SymbolicRegressor output support for multi-dimensional y

[superminghui]

My outputnode corresponds to an array, not a single value. So my y is a two-dimensional array
But when I was training, something went wrong, he told me:
ValueError: y should be a 1d array, got an array of shape (15, 1600) instead.
My output y looks like this:
[[0. 0. 0. ... 0. 0. 0.] --------> y1
[0. 0. 0. ... 0. 0. 0.] --------> y2
[0. 0. 0. ... 0. 0. 0.] --------> y3
...
[0. 0. 0. ... 0. 0. 0.]
[0. 0. 0. ... 0. 0. 0.]
[0. 0. 0. ... 0. 0. 0.]]
Each individual array represents an output.

So I want to ask if there is any way to solve this problem?

[superminghui]

The output y in example SymbolicRegressor like :
[1,
2,
4,
5,
...
n]
But my output y is array like this:
[[0. 0. 0. ... 0. 0. 0.] --------> y1
[0. 0. 0. ... 0. 0. 0.] --------> y2
[0. 0. 0. ... 0. 0. 0.] --------> y3
...
[0. 0. 0. ... 0. 0. 0.]
[0. 0. 0. ... 0. 0. 0.]
[0. 0. 0. ... 0. 0. 0.]]

So i met problem....
How could i solve this?

[superminghui]

I find the from sklearn.multioutput imported MultiOutputRegressor may can solve this problem
but how could i add it to the gplearn?
Is there any way? Please advise
Thanks soooo much

[trevorstephens]

It looks like gplearn should be compatible with that wrapper, do you run into any issues when trying to follow the syntax in that example with your data? Or maybe a small slice of your data to test with? https://scikit-learn.org/stable/modules/generated/sklearn.multioutput.MultiOutputRegressor.html

[superminghui]

It looks like gplearn should be compatible with that wrapper, do you run into any issues when trying to follow the syntax in that example with your data? Or maybe a small slice of your data to test with? https://scikit-learn.org/stable/modules/generated/sklearn.multioutput.MultiOutputRegressor.html

By using MultiOutputRegressor like this?:

est_gp = MultiOutputRegressor(SymbolicRegressor(
    population_size=500,
    generations=80, stopping_criteria=0.1,
    p_crossover=0.6,
    p_subtree_mutation=0.2, p_hoist_mutation=0.1,
    p_point_mutation=0.1,
    max_samples=0.9, verbose=1,
    parsimony_coefficient=0.01, random_state=0
))

est_gp.fit(train_x, train_y)

Is this correct? Can run

[superminghui]

It looks like gplearn should be compatible with that wrapper, do you run into any issues when trying to follow the syntax in that example with your data? Or maybe a small slice of your data to test with? https://scikit-learn.org/stable/modules/generated/sklearn.multioutput.MultiOutputRegressor.html

But if i use MultiOutputRegressor like this:
est_gp = MultiOutputRegressor(SymbolicRegressor(
population_size=500,
generations=80, stopping_criteria=0.1,
p_crossover=0.6,
p_subtree_mutation=0.2, p_hoist_mutation=0.1,
p_point_mutation=0.1,
max_samples=0.9, verbose=1,
parsimony_coefficient=0.01, random_state=0
))

est_gp.fit(train_x, train_y)

then i cant use the funciton _program to get the solution.

[trevorstephens]

With this module in sklearn there will be several estimators, one for each target. You will need to access the estimators_ attribute which will be a list of n_output gplearn estimators which will each have a _program attribute hopefully 🤞

[galenseilis]

Any progress in understanding how to setup having multiple outputs since this thread was last updated?

[trevorstephens]

Also requested in #218

[trevorstephens]

No progress right now @galenseilis , going through the issue traker right now to determine what makes it to the next release though.