That is evidently a problem of psutil; do you have an old version of it? I have tested with 5.2.2 and 5.1.2. Earlier, psutil seemed to have had a different convention. But both is not related to Scanpy.

Does running pip install psutil --upgrade help? If yes, I will update the requirements with a minimal version of psutil so that others don't run into the same problem.

Hm, I researched a bit more. psutil doesn't seem to cause problems and also, this has not been a problem within Scanpy for any user up to now. If you start a terminal with python and type

import psutil
psutil.process_iter()

does this throw an error? I'd really like to know what's going on.

If you want a quick fix; you can simply comment out line 773 in your file /ifs/devel/hashem/sw-v1/conda/lib/python3.6/site-packages/scanpy/readwrite.py; this should cause no problem for your applications.

Btw: Scanpy now has an initial documentation at https://scanpy.readthedocs.io.

Hi Alex,
The psutil issue by updating it has apparently gone away, however later on when I call sc.tl.louvain(adata_corrected, n_neighbors=10, resolution=1.3, recompute_graph=True)
, I get the following error. The igraph I am using is V 0.1.11.
Many thanks
Hashem
`DeprecationWarning Traceback (most recent call last)
in ()
1
----> 2 sc.tl.louvain(adata_corrected, n_neighbors=10, resolution=1.3, recompute_graph=True)
3

/ifs/devel/hashem/sw-v1/conda/lib/python3.6/site-packages/scanpy/tools/louvain.py in louvain(adata, n_neighbors, resolution, n_pcs, random_state, flavor, directed, recompute_pca, recompute_distances, recompute_graph, n_dcs, n_jobs, copy)
78 directed = False
79 if not directed: logg.m(' using the undirected graph', v=4)
---> 80 g = utils.get_igraph_from_adjacency(adjacency, directed=directed)
81 if flavor == 'vtraag':
82 import louvain

/ifs/devel/hashem/sw-v1/conda/lib/python3.6/site-packages/scanpy/utils.py in get_igraph_from_adjacency(adjacency, directed)
41 def get_igraph_from_adjacency(adjacency, directed=None):
42 """Get igraph graph from adjacency matrix."""
---> 43 import igraph as ig
44 sources, targets = adjacency.nonzero()
45 weights = adjacency[sources, targets]

/ifs/devel/hashem/sw-v1/conda/lib/python3.6/site-packages/igraph/init.py in ()
6 license = "MIT"
7
----> 8 raise DeprecationWarning("To avoid name collision with the igraph project, "
9 "this visualization library has been renamed to "
10 "'jgraph'. Please upgrade when convenient.")

DeprecationWarning: To avoid name collision with the igraph project, this visualization library has been renamed to 'jgraph'. Please upgrade when convenient.`

This is again a configuration problem. You don't have http://igraph.org/python/ installed. That's a library with thousands of users and citations. It has a very powerful and fast C++ core that allows treating dataset sizes with a million cells.

I realize that I misspecified this in Scanpy's automatic installation in the requirements file. I just updated this and will push it to the master branch.

You simply need to type pip install python-igraph and then everything should work.

I think I have the most recent version of python-graph installed ( python-igraph 0.7.1.post6 ).
pip install python-igraph doesnt solve the issue since it is already updated.

I think this was happening because of confusion between igraph and python-igraph, by un-installing both and reinstalling just python-igraph, the issue was solved. Interesting enough installation with conda didnt work, only installation with pip worked.
thanks again.

Sorry for being offline for a few hours and great that you could resolve it. In the next version, this confusion will not appear again.

Even though, obviously, any Scanpy release reproduces all examples on https://github.com/theislab/scanpy_usage, we're still at an early stage in the package. Things are progressing very fast and structure and maintenance of the package are becoming more and more professional. Also, soon, many new features and examples where other Python packages are used will be added.

Thank you for a bit of patience at this stage.
Alex

PS: We now have an initial version of the documentation: https://scanpy.readthedocs.io.

No worries and thank you for usually very prompt suggestions.
The idea that scanpy can handle many cells efficiently is great and therefore I have been trying it in a computing cluster (and not my local machine) for the future usage. This in turn makes configuration just a bit more difficult.
Looking forward to a more stable version with more added function.
Thank you
Hashem