hello,can you help me ? #29
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
TODO: port AEH, investigate caching distance matrices for fixed inducers
instead use a sum of a spectral mixture kernel (Wilson + Adams 2013) and a linear kernel, ARD should figure out the rest This also changes the default settings to fixed inducers and ard=True
- implement sparse GPs according to Titisas (2009) - use score test with BH correction - reorganize code
note that the results data frame no longer contains null model results
Previously, a score test was performed for every lengthscale. Now, only the maximum likelihood model is tested.
The standard workflow now is run() to detect spatially variable genes followed by fit_patterns, which fits spectral mixture kernels to the spatially variable genes. Alternatively, one can use run_detailed() to directly fit spectral mixture kernels
this makes p-values under the null uniformly distributed
this prevents significant hits that have one spot with many reads and very little reads overall
the previous strategy may have been generating false-positives due to the score test being performed with the fitted kernel. The current workflow is to first run the score test on multiple kernels, and then analyze the detected spatially variable genes in more detail with GPs. Based on extensive benchmarking, the Cauchy combination seems to perform best in terms of statistical power, so it is the default. If CPU time is a bottleneck, the omnibus test can be used instead. It may lose some power, but its runtime is independent of the number of kernels to be tested. In the default configuration, it is 10x faster than the Cauchy combination. This commit also ports the score test to use Tensorflow, so it can run on GPUs. This gives a 100x speed-up compared to CPU.
this uses a Hidden Markov Random Field with a Dirichlet process prior to automatically detect the number of cell types present in the given sample. It uses a Poisson likelihood with a Gamma prior. If no spatial information is given, the model reduces to a Dirichlet process Poisson mixture.
previous implementation relied on TensorFlow's arithmetic optimizer to remove redundant cholesky factorizations. Visualizing the graph in TensorBoard showed that this was not the case, one superfluous cholesky factorization was performed in every iteration
a remnant of ScoreTest.dtype was left after commit d2dd0bf
this plays better with ScanPy's plotting functions
this plays better with ScanPy's plotting
parameter fits for null model are always done in float64, but the test itself uses default_float
Disabling caching is useful to reduce memory footprint when testing many data points. It would be nice to do this automatically, but Tensorflow doesn't provide a way to get the total available memory, so...
this removes all the useless clutter introduced by autosummary
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
AttributeError Traceback (most recent call last)
~\AppData\Local\Temp\ipykernel_9316\2717546558.py in
----> 1 svg_full, _ = SpatialDE.test(data, omnibus=True)
2 svg_full["total_counts"] = np.asarray(data.X.sum(axis=0)).squeeze()
3 svg_full.to_pickle("ST8059048_svg_full.pkl")
AttributeError: module 'SpatialDE' has no attribute 'test'