-
Notifications
You must be signed in to change notification settings - Fork 47
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Exp time kernel #595
Exp time kernel #595
Conversation
Codecov Report
@@ Coverage Diff @@
## dev #595 +/- ##
==========================================
- Coverage 87.78% 86.01% -1.78%
==========================================
Files 63 65 +2
Lines 7885 8107 +222
==========================================
+ Hits 6922 6973 +51
- Misses 963 1134 +171
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
Hi Mike, first comment, when trying to import with ---------------------------------------------------------------------------
ModuleNotFoundError Traceback (most recent call last)
~/Projects/cellrank/cellrank/external/kernels/_wot_kernel.py in <module>
17 try:
---> 18 import wot
19
ModuleNotFoundError: No module named 'wot'
The above exception was the direct cause of the following exception:
ImportError Traceback (most recent call last)
<ipython-input-20-d38dddb5f66c> in <module>
----> 1 otk = WOTKernel(adata)
~/Projects/cellrank/cellrank/external/kernels/_wot_kernel.py in __init__(self, adata, backward, time_key, compute_cond_num)
57 compute_cond_num: bool = False,
58 ):
---> 59 super().__init__(
60 adata,
61 backward=backward,
~/Projects/cellrank/cellrank/external/_error_mixin.py in __init__(self, *args, **kwargs)
24
25 def __init__(self, *args: Any, **kwargs: Any):
---> 26 raise ImportError(self.__import_error_message__) from self.__error__
27
28 def __init_subclass__(cls, error: Optional[Exception] = None, **kwargs: Any):
ImportError: Unable to import the kernel. This is of course expected as I don't have the package installed - however, could we print a message, telling me to install it from pip? |
Hey @Marius1311 , done. I've also exposed |
I installed WOT and initalized the kernel, as above. I get the following warning: |
Next point (I'm just going through as I experiment, please excuse the absence of structure here): I print my initialized kernel object, not transition matrix computed yet: I'm just wondering why so many of these are object level properties - I suppose they are specified when I compute a transition matrix? Just as a comparison with the otk = WOTKernel(adata, time_key='day', )
pk = PseudotimeKernel(adata) Print otk gives: Print i.e. the pseudotime kernel is "blank" before I compute the transition matrix, the WOT kernel has lots of attributes already. Why is it different? |
It's in ascending order, changed the message.
There's even more if you supply kwargs, see all options here. I don't think this is problematic though. |
I updated my post from above - I don't think it's problematic either, I just wonder why we handle it differently here than in other kernels. |
Next point: I think |
Can't reproduce, only after
Ok.
Can't do, afaik there are no online docs. https://broadinstitute.github.io/wot/ only has tutorials + cmd line interface + additional stuff. |
You are right, I must have messed it up somehow. All good here.
It's fine to link to that page, it's all explained in the tutorials in length, that's where I got the info from. |
Is |
Some more open points:
|
Another point: can we write the final growth rates (last round of iteration) back to adata.obs please? I think that's an important quantity. |
No, and I don't think it should be - the OT solver is already consuming all cores.
Ok, can you please add this? I also don't know which pre-processing should be used (would be best), just used the imputed data from scVelo (standard preprocessing).
Ok, done.
How do you propose to do that? We don't have neither prolif./apop. info. 2 options I have in mind:
|
Hi Mike!
Thanks! We're getting there. |
This is already done:
We're using WOT's default now (pairwise PCA).
Ok, agreed, will add this. |
Done as # preprocessing
ok = cr.external.kernels.WOT(adata, time_key=...)
ok.compute_initial_growth_rates("prolif_key", "apop_key", ..., key_added="initial_growth_rates") # optional
ok.compute_transition_matrix(..., growth_rate_key="initial_growth_rates") |
Also, I think we should add a tutorial for this, maybe in combination with StationaryOT. I can take care of this. |
Is scanpy used under the hood to compute PCA? |
The transition matrix has the same order of cells as the original one.
No. I'm staying 100% true to WOT's implementation.
Sounds good, I'd wait for 1.4. |
So will the row/column ordering be the same as in
What do you use? Is highly variable gene annotation taken into account? |
Yes.
No. Can add option to subset the adata before to use only HVGs only. |
Re HVGs, that would be good - I would have a parameter |
We should also rename that |
Note to self: we should add to the docstring how we assign transition probs to the last time-point. |
Because I can't justify using only diagonal.
Agreed, once we converge on solution, could you please update it? |
I get that point - in case we know nothing, you say let's use the maximum entropy distribution here which would be uniform. However, taking into account computational considerations, I would go for sparse diagonal because it's
Since there is no obvious choice here, how about we leave it to the user?
Will do! |
Done as |
Fantastic, thanks! |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is fine from my side. Please review my suggested workflow and modify if needed, in particular, if you think it can be shortened. Otherwise, feel free to merge.
We'll have to look into downstream applications, i.e. inferrint the long-range temporal couplings, but that can be done after this merge, which will be the initial step. |
Nice! Should we include these or not?
Visually, they seem similar.
Ok, thanks, will take a look at this. |
For the gene sets, they will be specific to the organism (mouse, in this case), so I wouldn't include them. We can add a link to the tutorial again where they can be downloaded. |
* Smoothing for Pseudotime kernel (#514) * Rename PalantirKernel -> PseudotimeKernel * Add initial abstraction * Add Palantir scheme * Add soft PT * Remove old test * Fix wrong condition * Update docs and defaults * Fix Pseudotime kernel docstring * Fix more docstrings * Update docs * Add tests, pin down numba Co-authored-by: Marius Lange <marius.lange@t-online.de> * Add compute_projection (#520) * Add compute_projection * Swap order of args * Add tests * Fix params, add plotting test * [ci skip] Update docstring * Pin down numba Co-authored-by: Marius Lange <marius.lange@t-online.de> * Fix bugs, QoL improvements (#535) * Fix bugs, QoL improvements * Add QoL changes v2 * Enable CI * Cytotrace kernel (#527) * Add CytoTRACE kernel * Add new kernel to docs * Fix literal * Work on the docstring * Rename variables * Rename the gcs variable * Update the docstring * Add docstring to kernel class * Move the workflow example * [ci skip] Enable CI * Move to CytoTRACEKernel * Improve logging, remove arbitrary restriction * Do not use short kernel names * Add CytoTRACE test skeletons * Set use_raw=False, add tests * Improve tests * Fix raw and normal adata with different n_vars * Add cytotrace score to adata_200 * Add regression test for ct_score * Add regression test for gene_corr * Fix tests Co-authored-by: Marius Lange <marius.lange@t-online.de> * Add external model (#522) * Add external module, port statOT * Add docs * Add typing_extensions * Add some sanity check * Move external to separate page * Add tests * Temp. pin down numba * Add explicit POT mention in docs, add to external * Use raise from * Fix init subclass * Add rest error mixins * Add tox * Update CONTRIBUTING.rst * Update CONTRIBUTING.rst * Fix sections * Update CONTRIBUTING.rst * Update test requirements * [ci skip] Start addressing comments * [ci skip] Update CONTRIBUTING.rst * Add tutorials to CONTRIBUTING.rst * Different import, enable CI on dev * Update GA issues/discussions * Update OTKernel to reflect new changes * Fix wrong import * [ci skip] Update dev install, docs * [ci skip] Fix link * Remove edit_on_github (#538) * Random walk simulation (#537) * Add random walk class * Parallelize * Random walk plotting * Add docstrings * [ci skip] Fix some imports * [ci skip] Add error messages * Add random walks test skeleton * [ci skip] Add plotting test skeletons * Add RW tests * Add plotting tests * Add legend loc for start/stop indices, update docs * Work on docstrings * Add first/last cell description Co-authored-by: Marius Lange <marius.lange@t-online.de> * Fix dist -> conn, update docstrings, remove copy (#536) * Fix dist -> conn, update docstrings, remove copy * Change default * Work on the docstring * Exchange k for fract_to_keep * Make fract_to_keep a float * Fix tests Co-authored-by: Marius Lange <marius.lange@t-online.de> * Fix plot lin drivers (#543) * Fix lineage drivers plotting * Add tests * Add forgotten GT images * Fix nan in emb when projection tmat (#546) * [ci skip] Add Zebrafish dataset (#547) * [ci skip] Remove fig.show() * Fix some logging messages * Release notes 1.3 (#539) * Fix docs, move latest additions * Change latest to stable * Add most of the release notes * [ci skip] Update release notes - new dataset * Go over release notes * Update release notes: move Zebrafish, add pygpcca * Correct tutorial names * Regenerate some figures to accomodate mpl==1.4.0 Co-authored-by: Marius Lange <marius.lange@t-online.de> * Release 1.3 (#548) * Fix typos, update _static * Add forgotten docstring to abstract method * Fix tutorial links in docs * Update the index * Make it a bit shorter * Add links and contributions seciton * Fix spelling * Fix same plot (#556) * Fix reading from adata * Regenerate figure * Add warning * Compute Schur only for num. states, not +1 * Add test * Update README.rst * Minor update * Remove solver heuristics (#558) * Remove solver/PETSc heuristic * Change tol to 1e-6 * Change test tol, regen image * Add check probs sum to 1 (#566) * Add abs. probs check * Add nicer message, add tests * Increase tolerance to 1e-3 * Fix test * [ci skip] Fix test names * Fix not using percentile (#568) * Fix wrong pyGPCCA attribute in logging * First version of req. pruning (#571) * Fix inconsistent state (#563) * Fix inconsistent cat/cont macrostates/lineages * Add tests, fix 1 minor bug * Add force recompute tmat (#577) * Add force tmat recomputation, fix typo bug * Add test * Change circ. projection TSP default heuristic * Remove -1 in hard scheme, modify test (#582) * [ci skip] Remove sphinx-paramlinks (#584) * Pt kernel changes (#583) * Change PT kernel defaults, remove perc/dnorm * Remove density normalize from tests * Fix docs * [ci skip] Remove sphinx-paramlinks * Fix test * Remove dens correction from docstring * Dont force sparse tmat if already dense (#586) * Do not force tmat to be sparse * Add test * Fix tests expecting sparse matrix * Parallelize pt kernel (#587) * Parallelize PT kernel * [ci skip] Add forgotten kwargs in docs * Fix custom connectivities key (#590) * Fix plot_coarse_T annotation * Exp time kernel (#595) * Resurrect exp. time kernel * Add WOT skeleton class * Add WOT * Add tests, add wot to external requirements * Skip WOT tests on CI, update WOT installation * Better import error message, save tmats, last 1s * Clear order info message, fix test skip * Expose more args * Save estimated growth rates to adata * [ci skip] Add compute_initial_growth_rates * Add to the docstrings * Add HVGs subset, fix docs, rename StatOT kernel * Change debug to info, add uniform * Add example workflow * [ci skip] Polish docs Co-authored-by: Marius Lange <marius.lange@t-online.de> * Wot kernel graph (#600) * Don't require connectivities for external kernels * Fix commented skip * Enable bibtex (#601) * Use bibtex for references * Polish docs * Fix dois * Update CytoTRACE workflow * Update WOT example workflow * Remove old examples, add new skeleton ones (#602) * Remove old examples, add new skeleton ones * Move old release notes * Add 2 new examples - projection and RW * Update latest additions * Update RW docstring * Update docstring to manually set terminal states (#607) * Update docstring to manually set terminal states * Fix bug with doc insertion * Add option to set term states from clusters Co-authored-by: Michal Klein <michal.klein@protonmail.com> * Add option to use connectivities for WOT (#609) * Fix tqdm/ipywidgets (#616) * Vein plot (#615) * Start vein plot * Specialize plot_flow for exp. time kernel * Add first working impl. * Update plotting * Add plotting options * Add flow threshold * Rename argument * Fix cyclic import * Fix plot_flow in ExpTimeKernel * Add tmat kernel * Subset clusters, fix bwd * Allow cluster subset and ordering * Add time point filtering * [ci skip] Fix bugs, start abstraction * [ci skip] Clean code * [ci skip] Remove dicts * [ci skip] Fix indexing error * Remove plot_flow specification for TMatKernel * [ci skip] Control ticks * [ci skip] Order legend * Polish docs * Polish docs * Add option to remove empty clusters * Fix linting * [ci skip] Fix logging and docs * Add skeleton tests, fix typo, name to single flow * Add first batch of tests * Add plotting tests * [ci skip] Fix doi * [ci skip] Fix exp. time kernel copy (#623) * Fix lineage drivers to allow for combinations (#624) * Fix parallelize too many jobs (#633) * Cluster lin clusters (#634) * [ci skip] Add cluster key to cluster_lineage * Fix limits * Use more xticks * Add multiple covariates, expose ratio and cmap * Add tests * Add WOT dataset infra (#631) * Add WOT dataset infra * Add figshare link * Fix indentation in docs, missing dot. Co-authored-by: Marius Lange <marius.lange@t-online.de> * Increate timeout * Prolif apop markers (#630) * Add marker fetching class * Add prolif/apop infra * Remove rat * Add human markers * Add mouse markers * Add references * Add test * Minor dosctring changes Co-authored-by: Marius Lange <marius.lange@t-online.de> * Speed up calculation of absorption probabilities (#638) * Refactor `compute_absorption_probabilities` Refactors the calculation of absorption probabilities to first define a single artificial cell in the macro state by summing all cells identifying the macro state. This is equivalent to solving the linear system for each individual cell and then summing the absorption probabilities. * Update test_cflare.py Updates `TestCFLARE.test_abs_probs_negative` to work correctly with the refactored version of `BaseEstimator.compute_absorption_probabilities`. Co-authored-by: michalk8 <46717574+michalk8@users.noreply.github.com> * Stacked histogram (#641) * Add plot macrostate composition * Set margins * Fix cyclic import * Add test skeletons * Add tests * Remove return ax test * Set yticks * Regenerate figures Co-authored-by: Marius Lange <marius.lange@t-online.de> * Log odds (#642) * Initialize log_odds * Move Marius' function * Implement log odds * Add docs, fix TODOs * Add test skeletons * Add colorbar * Add legend_loc * Fix categorical bug, jitter, seed, add tests * Remove testing artifact * Fix copy * Change threshold * Add show * Add key-specific threshold * Add fontsize * Add xticks step size * Add tests, fix tick step size None * Change verbosity to debug * Add plot gene corr (#640) * Add plot gene corr * Add test skeletons * Add tests, fix raw bug * Fix adjustText * Change defaults, ERROR logging * Fix using wrong params Co-authored-by: Marius Lange <marius.lange@t-online.de> * Update precommit (#645) * Update precommit * Update reqs, setup.py, pyproject.toml * Install Cython for 3.9 * Install Cython v2 * POT v3 * Try latest numba version * Disable 3.9 CI * Try installing umap-learn on CI/tox * Fix macOS, docs umap-learn version * Fix docs build * Add some references * Update release notes (#647) * Update release notes * Pin numba in reqs again * Update test reqs., CI Python hash * Purge CI cache * Remove 3.9 CI * Add links to index.rst * Finalize release notes * Fix reference, add missing doi, download notebook Co-authored-by: Marius Lange <marius.lange@t-online.de> Co-authored-by: Marius Lange <marius.lange@t-online.de> Co-authored-by: Marius Lange <marius.lange@tum.de> Co-authored-by: Philipp Weiler <philipp.weiler@tum.de>
IMPORTANT: Please search among the Pull request before creating one.
Title
Types of changes
Description
Add Waddington OT kernel, see related issue broadinstitute/wot#91. I've also added option to use existing obs features (PCA, imputed expression) to compute the cost matrices.
Also adds
ExpTimeKernel
class which can be later used to derive kernels with experimental time.@Marius1311 I've decided against adding anything to the block diagonal, don't think it's a nice solution - user can always weight with full
ConnectivitiesKernel
and afaik, the GPCCA stability issues were gone (I've used small weights like 0.05 or so for connectivities). Maybe you add some nice note in.compute_transition_matrix
, belowReturns
asHow has this been tested?
I've added a few.
Closes
closes #551