf2/deepsphere/healpy_networks.py in init(self, nside, indices, layers, n_neighbors)
85 hp_nn.Healpy_Transformer,hp_nn.HealpyBernstein)):
86 # we need to calculate the current L and get the actual layer
---> 87 sphere = SphereHealpix(subdivisions=current_nside, indexes=current_indices, nest=True,
88 k=self.n_neighbors, lap_type='normalized')
89 current_L = sphere.L

~/.local/lib/python3.9/site-packages/pygsp/graphs/nngraphs/spherehealpix.py in init(self, subdivisions, indexes, nest, **kwargs)
114 ' {} subdivisions.'.format(k, subdivisions))
115
--> 116 super(SphereHealpix, self).init(coords, k=k,
117 kernel_width=kernel_width,
118 **kwargs)

~/.local/lib/python3.9/site-packages/pygsp/graphs/nngraphs/nngraph.py in init(self, features, standardize, metric, order, kind, k, radius, kernel, kernel_width, backend, **kwargs)
417
418
--> 419 neighbors, distances = function(features, metric, order,
420 kind, k, radius, kwargs)
421 # ------ MARTINO's MODIFICATION ------

~/.local/lib/python3.9/site-packages/pygsp/graphs/nngraphs/nngraph.py in _scipy_ckdtree(features, _, order, kind, k, radius, params)
64 params['k'] = features.shape[0] # number of vertices
65 params['distance_upper_bound'] = radius
---> 66 distances, neighbors = tree.query(features, **params)
67 if kind == 'knn':
68 return neighbors, distances

_ckdtree.pyx in scipy.spatial._ckdtree.cKDTree.query()

_ckdtree.pyx in scipy.spatial._ckdtree.get_num_workers()

TypeError: Unexpected keyword argument {'n_jobs': -1}

and also I tested installation via pytest test and got same error in
test_gnn_transformers.py
tests/test_healpy_networks.py .
Now I'm using python 3.9 and I'm program noob. If you have some advice to fix this, it's great to get that.

I created a conda environment, activated it and installed pip in it; I then followed steps 1 and 2 in https://github.com/deepsphere/deepsphere-cosmo-tf2, but it returns

"
note: This error originates from a subprocess, and is likely not a
problem with pip.
ERROR: Failed building wheel for scipy
Building wheel for PyGSP (setup.py) ... done
Created wheel for PyGSP: filename=PyGSP-0.5.1-py3-none-any.whl
size=1886744
sha256=622eda0ca58242241bf95de9e2aac471594f6db650332a7def884a07cd1d967c
Stored in directory:
/private/var/folders/5l/8sxy3dbs4_sbyqdhz5g0tv2w0000gn/T/pip-ephem-wheel-cac
he-1xrlcgi6/wheels/02/45/86/e3968e70e73a8a08f3f802ca85992fede12b1a01485f22f3
29
Successfully built PyGSP
Failed to build scipy
ERROR: Could not build wheels for scipy, which is required to install
pyproject.toml-based projects "

Installing scipy by hand doesn't solve the problem.
What can I do?
The python version is 3.11.5

Hello, I was wondering if it's possible to use this library with a TPU, or if there are some limitations that prevent one to do it.

Thank you very much for your help!

The polynomial values are computed in a loop:

For commonly used degree of 5, that requires 5 consecutive steps. It may be faster to execute this using a single sparse-dense-matmul.
That would require building a sparse laplacian for each polynomial component, which can be pre-computed.
The single sparse-dense-matmul can be launched once with input of
sparse-dense-matmul( block sparse-L, tile input )
followed by a reshape and a sum.
For N=polynomial degree and M=map length and ignoring channels and batch dimensions, the matrix multiplication could look like this

[ N*M x N*M ] is the block-sparse L
[ N*M x 1 ] is the input tiled N times
[ N*M x 1 ] = sparse-dense-matmul( [ N*M x N*M ], [ N*M x 1 ] )
[ N x M x 1 ] = reshape( [ N*M x 1 ], dims=[ N, M, 1 ] )

which gives N feature maps.

Any chance that would make sense?

Also tagging @Arne-Thomsen

Hi,

I am trying to install deepsphere tf2. Although, it shows that the installation is completed, however, when trying to import dependencies from deepsphere, the following error occurs:

cannot import name 'SphereHealpix' from 'pygsp.graphs'

How should I solve this?

Many Thanks

Hello,
I am getting an error when I am running either 'pytest tests' or the quick_start notebook. This error is: "ValueError: All layers added to a Sequential model should have unique names. Name "" is already the name of a layer in this model. Update the name argument to pass a unique name.";
which appears during the creation of the model:
model = HealpyGCNN(nside=nside, indices=indices, layers=layers, n_neighbors=20).

I am running deepsphere-cosmotf2 in tf.2.7 and python3.8.10. Anyone has ideas to solve the problem?
Thank you very much.