Cache chunk boundaries for integer slicing #4923
Conversation
This is an alternative to dask#4909, to implement dask#4867. Instead of caching in the class as in dask#4909, use functools.lru_cache. This unfortunately has a fixed cache size rather than a cache entry stored with each array, but simplifies the code as it is not necessary to pass the cached value from the Array class down through the call tree to the point of use. A quick benchmark shows that the result for indexing a single value from a large array is similar to that from dask#4909, i.e., around 10x faster for constructing the graph. This only applies the cache in `_slice_1d`, so should be considered a proof-of-concept.
|
@mrocklin this is based on your idea for using an lru_cache to cache chunk boundaries - thanks for the suggestion! If you're happy with it I can go looking for other places it can be used. I'm not sure which module cached_cumsum belongs in. It felt a little wrong to have to add numpy to the imports for dask.utils. |
bmerry
force-pushed
the
cache-chunk-boundaries2
branch
from
41e8c95 to
9cfeb81
Compare
|
Oops, looks like I pushed a slightly older version of this. I've just force-pushed my corrected version, which hopefully will actually pass CI. |
dask/utils.py
Outdated
| return 1 + ndimlist(seq[0]) | ||
|
|
||
|
|
||
| class _HashIdWrapper(object): |
There was a problem hiding this comment.
I'm curious, why do we need this?
There was a problem hiding this comment.
Hashing a large tuple takes time proportional to the length of the tuple (about 2ms for 10^6 elements on my machine), which would somewhat negate the benefits of caching (although it is still faster than np.cumsum, mostly because converting a tuple to an ndarray is surprisingly slow). So I'm trading off poorer cache usage (because tuples with the same value but different identity require separate cache slots) to get higher performance lookups.
There was a problem hiding this comment.
If we didn't add the HashIdWrapper, and hashed tuples directly, how much slower would that make your workloads?
When I look at the difference I see about a 20x difference
In [1]: t = tuple(range(1000000))
In [2]: d = {}
In [3]: %time d[t] = 1
CPU times: user 2.54 ms, sys: 0 ns, total: 2.54 ms
Wall time: 2.55 ms
In [4]: %timeit d[t]
2.31 ms ± 151 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
In [5]: import numpy as n
In [6]: import numpy as np
In [7]: %timeit np.cumsum(t)
43.1 ms ± 477 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)I'm hoping that this is fast enough, and that something else becomes the bottleneck instead, especially given your statement here:
A quick benchmark shows that the result for indexing a single value from
a large array is similar to that from #4909, i.e., around 10x faster for
constructing the graph.
There was a problem hiding this comment.
I'll take a look, but my goal is to make slicing M chunks out of an N-chunk array take O(M + log N) rather than O(N). Are you objecting to the wrapper because it reduces cache re-use or just because it adds complexity?
There was a problem hiding this comment.
Just because of complexity. It's not a big deal, we can keep it (it's well isolated), I just thought I'd ask in case it wasn't necessary in practice.
There was a problem hiding this comment.
Here's some benchmarking. It's somewhat pathological (an enormous number of very small chunks) so I don't expect the differences to be as significant in real use, but it shows that the wrapper does improve performance.
#!/usr/bin/env python
import time
from katdal.lazy_indexer import dask_getitem
import dask.array as da
import numpy as np
a = da.ones((1000000, 100), chunks=(1, 100))
step = 100
start = time.monotonic()
for i in range(0, a.shape[0], step):
dask_getitem(a, np.s_[i : i + step]).compute()
stop = time.monotonic()
print(stop - start)- master: 1301 seconds
- branch: 278 seconds
- branch without
_HashIdWrapper: 339 seconds
There was a problem hiding this comment.
It improves things, but not by a large amount. My guess is that if we didn't use _HashIdWrapper that you wouldn't notice enough to add it.
I'm happy either way, with a slight preference for simpler/slower code. It sounds like you prefer HashIdWrapper though, given that you seem to be making a few changes these days I'm happy to defer to your preference here.
In fact I see the tests fail because they don't like |
Right, the core dask library doesn't have any dependencies. Depending on NumPy would definitely be unwelcome. Perhaps a dask/array/utils.py file? |
There is one already, but it depends on array/core, so if I add it there then you end up with a cycle of dependencies array/slicing -> array/utils -> array/core -> array/slicing. I could also add it to array/slicing (which doesn't depend on anything else from array), but it's not really specific to slicing. |
It can't go in dask/utils.py because the top level is not supposed to depend on numpy.
The underlying _cumsum is first called with _HashIdWrapper, which will hit (very cheaply) if we've seen this tuple object before. If not, it will call itself again without the wrapper, which will hit (but at a higher cost for tuple.__hash__) if we've seen the same value before but in a different tuple object.
|
I've put it in array/slicing for now. I also tweaked the cache lookup so that it looks up by value if there is a miss in the lookup by name, which should give the benefits of both without noticeably increasing storage (because each entry is cached twice but the key and value objects are shared rather than duplicated). |
|
@mrocklin It sounds like you're happy with this overall approach (modulo the concerns about |
|
If it's easy (<1hr ?) then I recommend that we do it here. If we're going to add things then we might as well add them consistently. |
|
Ok, I'll look for low-hanging fruit tomorrow. There are also a number of places that use toolz.accumulate rather than np.cumsum. Those might need more work to check that it's safe to switch out one for the other. |
|
I've picked off the low-hanging fruit. |
|
OK. Merging this in. Thanks for all the work @bmerry ! |
commit 255cc5b Author: Justin Waugh <[email protected]> Date: Mon Jun 17 08:18:26 2019 -0600 Map Dask Series to Dask Series (dask#4872) * index-test needed fix * single-parititon-error * added code to make it work * add tests * delete some comments * remove seed set * updated tests * remove sort_index and add tests commit f7d73f8 Author: Matthew Rocklin <[email protected]> Date: Mon Jun 17 15:22:35 2019 +0200 Further relax Array meta checks for Xarray (dask#4944) Our checks in slicing were causing issues for Xarray, which has some unslicable array types. Additionally, this centralizes a bit of logic from blockwise into meta_from_array * simplify slicing meta code with meta_from_array commit 4f97be6 Author: Peter Andreas Entschev <[email protected]> Date: Mon Jun 17 15:21:15 2019 +0200 Expand *_like_safe usage (dask#4946) commit abe9e28 Author: Peter Andreas Entschev <[email protected]> Date: Mon Jun 17 15:19:24 2019 +0200 Defer order/casting einsum parameters to NumPy implementation (dask#4914) commit 76f55fd Author: Matthew Rocklin <[email protected]> Date: Mon Jun 17 09:28:07 2019 +0200 Remove numpy warning in moment calculation (dask#4921) Previously we would divide by 0 in meta calculations for dask array moments, which would raise a Numpy RuntimeWarning to users. Now we avoid that situation, though we may also want to investigate a more thorough solution. commit c437e63 Author: Matthew Rocklin <[email protected]> Date: Sun Jun 16 10:42:16 2019 +0200 Fix meta_from_array to support Xarray test suite (dask#4938) Fixes pydata/xarray#3009 commit d8ff4c4 Author: jakirkham <[email protected]> Date: Fri Jun 14 10:35:00 2019 -0400 Add a diagnostics extra (includes bokeh) (dask#4924) * Add a diagnostics extra (includes bokeh) * Bump bokeh minimum to 0.13.0 * Add to `test_imports` commit 773f775 Author: btw08 <[email protected]> Date: Fri Jun 14 14:34:34 2019 +0000 4809 fix extra cr (dask#4935) * added test that fails to demonstrate the issue in 4809 * modfied open_files/OpenFile to accept a newline parameter, similar to io.TextIOWrapper or the builtin open on py3. Pass newline='' to open_files when preparing to write csv files. Fixed dask#4809 * modified newline documentation to follow convention * added blank line to make test_csv.py flake8-compliant commit 419d27e Author: Peter Andreas Entschev <[email protected]> Date: Fri Jun 14 15:18:42 2019 +0200 Minor meta construction cleanup in concatenate (dask#4937) commit 1f821f4 Author: Bruce Merry <[email protected]> Date: Fri Jun 14 12:49:59 2019 +0200 Cache chunk boundaries for integer slicing (dask#4923) This is an alternative to dask#4909, to implement dask#4867. Instead of caching in the class as in dask#4909, use functools.lru_cache. This unfortunately has a fixed cache size rather than a cache entry stored with each array, but simplifies the code as it is not necessary to pass the cached value from the Array class down through the call tree to the point of use. A quick benchmark shows that the result for indexing a single value from a large array is similar to that from dask#4909, i.e., around 10x faster for constructing the graph. This only applies the cache in `_slice_1d`, so should be considered a proof-of-concept. * Move cached_cumsum to dask/array/slicing.py It can't go in dask/utils.py because the top level is not supposed to depend on numpy. * cached_cumsum: index cache by both id and hash The underlying _cumsum is first called with _HashIdWrapper, which will hit (very cheaply) if we've seen this tuple object before. If not, it will call itself again without the wrapper, which will hit (but at a higher cost for tuple.__hash__) if we've seen the same value before but in a different tuple object. * Apply cached_cumsum in more places commit 66531ba Author: jakirkham <[email protected]> Date: Thu Jun 13 12:13:55 2019 -0400 Drop size 0 arrays in concatenate (dask#4167) * Test `da.concatenate` with size 0 array Make sure that `da.concatenate` does not include empty arrays in the result as they don't contribute any data. * Drop size 0 arrays from `da.concatenate` If any of the arrays passed to `da.concatenate` has a size of 0, then it won't contribute anything to the array created by concatenation. As such make sure to drop any size 0 arrays from the sequence of arrays to concatenate before proceeding. * Handle dtype and all 0 size case * Cast inputs with asarray * Coerce all arrays to concatenate to the same type * Drop obsoleted type handling code * Comment on why arrays are being dropped * Use `np.promote_types` for parity w/old behavior * Handle endianness during type promotion * Construct empty array of right type Avoids the need to cast later and the addition of another node to the graph. * Promote types in `concatenate` using `_meta` There was some left over type promotion code for the arrays to concatenate using their `dtype`s. However this should now use the `_meta` information instead since that is available. * Ensure `concatenate` is working on Dask Arrays * Raise `ValueError` if `concatenate` gets no arrays NumPy will raise if no arrays are provided to concatenate as it is unclear what to do. This adds a similar exception for Dask Arrays. Also this short circuits handling unusual cases later. Plus raises a clearer exception than one might see if this weren't raised. * Test `concatenate` raises when no arrays are given * Determine the concatenated array's shape Needed to handle the case where all arrays have trivial shapes. * Handle special sequence cases together * Update dask/array/core.py Co-Authored-By: James Bourbeau <[email protected]> * Drop outdated comment * Assume valid `_meta` in `concatenate` Simplifies the `_meta` handling logic in `concatenate` to assume that `_meta` is valid. As all arguments have been coerced to Dask Arrays, this is a reasonable assumption to make. commit 46aef58 Author: James Bourbeau <[email protected]> Date: Thu Jun 13 11:04:47 2019 -0500 Overload HLG values method (dask#4918) * Overload HLG values method * Return lists for keys, values, and items * Add tests for keys and items commit f9cd802 Author: mcsoini <[email protected]> Date: Thu Jun 13 18:03:55 2019 +0200 Merge dtype warning (dask#4917) * add test covering the merge column dtype mismatch warning * for various merge types: checks that the resulting dataframe has either no nans or that a UserWarning has been thrown * Add warning for mismatches between column data types * fixes issue dask#4574 * Warning is thrown if the on-columns of left and right have different dtypes * flake8 fixes * fixes * use asciitable for warning string commit c400691 Author: Hugo <[email protected]> Date: Thu Jun 13 17:38:37 2019 +0300 Docs: Drop support for Python 2.7 (dask#4932) commit 985cdf2 Author: Benjamin Zaitlen <[email protected]> Date: Thu Jun 13 10:38:15 2019 -0400 Groupby Covariance/Correlation (dask#4889) commit 6e8c1b7 Author: Jim Crist <[email protected]> Date: Wed Jun 12 15:55:11 2019 -0500 Drop Python 2.7 (dask#4919) * Drop Python 2.7 Drops Python 2.7 from our `setup.py`, and from our test matrix. We don't drop any of the compatability fixes (yet), but won't be adding new ones. * fixup commit 7a9cfaf Author: Ian Bolliger <[email protected]> Date: Wed Jun 12 11:44:26 2019 -0700 keep index name with to_datetime (dask#4905) * keep index name with to_datetime * allow users to pass meta * Update dask/dataframe/core.py put meta as explicit kwarg Co-Authored-By: Matthew Rocklin <[email protected]> * Update dask/dataframe/core.py remove meta kwargs.pop Co-Authored-By: Matthew Rocklin <[email protected]> * remove test for index * allow index commit abc86d3 Author: jakirkham <[email protected]> Date: Wed Jun 12 14:20:59 2019 -0400 Raise ValueError if concatenate is given no arrays (dask#4927) * Raise `ValueError` if `concatenate` gets no arrays NumPy will raise if no arrays are provided to concatenate as it is unclear what to do. This adds a similar exception for Dask Arrays. Also this short circuits handling unusual cases later. Plus raises a clearer exception than one might see if this weren't raised. * Test `concatenate` raises when no arrays are given commit ce2f866 Author: jakirkham <[email protected]> Date: Wed Jun 12 14:09:35 2019 -0400 Promote types in `concatenate` using `_meta` (dask#4925) * Promote types in `concatenate` using `_meta` There was some left over type promotion code for the arrays to concatenate using their `dtype`s. However this should now use the `_meta` information instead since that is available. * Ensure `concatenate` is working on Dask Arrays Merge remote-tracking branch 'upstream/master' into dataframe-warnings
3 participants
This is an alternative to #4909. Fixes #4867.
Instead of caching in the class as in #4909, use functools.lru_cache.
This unfortunately has a fixed cache size rather than a cache entry
stored with each array, but simplifies the code as it is not necessary
to pass the cached value from the Array class down through the call tree
to the point of use.
A quick benchmark shows that the result for indexing a single value from
a large array is similar to that from #4909, i.e., around 10x faster for
constructing the graph.
This only applies the cache in
_slice_1d, so should be considered aproof-of-concept.
flake8 dask