Skip to content

Latest commit

 

History

History
267 lines (231 loc) · 21.8 KB

CHANGELOG.md

File metadata and controls

267 lines (231 loc) · 21.8 KB
Raw

v1.0.1

Bug fixes

v1.0.0

New features / Highlights

  • #660 NN module for data parallel neural networks
  • #699 Support for complex numbers; New functions: angle, real, imag, conjugate
  • #702 Support channel stackoverflow
  • #728 DASO optimizer
  • #757 Major documentation overhaul, custom docstrings formatting

Bug fixes

  • #706 Bug fix: prevent __setitem__, __getitem__ from modifying key in place
  • #709 Set the encoding for README.md in setup.py explicitly.
  • #716 Bugfix: Finding clusters by spectral gap fails when multiple diffs identical
  • #732 Corrected logic in DNDarray.__getitem__ to produce the correct split axis
  • #734 Fix division by zero error in __local_op with out != None on empty local arrays.
  • #735 Set return type to bool in relational functions.
  • #744 Fix split semantics for reduction operations
  • #756 Keep track of sent items while balancing within sort()
  • #764 Fixed an issue where repr was giving the wrong output.
  • #767 Corrected std to not use numpy

DNDarray

  • #680 New property: larray: extract local torch.Tensor
  • #683 New properties: nbytes, gnbytes, lnbytes
  • #687 New property: balanced

Factories

  • #707 New feature: asarray()

I/O

  • #559 Enhancement: save_netcdf allows naming dimensions, creating unlimited dimensions, using existing dimensions and variables, slicing

Linear Algebra

  • #658 Bugfix: matmul on GPU will cast away from ints to floats for the operation and cast back upon its completion. This may result in numerical inaccuracies for very large int64 DNDarrays

Logical

  • #711 isfinite(), isinf(), isnan()
  • #743 isneginf(), isposinf()

Manipulations

  • #677 New features: split, vsplit, dsplit, hsplit
  • #690 New feature: ravel
  • #690 Enhancement: reshape accepts shape arguments with one unknown dimension
  • #690 Enhancement: reshape accepts shape arguments with one unknown dimension.
  • #706 Bug fix: prevent __setitem__, __getitem__ from modifying key in place

Neural Networks

  • #660 New submodule: nn.DataParallel for creating and training data parallel neural networks
  • #660 New feature: Synchronous and Asynchronous gradient updates availble for ht.nn.DataParallel
  • #660 New feature: utils.data.datatools.DataLoader for created a local torch.utils.data.Dataloader for use with ht.nn.DataParallel
  • #660 New feature: utils.data.datatools.Dataset for created a local torch.utils.data.Dataset for use with ht.nn.DataParallel
  • #660 Added MNIST example to example/nn to show the use of ht.nn.DataParallel. The MNISTDataset can be found in ht.utils.data.mnist.py
  • #660 New feature: Data loader for H5 datasets which shuffles data in the background during training (utils.data.partial_dataset.PartialH5Dataset)
  • #728 New feature: nn.DataParallelMultiGPU which uses torch.distributed for local communication (for use with optim.DASO)
  • #728 New feature: optim.DetectMetricPlateau detects when a given metric plateaus.

Statistical Functions

  • #679 New feature: histc() and histogram()

Types

  • #712 New function: issubdtype
  • #738 iscomplex(), isreal()

Unit testing / CI

  • #717 Switch CPU CI over to Jenkins and pre-commit to GitHub action.
  • #720 Ignore test files in codecov report and allow drops in code coverage.
  • #725 Add tests for expected warnings.
  • #736 Reference Jenkins CI tests and set development status to Beta.

v0.5.1

  • #678 Bugfix: Internal functions now use explicit device parameters for DNDarray and torch.Tensor initializations.
  • #684 Bug fix: distributed reshape now works on booleans as well.

v0.5.0

  • #488 Enhancement: Rework of the test device selection.
  • #569 New feature: distributed percentile() and median()
  • #572 New feature: distributed pad()
  • #573 Bugfix: matmul fixes: early out for 2 vectors, remainders not added if inner block is 1 for split 10 case
  • #575 Bugfix: Binary operations use proper type casting
  • #575 Bugfix: where() and cov() convert ints to floats when given as parameters
  • #577 Add DNDarray.ndim property
  • #578 Bugfix: Bad variable in reshape()
  • #580 New feature: distributed fliplr()
  • #581 New Feature: DNDarray.tolist()
  • #583 New feature: distributed rot90()
  • #593 New feature distributed arctan2()
  • #594 New feature: Advanced indexing
  • #594 Bugfix: distributed __getitem__ and __setitem__ memory consumption heavily reduced
  • #596 New feature: distributed outer()
  • #598 Type casting changed to PyTorch style casting (i.e. intuitive casting) instead of safe casting
  • #600 New feature: shape()
  • #608 New features: distributed stack(), column_stack(), row_stack()
  • #614 New feature: printing of DNDarrays and __repr__ and __str__ functions
  • #615 New feature: distributed skew()
  • #615 New feature: distributed kurtosis()
  • #618 Printing of unbalanced DNDarrays added
  • #620 New feature: distributed knn
  • #624 Bugfix: distributed median() indexing and casting
  • #629 New features: distributed asin, acos, atan, atan2
  • #631 Bugfix: get_halo behaviour when rank has no data.
  • #634 New features: distributed kmedians, kmedoids, manhattan
  • #633 Documentation: updated contributing.md
  • #635 DNDarray.__getitem__ balances and resplits the given key to None if the key is a DNDarray
  • #638 Fix: arange returns float32 with single input of type float & update skipped device tests
  • #639 Bugfix: balanced array in demo_knn, changed behaviour of knn
  • #648 Bugfix: tensor printing with PyTorch 1.6.0
  • #651 Bugfix: NotImplemented is now NotImplementedError in core.communication.Communication base class
  • #652 Feature: benchmark scripts and jobscript generation
  • #653 Printing above threshold gathers the data without a buffer now
  • #653 Bugfixes: Update unittests argmax & argmin + force index order in mpi_argmax & mpi_argmin. Add device parameter for tensor creation in dndarray.get_halo().
  • #659 New feature: distributed random.permutation + random.randperm
  • #662 Bugfixes: minimum() and maximum() split semantics, scalar input, different input dtype
  • #664 New feature / enhancement: distributed random.random_sample, random.random, random.sample, random.ranf, random.random_integer
  • #666 New feature: distributed prepend/append for diff().
  • #667 Enhancement reshape: rename axis parameter
  • #674 New feature: repeat
  • #670 New Feature: distributed bincount()
  • #672 Bug / Enhancement: Remove MPIRequest.wait(), rewrite calls with capital letters. lower case wait() now falls back to the mpi4py function

v0.4.0

  • Update documentation theme to "Read the Docs"
  • #429 Create submodule for Linear Algebra functions
  • #429 Implemented QR
  • #429 Implemented a tiling class to create Square tiles along the diagonal of a 2D matrix
  • #429 Added PyTorch Jitter to inner function of matmul for increased speed
  • #483 Bugfix: Underlying torch tensor moves to the right device on array initialisation
  • #483 Bugfix: DNDarray.cpu() changes heat device to cpu
  • #496 New feature: flipud()
  • #498 Feature: flip()
  • #499 Bugfix: MPI datatype mapping: torch.int16 now maps to MPI.SHORT instead of MPI.SHORT_INT
  • #501 New Feature: flatten
  • #506 Bugfix: setup.py has correct version parsing
  • #507 Bugfix: sanitize_axis changes axis of 0-dim scalars to None
  • #511 New feature: reshape
  • #515 ht.var() now returns the unadjusted sample variance by default, Bessel's correction can be applied by setting ddof=1.
  • #518 Implementation of Spectral Clustering.
  • #519 Bugfix: distributed slicing with empty list or scalar as input; distributed nonzero() of empty (local) tensor.
  • #520 Bugfix: Resplit returns correct values now.
  • #520 Feature: SplitTiles class, used in new resplit, tiles with theoretical and actual split axes
  • #521 Add documentation for the dtype reduce_op in Heat's core
  • #522 Added CUDA-aware MPI detection for MVAPICH, MPICH and ParaStation.
  • #524 New Feature: cumsum & cumprod
  • #526 float32 is now consistent default dtype for factories.
  • #531 Tiling objects are not separate from the DNDarray
  • #534 eye() supports all 2D split combinations and matrix configurations.
  • #535 Introduction of BaseEstimator and clustering, classification and regression mixins.
  • #536 Getting rid of the docs folder
  • #541 Introduction of basic halo scheme for inter-rank operations
  • #558 sanitize_memory_layout assumes default memory layout of the input tensor
  • #558 Support for PyTorch 1.5.0 added
  • #562 Bugfix: split semantics of ht.squeeze()
  • #567 Bugfix: split differences for setitem are now assumed to be correctly given, error will come from torch upon the setting of the value

v0.3.0

  • #454 Update lasso example
  • #474 New feature: distributed Gaussian Naive Bayes classifier
  • #473 Matmul now will not split any of the input matrices if both have split=None. To toggle splitting of one input for increased speed use the allow_resplit flag.
  • #473 dot handles 2 split None vectors correctly now
  • #470 Enhancement: Accelerate distance calculations in kmeans clustering by introduction of new module spatial.distance
  • #478 ht.array now typecasts the local torch tensors if the torch tensors given are not the torch version of the specified dtype + unit test updates
  • #479 Completion of spatial.distance module to support 2D input arrays of different splittings (None or 0) and different datatypes, also if second input argument is None

v0.2.2

This version adds support for PyTorch 1.4.0. There are also several minor feature improvements and bug fixes listed below.

  • #443 added option for neutral elements to be used in the place of empty tensors in reduction operations (operations.__reduce_op) (cf. #369 and #444)
  • #445 var and std both now support iterable axis arguments
  • #452 updated pull request template
  • #465 bug fix: x.unique() returns a DNDarray both in distributed and non-distributed mode (cf. [#464])
  • #463 Bugfix: Lasso tests now run with both GPUs and CPUs

v0.2.1

This version fixes the packaging, such that installed versions of HeAT contain all required Python packages.

v0.2.0

This version varies greatly from the previous version (0.1.0). This version includes a great increase in functionality and there are many changes. Many functions which were working previously now behave more closely to their numpy counterparts. Although a large amount of progress has been made, work is still ongoing. We appreciate everyone who uses this package and we work hard to solve the issues which you report to us. Thank you!

Updated Package Requirements

  • python >= 3.5
  • mpi4py >= 3.0.0
  • numpy >= 1.13.0
  • torch >= 1.3.0

Optional Packages

  • h5py >= 2.8.0
  • netCDF4 >= 1.4.0, <= 1.5.2
  • pre-commit >= 1.18.3 (development requirement)

Additions

GPU Support

#415 GPU support was added for this release. To set the default device use ht.use_device(dev) where dev can be either "gpu" or "cpu". Make sure to specify the device when creating DNDarrays if the desired device is different than the default. If no device is specified then that device is assumed to be "cpu".

Basic Operations

Basic Multi-DNDarray Operations

Developmental

  • Code of conduct
  • Contribution guidelines
    • pre-commit and black checks added to Pull Requests to ensure proper formatting
  • Issue templates
  • #357 Logspace factory
  • #428 lshape map creation
  • Pull Request Template
  • Removal of the ml folder in favor of regression and clustering folders
  • #365 Test suite

Linear Algebra and Statistics

Regression, Clustering, and Misc.

  • #307 lasso regression example
  • #308 kmeans scikit feature completeness
  • #435 Parter matrix

Bug Fixes

  • KMeans bug fixes
    • Working in distributed mode
    • Fixed shape cluster centers for init='kmeans++'
  • __local_op now returns proper gshape
  • allgatherv fix -> elements now sorted in the correct order
  • getitiem fixes and improvements
  • unique now returns a distributed result if the input was distributed
  • AllToAll on single process now functioning properly
  • optional packages are truly optional for running the unit tests
  • the output of mean and var (and std) now set the correct split axis for the returned DNDarray