MilesCranmer
commented
1 year ago Good idea. So far there is only precision in [16, 32, 64], but in principle, BigFloat should be easy to support.
You can hack it like this:
import pysr
jl = pysr.julia_helpers.init_julia()
# First, set the precision:
jl.setprecision(200) # 200 bits
# Define `randn` so it works for BigFloat:
jl.eval("""
using Random: AbstractRNG
try
# In case randn(BigFloat) gets defined in the future
randn(BigFloat)
catch e
if isa(e, MethodError)
Base.randn(::Type{BigFloat}, args::Integer...) = BigFloat.(randn(Float64, args...))
Base.randn(rng::AbstractRNG, ::Type{BigFloat}, args::Integer...) = BigFloat.(randn(rng, Float64, args...))
else
throw(e)
end
end
""")Then, you need to edit pysr/sr.py, at this line: https://github.com/MilesCranmer/PySR/blob/46fe8c163f147d8447993c80cba4bf0d2f43d461/pysr/sr.py#L1638.
Add the lines:
Main.X = Main.eval("BigFloat.(X)")
Main.y = Main.eval("BigFloat.(y)")
if weights is not None:
Main.weights = Main.eval("BigFloat.(weights)")I think that should be it. Let me know if it works. I'd also love a pull request if you can figure out a nice way to make this configurable using the precision kwarg.
Cheers, Miles
dbl001
Is your feature request related to a problem? Please describe. I tried running 'pysr' on a 1,000 row array with 4 integer input variables and one integer output variable - a Goedel Number.
From Mathematica:
E.g.
The model returned:
I am just learning 'pysr' and maybe it's just 'user error'. However, Inf and Nan suggest that Goedel numbers may exceed Float64.
Describe the solution you'd like Not sure what happened, because the largest Goedel number in the input is: 1.6679880978201e+23
Additional context I didn't see any parameters to set 'verbose' mode or 'debugging' information.
GoedelTableFourParameters.txt