Julia1.0 #4
Julia1.0 #4
Conversation
Codecov Report
@@ Coverage Diff @@
## master #4 +/- ##
==========================================
- Coverage 49.53% 43.41% -6.13%
==========================================
Files 6 7 +1
Lines 216 205 -11
==========================================
- Hits 107 89 -18
- Misses 109 116 +7
Continue to review full report at Codecov.
|
examples/HydroValleys/case3.jl
Outdated
| using Base.Test | ||
| @test isapprox(results["simulations"][1]["objective"], 12400.00, atol=1e-2) | ||
| using Test | ||
| @test isapprox(results[:simulations][1][1][:objective], 10496.09, atol=1) |
There was a problem hiding this comment.
Hmm. To get the old behaviour you need something like
sum(s[:stage_objective] for s in results[:simulations][1])We could make this nicer.
| objective_bound = 0.0 | ||
| optimizer = with_optimizer(params["optimizer"]), | ||
| lower_bound = 0.0, | ||
| direct_mode=false |
There was a problem hiding this comment.
Hmm. I should probably document this better.
| using Base.Test | ||
| @test isapprox(results["simulations"][1]["objective"], 12400.00, atol=1e-2) | ||
| using Test | ||
| @test isapprox(sum(s[:stage_objective] for s in results[:simulations][1]), 9400.0, atol=1) |
There was a problem hiding this comment.
Why did the value (12400) change? Is it because you run more iterations? (I can't guarantee that SDDP.jl is bug-free, so changes like this are worth investigating.)
There was a problem hiding this comment.
It doesn't seem to be because I run more iterations. I augmented it because with 60 iterations, in the new framework, the sddp was not returning a stable solution.
My best guess is that the scenarios are not being sampled in the same way. I tried to seed it, but no luck. Do you believe this could be the reason?
There was a problem hiding this comment.
Oh, it's probably because I turn cut selection on by default now. You can set the cut_deletion_minimum argument in train to a large number like 1_000_000 if you want to turn it off.
There was a problem hiding this comment.
Just added the cut_deletion_minimum argument. However, I was not doing this on the old release. Couldn't this be because the random generator has actually changed and my seed doesn't work the same way?
I was advise to start testing the lower-bound instead of the simulation cost.
There was a problem hiding this comment.
Potentially. I usually check the lower bound:
https://github.com/odow/SDDP.jl/blob/9fea8fcd3b6ab028b94835ccb6617f4d4d9a4b5c/examples/FAST_hydro_thermal.jl#L37
There was a problem hiding this comment.
Will do this from now on!
| using Base.Test | ||
| @test isapprox(results["simulations"][1]["objective"], 59800.0, atol=1e-2) | ||
| using Test | ||
| @test isapprox(results[:simulations][1][1][:objective], 58528.47, atol=1) |
There was a problem hiding this comment.
Changed to sum(s[:stage_objective] for s in results[:simulations][1]), but still different.
|
Use import or using? |
|
Probably using. It's more user friendly. Within a package, I'm trying to mainly use import. The reexport is a thorn though. |
Code to julia 1.0.