89 lines
1.8 KiB
Python
89 lines
1.8 KiB
Python
import numpy as np
|
|
from function import *
|
|
import wandb
|
|
import math
|
|
|
|
z_history = []
|
|
x1_history = []
|
|
x2_history = []
|
|
# fName = 'semicircle'
|
|
# fName = 'rosenbrock'
|
|
fName = 'semicircle2'
|
|
fName = 'himmelblau'
|
|
dfName = 'df_'+fName
|
|
def objective():
|
|
x1 = 0.1
|
|
x2 = 0.1
|
|
lr = 0.01
|
|
# x1 = trial.suggest_float("x1", 0, 3)
|
|
# x2 = trial.suggest_float("x2", 0, 3)
|
|
beta = 0.5
|
|
# lr = trial.suggest_float("lr", 0, 0.1)
|
|
epochs = 15
|
|
|
|
z = eval(fName)(x1, x2)
|
|
z_history.append(z)
|
|
|
|
v1,v2 = 0,0
|
|
for epoch in range(epochs):
|
|
dx1, dx2 = eval(dfName)(x1, x2)
|
|
v1 = beta * v1 + (1-beta)* dx1
|
|
v2 = beta * v2 + (1-beta) * dx2
|
|
|
|
x1 -= lr * v1
|
|
x2 -= lr * v2
|
|
|
|
x1_history.append(x1)
|
|
x2_history.append(x2)
|
|
z = eval(fName)(x1, x2)
|
|
z_history.append(z)
|
|
wandb.log({
|
|
'loss': z,
|
|
})
|
|
|
|
return eval(fName)(x1, x2)
|
|
|
|
def RMS():
|
|
x1 = 0.1
|
|
x2 = 0.1
|
|
lr = 0.2
|
|
# x1 = trial.suggest_float("x1", 0, 3)
|
|
# x2 = trial.suggest_float("x2", 0, 3)
|
|
beta = 0.5
|
|
# lr = trial.suggest_float("lr", 0, 0.1)
|
|
epochs = 15
|
|
|
|
z = eval(fName)(x1, x2)
|
|
z_history.append(z)
|
|
|
|
s1 ,s2 = 0,0
|
|
for epoch in range(epochs):
|
|
|
|
dx1, dx2 = eval(dfName)(x1, x2)
|
|
|
|
s1 = beta*s1+(1-beta)* dx1 ** 2
|
|
s2 = beta*s2+(1-beta)* dx2 ** 2
|
|
|
|
lr1,lr2 = lr/ math.sqrt(s1),lr/ math.sqrt(s2)
|
|
x1 -= lr1 * dx1
|
|
x2 -= lr2 * dx2
|
|
|
|
x1_history.append(x1)
|
|
x2_history.append(x2)
|
|
z = eval(fName)(x1, x2)
|
|
z_history.append(z)
|
|
wandb.log({
|
|
'loss': z,
|
|
})
|
|
return eval(fName)(x1, x2)
|
|
|
|
def main():
|
|
RMS()
|
|
plotAll(fName,np.array(x1_history), np.array(x2_history))
|
|
|
|
|
|
|
|
if __name__ == '__main__':
|
|
wandb.init(project="优化算法", name="RMS")
|
|
main()
|
|
wandb.finish() |