v0.4.1

CMBLensing v0.4.1

Diff since v0.4.0

Closed issues:

• Support non-square flat-sky maps (#15)
• Fix EB constructor (#24)

Merged pull requests:

• Upgrade to CUDA.jl 2 (#21) (@marius311)
• CompatHelper: bump compat for "JLD2" to "0.3" (#33) (@github-actions[bot])
• CompatHelper: bump compat for "StaticArrays" to "1.0" (#35) (@github-actions[bot])
• Allow gradients to bail out to Arrays, and some correctness fixes (#39) (@marius311)
• CompatHelper: bump compat for "AbstractFFTs" to "1.0" (#43) (@github-actions[bot])
• CompatHelper: bump compat for "Adapt" to "3.0" (#44) (@github-actions[bot])
• CompatHelper: bump compat for "KahanSummation" to "0.2" (#45) (@github-actions[bot])

v0.4.0

CMBLensing v0.4.0

Diff since v0.3.0

Closed issues:

• Give argmaxf_lnP a more reasonable default tolerance (#16)
• Document that MKL should be used (#17)

Merged pull requests:

• Support for rectangular flat-sky maps (#23) (@kimmywu)
• Add QQ, UU, QU indexing to Fourier-diagonal EB covariances (#22) (@cailmdaley)
• Add documentation for GPU (#14) (@marius311)
• Add CompatHelper.yml (#25) (@RafaelArutjunjan)
• CompatHelper: bump compat for "JLD2" to "0.2" (#26) (@github-actions[bot])
• CompatHelper: bump compat for "DataStructures" to "0.18" (#27) (@github-actions[bot])

Other notable changes:

• Upgrade from CuArrays.jl to CUDA.jl
• Switch to using OptimKit.jl-based LBFGS solver for MAP_joint.

v0.3.0

CMBLensing v0.3.0

Diff since v0.2.1

Closed issues:

• passing operators to load_sim_dataset is delicate on v0.2.1 (#13)

Merged pull requests:

• add batched fields for better GPU usage (#11) (@marius311)
• Preliminary API for defining custom DataSets (#12) (@marius311)

v0.2.1

CMBLensing v0.2.1

Diff since v0.2.0

Changes:

• Upgrade to CuArrays 2.2.0
• Add Chain and Chains wrapper objects returned by and documented in load_chains
• Some fixes to correctness and differentiability of BinRescaledOp and a some mixing operators

v0.2.0

CMBLensing v0.2.0

Diff since v0.1.2

Changes:

• Both Zygote and CuArrays have bugs which break CMBLensing in recent versions, so their version is capped here for now
• rfid parameter to load_sim_dataset is deprecated, instead do fiducial_θ=(r=...,)
• Saving chains to disk is now done incrementally (only new samples are written each time) for greatly improved performance. load_chains is available to read these new chain files (still one file per chain).
• Add init_GPU_workers and enable MPI parallel GPU runs
• Add resimulate and resimulate!, only the latter which mutates, and they both return all the resimulated fields (this was the behavior of return_truths=true before)
• Handling of CPU/GPU random number generation is cleaned up interally somewhat
• Add BinRescaledOp for bandpower estimation

Merged pull requests:

• set PACKAGECOMPILE to false to at least get docs building (#8) (@marius311)

v0.1.2

CMBLensing v0.1.2

Diff since v0.1.1

Merged pull requests:

• relax some dependency requirements so we can install on JuliaPro 1.3.1 (#5) (@marius311)
• Build documentation via Github Actions (#6) (@marius311)
• Add BilinearLens, improve MAP_marg (#7) (@marius311)

v0.1.1

CMBLensing v0.1.1

Diff since v0.1.0

v0.1.0

v0.1.0 (2020-03-02)

Diff since 2166eebe40134fd92c2c09091683fa81e612b972

Closed issues:

• Figure out a way to get MyBinder working again (#2)
• Bug loading CAMB Cls (#1)