Many model-fitting Python packages have a predict() method, which outputs a prediction of the fitted model given observations of the predictor(s).
Question: How would I use these predict() methods to predict a single value when the observation is a variable in a gekko model?
Below is a very simple reproducible example:
Note: The actual model that I am fitting is a B-spline using statsmodels.gam.smooth_basis.BSplines and statsmodels.gam.generalized_additive_model.GLMGam. However, I am hoping that this simple example with sklearn.linear_model.LinearRegression will translate to more complex model classes from other packages.
from sklearn.linear_model import LinearRegression
import numpy as np
import matplotlib.pyplot as plt
from gekko import GEKKO
# create example data
x = np.arange(100)[:, np.newaxis]
y = np.arange(100) * 2 + 10
plt.plot(x, y) # plot x vs y data
# plt.show()
model = LinearRegression() # instantiate linear model
model.fit(x, y) # fit model
x_predict = np.arange(100, 200)[:, np.newaxis] # create array of predictor observations
y_predict = model.predict(x_predict) # use model to make prediction
plt.plot(x_predict, y_predict) # plot prediction
# plt.show()
m = GEKKO() # instantiate gekko model
x2 = m.FV() # instantiate free variable
x2.STATUS = 1 # make variable available for solver
y2 = 50 # true value
# place x2 variable in numpy array to adhere to predict()'s argument requirements
x2_arr = np.array(x2).reshape(1, -1)
# minimize squared error between the true value and the model's prediction
m.Minimize((y2 - model.predict(x2_arr)) ** 2)
m.options.IMODE = 3
# solve for x2
m.solve(disp=True)
print(f"x2 = {x2.value[0]:3f}")
I get the following sequential errors:
TypeError: float() argument must be a string or a real number, not 'GK_FV'
ValueError: setting an array element with a sequence.
My first thought is that I would have to create a wrapper class around the gekko.gk_parameter.GK_FV class to modify the float() method, but that's where my knowledge and skills end.
The Gekko ML functions can import sklearn, gpflow, and tensorflow models. It requires the models to be fit and then the parameters and equations are imported into the gekko model.
Gekko_Model = ML.Gekko_LinearRegression(model,Xtrain,RMSE,Gekko_Model)