Closed Datseris closed 3 years ago
Datseris
commented
4 years ago About TimeseriesSurrogates: are they used in any of the methods or is it installed only because of the documentation (because I can also fix this; in general the documentation production is done in a separate project where you can install whatever you want without affecting the front-end user).
kahaaga
commented
4 years ago At the moment doing ]add CausalityTools installs Plots, GR, FFMPEG, and many other packages. It even installs Documenter (?!?!?):
This should have been taken care of already. As you say, plotting should be handled by RecipesBase.
In the latest tagged version, neither Plots nor Documenter are dependencies. Currently, the last tagged version of the package is v0.9.9, in which the dependencies are as follows
[deps]
CausalityToolsBase = "b1f25513-c649-51ee-8a1c-0e98d01ae897"
Conda = "8f4d0f93-b110-5947-807f-2305c1781a2d"
CrossMappings = "7011085d-2e9b-5a7b-b0c2-f102c05e50c2"
DelayEmbeddings = "5732040d-69e3-5649-938a-b6b4f237613f"
Distances = "b4f34e82-e78d-54a5-968a-f98e89d6e8f7"
Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
DynamicalSystems = "61744808-ddfa-5f27-97ff-6e42cc95d634"
HypothesisTests = "09f84164-cd44-5f33-b23f-e6b0d136a0d5"
Interpolations = "a98d9a8b-a2ab-59e6-89dd-64a1c18fca59"
NearestNeighbors = "b8a86587-4115-5ab1-83bc-aa920d37bbce"
PerronFrobenius = "260eed61-d0e8-5f1e-b040-a9756a401c97"
Pkg = "44cfe95a-1eb2-52ea-b672-e2afdf69b78f"
PyCall = "438e738f-606a-5dbb-bf0a-cddfbfd45ab0"
RecipesBase = "3cdcf5f2-1ef4-517c-9805-6587b60abb01"
Reexport = "189a3867-3050-52da-a836-e630ba90ab69"
Requires = "ae029012-a4dd-5104-9daa-d747884805df"
SimpleDiffEq = "05bca326-078c-5bf0-a5bf-ce7c7982d7fd"
Simplices = "d5428e67-3037-59ba-9ab1-57a04f0a3b6a"
StateSpaceReconstruction = "1441a9f6-6a74-5418-a591-cdf1d78a07f0"
StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"
TimeseriesSurrogates = "c804724b-8c18-5caa-8579-6025a0767c70"
TransferEntropy = "ea221983-52f3-5440-99c7-13ea201cd633"
UncertainData = "dcd9ba68-c27b-5cea-ae21-829cd07325bf"In your update message it says that v0.3.0 was installed. I'm not sure why it isn't upgrading to the latest version.
kahaaga
commented
4 years ago About TimeseriesSurrogates: are they used in any of the methods or is it installed only because of the documentation (because I can also fix this; in general the documentation production is done in a separate project where you can install whatever you want without affecting the front-end user).
TimeseriesSurrogates is installed because it is included used by the cross mapping routines. However, I have realised that this is unnecessary - the user can do this while preparing the input data - so was planning to remove this option from crossmap and convergentcrossmapping. This reminds me to do so, so I'm adding that to the 1.0 milestone!
However, surrogate testing will explicitly be part of a high-level causality test at some points, so it still needs to be included then. But for now, I think it is sufficient to just remove any explicit dependencies on TimeseriesSurrogates by updating CrossMappings.
Datseris
commented
4 years ago However, surrogate testing will explicitly be part of a high-level causality test at some points, so it still needs to be included then. But for now, I think it is sufficient to just remove any explicit dependencies on TimeseriesSurrogates by updating CrossMappings.
Packages used for only for testing should not be part of the normal dependency tree. For example, have a look here: https://github.com/JuliaDynamics/ChaosTools.jl/blob/master/Project.toml at the extras . OrdinaryDiffEq is used for testing, but is not a dependency of ChaosTools and thus it is not installed when a user adds it. it is only necessary when running the tests.
kahaaga
commented
4 years ago Packages used for only for testing should not be part of the normal dependency tree.
Sorry, I didn't mean testing as in julia code tests. Surrogates are used for hypothesis testing in the context of time series causality analysis. For example, one way to tell if transfer entropy (TE) is significant or not, for example the nonzero-TE in the non-causal direction we discussed in #122, is to compare the obtained TE value with that obtained for an ensemble of surrogate time series.
If there are any dependencies used only for testing left, they are remnants from some old version. In the latest tagged version (v0.9.9), I think I should have gotten rid of all testing-related packages.
kahaaga
commented
3 years ago All plotting is removed in v1 (#129).
At the moment doing
]add CausalityToolsinstalls Plots, GR, FFMPEG, and many other packages. It even installs Documenter (?!?!?):Notice also that this dependency list is excluding the dependencies of DynamicalSystems.jl. I think this large, (in fact too large) and many things have to be dropped before 1.0. For example, plotting must absolutely not be installed with this package, as this makes it unusable in clusters (where plotting is typically impossible to install). The plotting can use
RecipesBaseor be in a different repo altogether, or behind arequiresblock.I guess having Documenter as a dependency is a mistake which I'll fix when I make the documenter PR.
It is also evident that parts of the dependencies have not been updated to their 1.0 version, e.g. Combinatorics. I'd recommend to use CompatHelper.jl to make this process easier.