Error when running the Lorenz 96 Example

[benxro]

I made a fresh installation of Julia 1.7.2 on Ubuntu 20.04. Cloned the repository, then followed the installation instructions.
Running the test cases works without issues, passing 340 of 340 tests.
I cd into the directory of /EnsembleKalmanProcesses.jl/examples/Lorenz/.
The documentation under 'Examples - Lorenz' advises:

The L96 parameter estimation can be run using julia --project Lorenz_example.jl

After resolving all missing packages and executing the mentioned command, I receive the following error:

$ julia --project Lorenz_example.jl 

/home/ben/some_path/EnsembleKalmanProcesses.jl/examples/Lorenz
2
[8.0; 2.5;;]
ERROR: LoadError: MethodError: no method matching ParameterDistribution(::Vector{Dict{String, Any}})
Closest candidates are:
  ParameterDistribution(::Union{Array{PDType}, PDType}, ::Union{Array{CType}, Array, CType}, ::Union{Array{ST}, ST}) 
  where {PDType<:ParameterDistributionType, CType<:EnsembleKalmanProcesses.ParameterDistributions.ConstraintType, ST<:AbstractString} 
  at ~/.julia/packages/EnsembleKalmanProcesses/iWN0p/src/ParameterDistributions.jl:167
Stacktrace:
 [1] top-level scope
   @ ~/some_path/EnsembleKalmanProcesses.jl/examples/Lorenz/Lorenz_example.jl:98
in expression starting at /home/ben/some_path/Kalman_test/EnsembleKalmanProcesses.jl/examples/Lorenz/Lorenz_example.jl:85

I also tried the installation via the package manager, leading to the same error.
If you need any further information, please let me know.

On a side note, it also seems that the current version of Cloudy is incompatible with the EnsembleKalmanProcesses, as I wasn't able to install both into the same (empty) environment.

Kind regards,
Ben

[odunbar]

Hi Ben,
Sorry the examples haven't worked out instantly! We have just updated EnsembleKalmanProcesses to a new interface, however we have not yet made these into a release yet. We will try to sort this out soon with a new breaking release.

This means, if you added EnsembleKalmanProcesses using something like

julia --project
]
add EnsembleKalmanProcesses

This will take the released version, not the latest one compatible with the examples.

Solution

One option, is to add EnsembleKalmanProcesses from the github address, or from a local path.
Go to Lorenz and remove the package:

julia --project
]
rm EnsembleKalmanProcesses

then (1) load it from the github main branch

add https://github.com/CliMA/EnsembleKalmanProcesses.jl#main

or (2) if you have a local version, add it locally

add ../..

my output using (1) gives:

2
[8.0; 2.5;;]
Using truth values to compute covariance
[0.22112428787318822 -0.004533149180349995 -0.0019610730500580365 0.03904494950054565; -0.004533149180349995 0.0637033768682159 -0.0009818730618963512 0.0023911706426016713; -0.0019610730500580365 -0.0009818730618963512 0.01675340867722733 -0.006785591840294391; 0.03904494950054565 0.0023911706426016713 -0.006785591840294391 0.05597049454872317]
EKP inversion error:
┌ Warning: New ensemble covariance determinant is less than 0.01 times its previous value.
│ Consider reducing the EK time step.
└ @ EnsembleKalmanProcesses ~/.julia/packages/EnsembleKalmanProcesses/U0thB/src/EnsembleKalmanInversion.jl:180
Iteration: 1, Error: 566.4761551590692
Iteration: 2, Error: 6.472141069453601
Iteration: 3, Error: 0.4077423818530319
Iteration: 4, Error: 0.39488169110060684
Iteration: 5, Error: 0.3243814378039824
True parameters: 
[8.0; 2.5;;]

EKI results:
[7.974555791388336; 2.4647490689605016;;]

(and some plots)

[benxro]

That works, great!
I followed your instructions (2) and received the same output as you.

Thank you very much @odunbar :)

[odunbar]

No problem!
Also PR #159 releases the new EKP version so hopefully adding from the registry will work as desired.