Hugging Face's logo

Spaces:
MilesCranmer
/
PySR
Running

App Files Community
PySR / docs /options.md
MilesCranmer's picture
MilesCranmer
Fix formatting warnings
d90120f
preview code
|
raw
history blame
11.5 kB

Features and Options

Some configurable features and options in PySR which you may find useful include:

These are described below

The program will output a pandas DataFrame containing the equations to PySRRegressor.equations containing the loss value and complexity.

It will also dump to a csv at the end of every iteration, which is .hall_of_fame_{date_time}.csv by default. It also prints the equations to stdout.

Model selection

By default, PySRRegressor uses model_selection='best' which selects an equation from PySRRegressor.equations_ using a combination of accuracy and complexity. You can also select model_selection='accuracy'.

By printing a model (i.e., print(model)), you can see the equation selection with the arrow shown in the pick column.

Operators

A list of operators can be found on the operators page. One can define custom operators in Julia by passing a string:

PySRRegressor(niterations=100,
    binary_operators=["mult", "plus", "special(x, y) = x^2 + y"],
    extra_sympy_mappings={'special': lambda x, y: x**2 + y},
    unary_operators=["cos"])

Now, the symbolic regression code can search using this special function that squares its left argument and adds it to its right. Make sure all passed functions are valid Julia code, and take one (unary) or two (binary) float32 scalars as input, and output a float32. This means if you write any real constants in your operator, like 2.5, you have to write them instead as 2.5f0, which defines it as Float32. Operators are automatically vectorized.

One should also define extra_sympy_mappings, so that the SymPy code can understand the output equation from Julia, when constructing a useable function. This step is optional, but is necessary for the lambda_format to work.

Iterations

This is the total number of generations that pysr will run for. I usually set this to a large number, and exit when I am satisfied with the equations.

Cycles per iteration

Each cycle considers every 10-equation subsample (re-sampled for each individual 10, unless fast_cycle is set in which case the subsamples are separate groups of equations) a single time, producing one mutated equation for each. The parameter ncyclesperiteration defines how many times this occurs before the equations are compared to the hall of fame, and new equations are migrated from the hall of fame, or from other populations. It also controls how slowly annealing occurs. You may find that increasing ncyclesperiteration results in a higher cycles-per-second, as the head worker needs to reduce and distribute new equations less often, and also increases diversity. But at the same time, a smaller number it might be that migrating equations from the hall of fame helps each population stay closer to the best current equations.

Processors

One can adjust the number of workers used by Julia with the procs option. You should set this equal to the number of cores you want pysr to use.

Populations

By default, populations=100, but you can set a different number of populations with this option. More populations may increase the diversity of equations discovered, though will take longer to train. However, it is usually more efficient to have populations>procs, as there are multiple populations running on each core.

Weighted data

Here, we assign weights to each row of data using inverse uncertainty squared. We also use 10 processes for the search instead of the default.

sigma = ...
weights = 1/sigma**2

model = PySRRegressor(procs=10)
model.fit(X, y, weights=weights)

Max size

maxsize controls the maximum size of equation (number of operators, constants, variables). maxdepth is by default not used, but can be set to control the maximum depth of an equation. These will make processing faster, as longer equations take longer to test.

One can warm up the maxsize from a small number to encourage PySR to start simple, by using the warmupMaxsize argument. This specifies that maxsize increases every warmupMaxsize.

Batching

One can turn on mini-batching, with the batching flag, and control the batch size with batch_size. This will make evolution faster for large datasets. Equations are still evaluated on the entire dataset at the end of each iteration to compare to the hall of fame, but only on a random subset during mutations and annealing.

Variable Names

You can pass a list of strings naming each column of X with variable_names. Alternatively, you can pass X as a pandas dataframe and the columns will be used as variable names. Make sure only alphabetical characters and _ are used in these names.

Constraining use of operators

One can limit the complexity of specific operators with the constraints parameter. There is a "maxsize" parameter to PySR, but there is also an operator-level "constraints" parameter. One supplies a dict, like so:

constraints={'pow': (-1, 1), 'mult': (3, 3), 'cos': 5}

What this says is that: a power law $x^y$ can have an expression of arbitrary (-1) complexity in the x, but only complexity 1 (e.g., a constant or variable) in the y. So $(x_0 + 3)^{5.5}$ is allowed, but $5.5^{x_0 + 3}$ is not. I find this helps a lot for getting more interpretable equations. The other terms say that each multiplication can only have sub-expressions of up to complexity 3 (e.g., $5.0 + x_2$) in each side, and cosine can only operate on expressions of complexity 5 (e.g., $5.0 + x_2 exp(x_3)$).

Custom complexity

By default, all operators, constants, and instances of variables have a complexity of 1. The sum of the complexities of all terms is the total complexity of an expression. You may change this by configuring the options:

  • complexity_of_operators - pass a dictionary of <str>: <int> pairs to change the complexity of each operator. If an operator is not specified, it will have the default complexity of 1.
  • complexity_of_constants - supplying an integer will make all constants have that complexity.
  • complexity_of_variables - supplying an integer will make all variables have that complexity.

LaTeX and SymPy

After running model.fit(...), you can look at model.equations which is a pandas dataframe. The sympy_format column gives sympy equations, and the lambda_format gives callable functions. You can optionally pass a pandas dataframe to the callable function, if you called .fit on a pandas dataframe as well.

There are also some helper functions for doing this quickly.

  • model.latex() will generate a TeX formatted output of your equation.
  • model.sympy() will return the SymPy representation.
  • model.jax() will return a callable JAX function combined with parameters (see below)
  • model.pytorch() will return a PyTorch model (see below).

Exporting to numpy, pytorch, and jax

By default, the dataframe of equations will contain columns with the identifier lambda_format. These are simple functions which correspond to the equation, but executed with numpy functions. You can pass your X matrix to these functions just as you did to the model.fit call. Thus, this allows you to numerically evaluate the equations over different output.

Calling model.predict will execute the lambda_format of the best equation, and return the result. If you selected model_selection="best", this will use an equation that combines accuracy with simplicity. For model_selection="accuracy", this will just look at accuracy.

One can do the same thing for PyTorch, which uses code from sympytorch, and for JAX, which uses code from sympy2jax.

Calling model.pytorch() will return a PyTorch module which runs the equation, using PyTorch functions, over X (as a PyTorch tensor). This is differentiable, and the parameters of this PyTorch module correspond to the learned parameters in the equation, and are trainable.

torch_model = model.pytorch()
torch_model(X)

Warning: If you are using custom operators, you must define extra_torch_mappings or extra_jax_mappings (both are dict of callables) to provide an equivalent definition of the functions. (At any time you can set these parameters or any others with model.set_params.)

For JAX, you can equivalently call model.jax() This will return a dictionary containing a 'callable' (a JAX function), and 'parameters' (a list of parameters in the equation). You can execute this function with:

jax_model = model.jax()
jax_model['callable'](X, jax_model['parameters'])

Since the parameter list is a jax array, this therefore lets you also train the parameters within JAX (and is differentiable).

loss

The default loss is mean-square error, and weighted mean-square error. One can pass an arbitrary Julia string to define a custom loss, using, e.g., loss="myloss(x, y) = abs(x - y)^1.5". For more details, see the Losses page for SymbolicRegression.jl.

Here are some additional examples:

abs(x-y) loss

PySRRegressor(..., loss="f(x, y) = abs(x - y)^1.5")

Note that the function name doesn't matter:

PySRRegressor(..., loss="loss(x, y) = abs(x * y)")

With weights:

model = PySRRegressor(..., loss="myloss(x, y, w) = w * abs(x - y)") 
model.fit(..., weights=weights)

Weights can be used in arbitrary ways:

model = PySRRegressor(..., weights=weights, loss="myloss(x, y, w) = abs(x - y)^2/w^2")
model.fit(..., weights=weights)

Built-in loss (faster) (see losses). This one computes the L3 norm:

PySRRegressor(..., loss="LPDistLoss{3}()")

Can also uses these losses for weighted (weighted-average):

model = PySRRegressor(..., weights=weights, loss="LPDistLoss{3}()")
model.fit(..., weights=weights)

Model loading

PySR will automatically save a pickle file of the model state when you call model.fit, once before the search starts, and again after the search finishes. The filename will have the same base name as the input file, but with a .pkl extension. You can load the saved model state with:

model = PySRRegressor.from_file(pickle_filename)

If you have a long-running job and would like to load the model before completion, you can also do this. In this case, the model loading will use the csv file to load the equations, since the csv file is continually updated during the search. Once the search completes, the model including its equations will be saved to the pickle file, overwriting the existing version.