How to handle conditional constraints in Gekko nonlinear optimization ??

[tazuddin90]

Variable x1, X2;

Constraints:(with if statement)

If x1/x2>2: x1+x2>=10; Else: x1-x2>=5

Minimize: x1+x2+5

[APMonitor]

Use the m.if3() switch function to produce a 1 (True) or 0 (False) for the if statement. This binary switch is used to turn on or off equations.

from gekko import GEKKO

m = GEKKO()
x1,x2 = m.Array(m.Var,2,lb=-10,ub=10)
b = m.if3(x1/x2-2,0,1)         # binary switch
m.Equation(b*(x1+x2-10)>=0)    # x1+x2>=10 with True switch
m.Equation((1-b)*(x1-x2-5)>=0) # x1-x2>=5 with False switch
m.Minimize(x1+x2+5)
m.solve()

print(b.value[0])
print(x1.value[0])
print(x2.value[0])

Please ask future questions such as this on Stack Overflow with tag [gekko]. GitHub issues are typically geared towards feature enhancements or to report bugs with the software.

[tazuddin90]

Thank you very much for your help. Now it is working fine..

[tazuddin90]

Is it possible to use function as a constraint ?

Eg. Input Variables - x1, x2, x3

def fun(x1,x2,x3): if (x1>2): return x1+x2+x3+5; else: return x1-x2-3

I want to use this function as constraints like this.

m.Equation( fun(x1,x2,x3)>=2)

Minimize x1x2x3.

[APMonitor]

It isn't a problem to use a function but the condition needs to be as a m.if3() function. Here is additional information on conditional statements: https://apmonitor.com/wiki/index.php/Main/ConditionalStatements

def fun(x1,x2,x3):
   return m.if3(2-x1,x1+x2+x3+5,x1-x2-3)

m.Equation(fun(x1,x2,x3)>=2)
m.Minimize(x1*x2*x3)

[tazuddin90]

Thank you very much for your reply. Actually my function is much more complicated in real case. I am unable to put it in this format

The function is as follows-

Function defination

def fun (x1,x2,x3):
      import tflite_runtime.interpreter as tflite
      interpreter = tf.lite.Interpreter (model_path = "modelpath/fire_lite_model.tflite")

      interpreter.allocate_tensors()

      input_details = interpreter.get_input_details()
      output_details = interpreter.get_output_details()

      input = [[5*x1, 2*x2, 3*x5]]
      input = np.float32(input)

      interpreter.set_tensor(input_details[0]['index'], input)
      interpreter.invoke()

      output_data = interpreter.get_tensor(output_details[0]['index'])

      return output_data

The objective function is also like this.

Is it possible to handle using GEKKO ?

[APMonitor]

Gekko needs to have all of the equations expressed with Gekko variables and equations so that it can compile the equations into bytecode and provide exact derivatives to the gradient-based solvers. Black box models are not allowed with Gekko. If it is 2 variables then you can create a b-spline or if the function is one variable then you can create a c-spline to approximate the original function. Your function has 3 variables so there isn't currently an approximation method for those. Maybe try Scipy.optimize.minimize if you do need to optimize with a black box model.

[tazuddin90]

Thank you very much for your reply. Actually my function is much more complicated in real case. I am unable to put it in this format

The function is as follows-

Function defination

Variables = x1, x2,x3 def fun (x1,x2,x3): import tflite_runtime.interpreter as tflite

  interpreter = tf.lite.Interpreter (model_path = "modelpath/fire_lite_model.tflite")

  interpreter.allocate_tensors()

  input_details = interpreter.get_input_details()
  output_details = interpreter.get_output_details()

  input = [[5x1, 2x2, 3x5]]

  interpreter.set_tensor(input_details[0]['index'], input)
  interpreter.invoke()

  output_data = interpreter.get_tensor(output_details[0]['index'])

  return output_data

The objective function is also like this.

Is it possible to handle using GEKKO ?

Thank you so much for the info. I will try with scipy.