Why do I get this error "UndefVarError: cov not defined"

Specific Domains Optimization (Mathematical)
,
1

Could someone help me solve the error: UndefVarError: cov not defined of the test code taken from the site JuMPeR - Robust Optimization with JuMP — JuMPeR -- Julia for Mathematical Programming 0.0 documentation
Did this error arise due to the lack of a package?

using JuMP, JuMPeR, GLPKMathProgInterface, Distributions

N = 2000
NUM_ASSET = 3000

# Setup the robust optimization model
function solve_portfolio(past_returns, Gamma, pref_cuts)
    # Create covariance matrix and mean vector
    covar = cov(past_returns)
    means = mean(past_returns, 1)

    A = round(chol(covar),2)
    m = RobustModel(solver=GLPKSolverLP(OutputFlag=0))

    # Variables
    @variable(m, obj)  # Put objective as constraint using dummy variable
    @variable(m, x[1:NUM_ASSET] >= 0)
    # Uncertainties
    @uncertain(m,       r[1:NUM_ASSET]      )  # The returns
    @uncertain(m, -1 <= z[1:NUM_ASSET] <= 1 )  # The "standard normals"
    @uncertain(m,  0 <= y[1:NUM_ASSET] <= 1 )  # |z|/box

    @objective(m, Max, obj)

    # Portfolio constraint
    @constraint(m, sum([ x[i] for i=1:NUM_ASSET ]) == 1)
    # The objective constraint - uncertain
    @constraint(m, sum([ r[i]*x[i] for i=1:NUM_ASSET ]) - obj >= 0)

    # Build uncertainty set
    # First, link returns to the standard normals
    for asset_ind = 1:NUM_ASSET
        @constraint(m, r[asset_ind] ==
        sum([ A[asset_ind, j] * z[j] for j=1:NUM_ASSET ]) + means[asset_ind] )
    end

    # Then link absolute values to standard normals
    for asset_ind = 1:NUM_ASSET
        @constraint(m, y[asset_ind] >= -z[asset_ind] / box)
        @constraint(m, y[asset_ind] >=  z[asset_ind] / box)
    end

    # Finally, limit how much the standard normals can vary from means
    @constraint(m, sum([ y[j] for j=1:NUM_ASSET ]) <= Gamma)
    solveRobust(m, prefer_cuts=pref_cuts)
    return getValue(x)

end

function eval_gamma(Gamma)
    x = solve_portfolio(past_returns, 1, true)
    future_z = future_returns * x[:]
    sort!(future_z)
    println("Selected solution summary stats for Gamma $Gamma")
    println("10%:     ", future_z[int(NUM_FUTURE*0.1)])
    println("20%:     ", future_z[int(NUM_FUTURE*0.2)])
    println("30%:     ", future_z[int(NUM_FUTURE*0.3)])
    println("Mean:    ", mean(future_z))
    println("Maximum: ", future_z[end])
end

past_returns   = generate_data(1000)
future_returns = generate_data(1000)

eval_gamma(0)  # UndefVarError: cov not defined
eval_gamma(3)   # UndefVarError: cov not defined
2

What’s the trace? Can you give the complete error? Distributions defines cov.

3

Hi @crinders,

I used Distributions because I imagined that the package would compile the covariance which is defined as covar = cov (past_returns)
The complete error message: UndefVarError: cov not defined, appears when I try to compile:

eval_gamma (0)
eval_gamma (3)

Thank you!

4

What happens if you run

using Distributions
cov(1:3)

Do you get 1.0 or the same error? Which version of Distributions are you using? (Pkg.installed() to find out)

5

Hi, @JLDC
He continued with the same mistake. I’m using version 0.23.8

Thank you

6

What about

using Distributions
Distributions.cov(1:3)

is it still the same error?

7

cov is provided by Statistics. Instead of using Distributions, you need

using Statistics
6 Likes
8

For my personal understanding, as I am no Julia expert. Could you please explain why that matters? Isn’t StatsBase, which, as far as I understand includes cov required by Distributions? For instance, I can run cov and Distributions.cov by having loaded only Distributionsand without having Statistics in my list of packages.

9

It only makes sense to load Distributions if you actually need it, since it just reexports cov.

julia> using Distributions

julia> parentmodule(cov)
Statistics
2 Likes
10

@Tamas_Papp thanks for the reply!

I understand that, however, the person is following a tutorial, I assumed Distributions would be needed somewhere else in the tutorial. I think I haven’t express my question quite clearly, sorry. I am more confused as to why

using Statistics
cov(1:3)

would work, when

using Distributions
cov(1:3)

would not.

11

But it does work, at least with versions 0.22 and 0.23 of Distributions I tested with.

12

It does for me as well, but it doesn’t seem to work for the OP.

13

Hi @odow, thanks for the reply
What version of Julia are you using?

14

This will work on all 1.x versions of Julia.

1 Like
15

Hi @odow
I followed your instruction and it worked, thank you!!
However, a new error arose when compiling eval_gamma eval [0] and eval_gamma [3]
Error message: MethodError: no method matching getindex (:: typeof (eval_gamma), :: Int64)

16

Its eval_gamma(0), not eval_gamma[0].

1 Like
17

Hi @odow, Thanks for the help
Unfortunately the error continues, but now this error message has appeared: MethodError: no method matching getindex (:: typeof (mean), :: Array {Float64,2}, :: Int64)