forcing exp operator to be used in special form

[arasugurlu]

I'm working with PySR (latest version, using Julia backend) and I noticed that
the exp operator is never used in the discovered equations, even though it is included
in the unary_operators list.

from pysr import PySRRegressor
import numpy as np

X = data
y = data

model = PySRRegressor(
unary_operators=["exp"],
binary_operators=["+", "-", "*", "/"],
niterations=100,
procs=4,
populations=20,
population_size=100,
complexity_of_operators={"exp": 1},
)

model.fit(X, y)
print(model)

Even though the data clearly follows an exponential law,
the discovered expressions never contain exp().
It keeps returning polynomial or multiplicative combinations instead.

I have tried,

Increasing niterations, population_size, and populations

Adjusting complexity_of_operators={"exp": 1}

Using only "exp" as unary operator

At the end, even if exp shows up, it only shows up as exp(K) where K is a constant real number. I need to force "exp" operator to be used as exp (x1/x2) form where x1 and x2 the features given from the dataset.

I want my analytical equation to be in exp(x1/x2) form but i never made it.

Thanks for this amazing package

[MilesCranmer]

Does x2 ever get very close to 0? It could just blow up and hurt convergence, maybe.

[arasugurlu]

The y value is the 7th column, it can also be understood from the line " y_val = float(parts[6]) ".
I apologize for bothering you, I didn't know what format to use.
Here, I try to use exp(x6) = exp(x4/x0) and exp (x7) = exp(x5/x0) if possible.
It would be also great to optimize those expressions with a constant number, such as, exp(K*x4/x0) where K is a real constant number.
The dimensions of the constants also can be seen in the comments. I really appreciate your help. You can see the first 20 rows above.

import numpy as np
import matplotlib.pyplot as plt
from pysr import PySRRegressor

--- Example data ---
file_path = r"/arasugurlu/PySR-2/train_set.txt"

with open(file_path, "r", encoding="utf-8") as file:
for line in file:
parts = line.strip().split()
if len(parts) != 7:
continue
try:

        x0 = float(parts[0])  # m
        x1 = float(parts[1])  # N/m^2
        x2 = float(parts[2])  # N/m^2
        x3 = float(parts[3])  # N/m^2
        x4 = float(parts[4])  # m
        x5 = float(parts[5])  # m
        x6 = x4 / x0          # dimensionless
        x7 = x5 / x0          # dimensionless
        x8 = x4 / x5          # dimensionless
        g = 3*x3*(x1-x2)/(x1-x2+4*x3) # N/m^2
    y_val = float(parts[6]) # N/m^2

    X.append([x0, g, x6, x7])
        y.append(y_val)

X = np.array(X)
y = np.array(y)

X_units = ["m","N/m^2","",""]
y_units = "N/m^2"

--- Example elementwise loss (it is randomly chosen)---
elementwise_loss = """
function loss_fnc(predictions, targets)
total_loss = 0.0
for (p, t) in zip(predictions, targets)
# simple relative error
total_loss += abs(p - t) / (abs(t) + 1e-6)
end
return total_loss
end
"""

--- PySR model ---
model = PySRRegressor(
procs=64,
populations=192,
population_size=192,
ncycles_per_iteration=500,
niterations=1000,
maxsize=25,
maxdepth=25,
parsimony=0.001,
binary_operators=["*", "+", "-", "/"],
unary_operators=["exp"],
dimensionless_constants_only=True,
weight_randomize=0.2,
weight_optimize=0.001,
warm_start=True,
elementwise_loss=elementwise_loss
)

Fit model
model.fit(X, y)
y_pred = model.predict(X)

[MilesCranmer]

Hi, I am still unable to reproduce it, sorry. Please read through https://stackoverflow.com/help/minimal-reproducible-example for instructions on how to create such an example. Try to make it minimal and reproducible, so if I copy-paste your code on my machine, I see the same issue. You could have the dataset written in the script itself, for example.

[MilesCranmer]

And also please use triple backticks for pasting code

[MilesCranmer]

Oh. I think I can figure it out by looking at your code. You have dimensionless_constants_only=True. Which makes your problem impossible to solve. Because K * x4/x0 has dimensions, and therefore exp(K * x4 / x0) is invalid.

Hope this helps!

[arasugurlu]

Thank you, now i can use, however, there is dimensional inconsistency. How can i solve this?