Adding support for different weight vector types #250
Commits on May 6, 2017
-
Initial work on adding weight vector types with bias correction.
NOTE: This broke a lot of things, so for now I'm keeping the existing tests without correction * WeightVec -> AbstractWeights * Weights (aka default reliability weights) * FrequencyWeights (limited to vectors of integers) * All weights take a corrected argument which defaults to true. * Added an exponential function which creates a set of exponential `Weights`. * Updated existing test cases to work with the appropriate types and added false to `weights` calls to avoid breaking existing test cases.
-
Added corrected option to many stats methods.
* Reverts many test changes from the last commit * Changed a bunch of test cases to use `corrected=false` for now * This included a lot of little changes to method definition and some resulting formatting changes where necessary * Added a few test cases for the corrected variances * Added bias correction for ProbabililtyWeights. * Deprecated `WeightVec` * Added a macro for easier creation of weight types. * Renamed weight creation function to `fweights`, `pweights`, etc.
-
-
Updated the "default" bias correction code to iteratively compute the…
… sum squared norm vs using vectorized operations.
-
-
Changed function name
bias->cfactor.* Started dispatching on corrected value rather than repeating the same corrected check for each weights type. * Added extra documentation for the different weights type, including documenting the alternative names for each type. * `cfactor(::AbstractWeights, true` now throws an argument error to avoid return an incorrect result for weight types that don't provide bias correction.
-
Removed corrected flag from
skewnessandkurtosisand set default…… behaviour to corrected=false.
-
Removed
correctedoption frommoments, but kept it for_moment2…… for `varm` to call internally.
-
-
-
-
-
-
Reworked
varandstddefinitions and doc strings.* `corrected` is now an optional argument rather than a keyword so we can produce a deprecation warning about `corrected=false` becoming `corrected=true`. NOTE: this is contrary to the `std` and `var` methods in base, but base julia isn't even consistent about whether `corrected` is a keyword or not betwen `var`, `std` and `cov`. * The `var` and `std` methods now have doc strings explaining how bias correction works with weight vectors. * `mean_and_x` methods default to what they were before.
-
-
Updated mean_and_x docstrings to mention
correctedand point to the…… `var` or `std` documentation.
-
Added documentation about
correctedargument incovand `mean_and……_cov` docstring. Also, added deprecation for cov `corrected=false` behaviour.
-
-
Updated test/moments.jl to use
@testsetso that we can test against…… multiple weight types corrected vs uncorrected calculations.
-
Updated test/weights.jl to use
@testsetso that we can test against…… multiple weight types and added tests for `eweights`.
-
Switched positional corrected back to a keyword with `Union{Bool, Voi…
…d}` deprecation approach. * Reverted some of the code reorganization changes in moments.jl and cov.jl to make the PR review easier * Added a DepBool alias for `Union{Bool, Void}` * Added a `depcheck` method for checking handling deprecation of unset `corrected` keyword. -
Added generic
Weightstype and deprecatedWeightVecto it.Also, did a bunch of docs cleanup from the previous commit.
-
Removed eweights and fixed up test cases.
* Removed eweights from this PR. * Fixed up weights and moments tests cases to loop over weight types more often. * Fixed Probability weights n = length(w) -> n = count(!iszero, w)
-
-
-
-
Updated docs to refer to
AbstractWeighsvsWeightVecand included…… a brief description of different weight types.
-
-
-
Removed
fweightsfrom tests in favour ofweights(to reduces PR s……ize) or `f` where appropriate.
-
Moved description of different weight types in an Implementations sec…
…tions and decided to only reference the `var`, `std` and `cov` docstrings.
-
-
-
Not sure how much this helped, but tried to minimize the amount of `@…
…testset` changes (`cov` was kind of a lost cause).
-
Added comment about unsupported bias correction for the
Weightstyp……e in `var`, `std` and `cov` docstrings.
-
-
Removed more deprecation test hacks and convert
wv->win the ap……propriate docs.
-
-
