All arrays must be instances of SparseArray

[ealsop]

Running de.test.wald using an un-normalized adata.X matrix with only 2 samples and ~5000 cells to test differential expression in scanpy. All libraries (numby, scanpy, diffxpy, sparse and dask) are updated the most recent versions.

test = de.test.wald(data=adata, formula_loc="~ 1 + Group", factor_loc_totest="Group")

make initally make this run at all I needed to include:

%env SPARSE_AUTO_DENSIFY=1

Or I would receive the error "RuntimeError: Cannot convert a sparse array to dense automatically. To manually densify, use the todense method." mentioned in other issues.

With this solved I am now running through several iterations:

training location model: False training scale model: True iter 0: ll=74052044.544684 iter 1: ll=74052044.544684, converged: 0.00% (loc: 100.00%, scale update: False), in 0.00sec iter 2: ll=53700205.003682, converged: 16.18% (loc: 16.18%, scale update: True), in 883.76sec iter 3: ll=53700205.003682, converged: 16.18% (loc: 100.00%, scale update: False), in 0.01sec iter 4: ll=53407203.078969, converged: 84.27% (loc: 84.27%, scale update: True), in 692.52sec iter 5: ll=53407203.078969, converged: 84.27% (loc: 100.00%, scale update: False), in 0.00sec iter 6: ll=53376103.605985, converged: 94.97% (loc: 94.97%, scale update: True), in 178.97sec iter 7: ll=53376103.605985, converged: 94.97% (loc: 100.00%, scale update: False), in 0.00sec Fitting 1842 dispersion models: (progress not available with multiprocessing)

But getting this error / trace back:

ValueError Traceback (most recent call last)

in ----> 1 test = de.test.wald(data=adata, formula_loc="~ 1 + Group", factor_loc_totest="Group") ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/diffxpy/testing/tests.py in wald(data, factor_loc_totest, coef_to_test, formula_loc, formula_scale, as_numeric, init_a, init_b, gene_names, sample_description, dmat_loc, dmat_scale, constraints_loc, constraints_scale, noise_model, size_factors, batch_size, backend, train_args, training_strategy, quick_scale, dtype, **kwargs) 719 720 # Fit model. --> 721 model = _fit( 722 noise_model=noise_model, 723 data=data, ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/diffxpy/testing/tests.py in _fit(noise_model, data, design_loc, design_scale, design_loc_names, design_scale_names, constraints_loc, constraints_scale, init_model, init_a, init_b, gene_names, size_factors, batch_size, backend, training_strategy, quick_scale, train_args, close_session, dtype) 242 pass 243 --> 244 estim.train_sequence( 245 training_strategy=training_strategy, 246 **train_args ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/batchglm/models/base/estimator.py in train_sequence(self, training_strategy, **kwargs) 122 (x, str(d[x]), str(kwargs[x])) 123 ) --> 124 self.train(**d, **kwargs) 125 logger.debug("Training sequence #%d complete", idx + 1) 126 ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/batchglm/train/numpy/base_glm/estimator.py in train(self, max_steps, method_b, update_b_freq, ftol_b, lr_b, max_iter_b, nproc, **kwargs) 115 self.model.b_var = self.model.b_var + b_step 116 # Reverse update by feature if update leads to worse loss: --> 117 ll_proposal = - self.model.ll_byfeature_j(j=idx_update).compute() 118 idx_bad_step = idx_update[np.where(ll_proposal > ll_current[idx_update])[0]] 119 if isinstance(self.model.b_var, dask.array.core.Array): ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/dask/base.py in compute(self, **kwargs) 281 dask.base.compute 282 """ --> 283 (result,) = compute(self, traverse=False, **kwargs) 284 return result 285 ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/dask/base.py in compute(*args, **kwargs) 564 565 results = schedule(dsk, keys, **kwargs) --> 566 return repack([f(r, *a) for r, (f, a) in zip(results, postcomputes)]) 567 568 ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/dask/base.py in (.0) 564 565 results = schedule(dsk, keys, **kwargs) --> 566 return repack([f(r, *a) for r, (f, a) in zip(results, postcomputes)]) 567 568 ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/dask/array/core.py in finalize(results) 1081 while isinstance(results2, (tuple, list)): 1082 if len(results2) > 1: -> 1083 return concatenate3(results) 1084 else: 1085 results2 = results2[0] ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/dask/array/core.py in concatenate3(arrays) 4747 try: 4748 x = unpack_singleton(arrays) -> 4749 return _concatenate2(arrays, axes=tuple(range(x.ndim))) 4750 except TypeError: 4751 pass ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/dask/array/core.py in _concatenate2(arrays, axes) 348 type(max(arrays, key=lambda x: getattr(x, "__array_priority__", 0))) 349 ) --> 350 return concatenate(arrays, axis=axes[0]) 351 352 ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/sparse/_common.py in concatenate(arrays, axis, compressed_axes) 1253 from ._coo import concatenate as coo_concat 1254 -> 1255 return coo_concat(arrays, axis) 1256 else: 1257 from ._compressed import concatenate as gcxs_concat ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/sparse/_coo/common.py in concatenate(arrays, axis) 159 from .core import COO 160 --> 161 check_consistent_fill_value(arrays) 162 163 arrays = [x if isinstance(x, COO) else COO(x) for x in arrays] ~/anaconda3/envs/scanpy/lib/python3.8/site-packages/sparse/_utils.py in check_consistent_fill_value(arrays) 468 469 if not all(isinstance(s, SparseArray) for s in arrays): --> 470 raise ValueError("All arrays must be instances of SparseArray.") 471 if len(arrays) == 0: 472 raise ValueError("At least one array required.") ValueError: All arrays must be instances of SparseArray.

[Zethson]

@niklaslang and me ran into the same issue, although forcing dense arrays did not work either.

[oligomyeggo]

I am also having this same issue and was wondering if there was any resolution/fix?

I initially was getting the same error described in issue #145 but I was using diffxpy v0.7.4. I installed diffxpy v0.7.3 instead, and am now getting the ValueError: All arrays must be instances of SparseArray. error. I tried including %env SPARSE_AUTO_DENSIFY=1 but it did not fix the problem.

Not sure if relevant, but since I saw them listed in issue #145: I have batchglm v0.7.4, sparse v0.12.0 and dask v2021.4.0, which I believe are all the most current versions. Do any of these need to be downgraded as well?

[Zethson]

I am also having this same issue and was wondering if there was any resolution/fix?

I initially was getting the same error described in issue #145 but I was using diffxpy v0.7.4. I installed diffxpy v0.7.3 instead, and am now getting the ValueError: All arrays must be instances of SparseArray. error. I tried including %env SPARSE_AUTO_DENSIFY=1 but it did not fix the problem.

Not sure if relevant, but since I saw them listed in issue #145: I have batchglm v0.7.4, sparse v0.12.0 and dask v2021.4.0, which I believe are all the most current versions. Do any of these need to be downgraded as well?

Hey,

looks like the issue lies within some seemingly unrelated dependencies. Diffxpy should be more strict here.

Try to force:

dask==2021.4.0
sparse==0.9.1

This resolved it on our end.

[Zethson]

I'll likely submit a PR to pin sparse tomorrow or so. Please report back whether this also works for you, @oligomyeggo

[oligomyeggo]

Thanks @Zethson! Forcing dask==2021.4.0 and sparse==0.9.1 worked for me for both diffxpy==0.7.3 and diffxpy==0.7.4. Though now I am hitting a new error which also seems potentially related (?). I am following this tutorial (the first portion, for the standard Wald test), and now everything is working up until the test.plot_vs_ttest() function, which is returning the following error:

ValueError: Performing a mixed sparse-dense operation that would result in a dense array. Please make sure that func(sparse_fill_values, ndarrays) is a constant array.

This occurred with both diffxpy==0.7.3 and diffxpy==0.7.4.

[Zethson]

Thanks @Zethson! Forcing dask==2021.4.0 and sparse==0.9.1 worked for me for both diffxpy==0.7.3 and diffxpy==0.7.4. Though now I am hitting a new error which also seems potentially related (?). I am following this tutorial (the first portion, for the standard Wald test), and now everything is working up until the test.plot_vs_ttest() function, which is returning the following error:

ValueError: Performing a mixed sparse-dense operation that would result in a dense array. Please make sure that func(sparse_fill_values, ndarrays) is a constant array.

This occurred with both diffxpy==0.7.3 and diffxpy==0.7.4.

I'd suggest to create a new issue for this.