def plotpoledata():
fcorner = "results/plots/poledata_corner2D.csv"
fin = "results/plots/poledata_inside2D.csv"
import matplotlib.pyplot as plt
X1, X2 = tools.readData(fcorner)
plt.scatter(X1[1:], X2[1:], label="$X^{(corner)}$", s=5)
# print(np.c_[X1[1:10],X2[1:10]])
X1, X2 = tools.readData(fin)
plt.scatter(X1[1:500000], X2[1:500000], label="$X^{(in)}$", s=1)
plt.legend()
plt.savefig("results/plots/Ppoledata2D.png")
X1, X2 = tools.readData(fin)
def plotsamplingstrategies():
import matplotlib.pyplot as plt
import matplotlib as mpl
folder = "results/exp1/benchmarkdata/f1_"
lhsfile = folder + "lhs.txt"
splilhsfile = folder + "splitlhs.txt"
sgfile = folder + "sg.txt"
for file, name in zip([lhsfile, splilhsfile, sgfile],
["LHS", "d-LHS", "SG"]):
X, Y = tools.readData(file)
mpl.rc('text', usetex=True)
mpl.rc('font', family='serif', size=12)
mpl.rc('font', weight='bold')
mpl.rcParams['text.latex.preamble'] = [
r'\usepackage{sfmath} \boldmath'
]
# mpl.style.use("ggplot")
plt.figure(0, figsize=(15, 10))
plt.scatter(X[:, 0], X[:, 1], s=100, c='r')
plt.xlabel("$x_1$", fontsize=44)
plt.ylabel("$x_2$", fontsize=44)
plt.tick_params(labelsize=28)
plt.savefig("../../log/" + name + ".pdf", bbox_inches='tight')
plt.clf()
plt.close('all')
def runRappsipStrategy2(infile,runs, larr,l1strat="ho_p_q",box=np.array([[-1,1],[-1,1]]),trainingScale="0.5x",outfile="out.json",debug=0):
# l1strat="ho_p_q"
# l1strat="all_p_q"
X, Y = tools.readData(infile)
outJSON = {}
# runs = [[2,2],[3,?3],[4,4],[5,5],[6,6]]
for r in runs:
for l in larr:
pdeg=r[0]
qdeg=r[1]
if(l1strat == "ho_p_q"):
ppenaltybin = np.ones(pdeg+1)
ppenaltybin[pdeg] = 0
qpenaltybin = np.ones(qdeg+1)
qpenaltybin[qdeg] = 0
elif(l1strat == "all_p_q"):
ppenaltybin = np.zeros(pdeg+1)
qpenaltybin = np.zeros(qdeg+1)
rappsip = RationalApproximationSIP(
X,
Y,
m=pdeg,
n=qdeg,
trainingscale=trainingScale,
roboptstrategy="baron",
box=box,
strategy=2,
penaltyparam=l,
ppenaltybin=ppenaltybin.tolist(),
qpenaltybin=qpenaltybin.tolist()
)
outJSON["p%s_q%s_%.E"%(str(pdeg),str(qdeg),l)] = rappsip.asDict
if(debug == 1):
import json
with open("/tmp/s2_latest.json", "w") as f:
json.dump(outJSON, f,indent=4, sort_keys=True)
import json
with open(outfile, "w") as f:
json.dump(outJSON, f,indent=4, sort_keys=True)
import json
with open(fname, "w") as f:
json.dump(self.asDict, f, indent=indent, sort_keys=sort_keys)
def printDebug(self, msg):
import datetime
print("[d%d p%d q%d ts%s] [[%s]] %s" %
(self._dim, self._m, self._n, self._trainingscale,
datetime.datetime.now().strftime("%Y-%m-%d %H:%M"), msg))
if __name__ == "__main__":
import sys
infilePath11 = "../benchmarkdata/f16.txt"
# infilePath1 = "../benchmarkdata/f1_noise_0.1.txt"
X, Y = tools.readData(infilePath11)
r = RationalApproximationSIPONB(
X,
Y,
m=2,
n=3,
trainingscale="Cp",
roboptstrategy='ms',
localoptsolver='scipy',
fitstrategy='filter',
strategy=0,
)
r2 = apprentice.RationalApproximationSIP(
X,
Y,
def runCrossValidation(infile,box=np.array([[-1,1],[-1,1]]),outfile="out.json",debug=0):
trainingScale = "Cp"
X, Y = tools.readData(infile)
# Some param overrides for debug
larr = np.array([10**i for i in range(3,-13,-1)])
# larr = np.array([10**i for i in range(0,-5,-1)])
k=10
# k=2
outJSON = {}
for pdeg in range(2,5):
# for pdeg in range(3,5):
ppenaltybin = np.zeros(pdeg+1)
for qdeg in range(2,5):
# for qdeg in range(3,5):
qpenaltybin = np.zeros(qdeg+1)
avgerror = np.zeros(len(larr))
avgerror_k = np.zeros(len(larr))
for index in range(len(larr)):
l = larr[index]
kfold = KFold(k)
error_l = 0
for train, test in kfold.split(X):
rappsip = RationalApproximationSIP(
X[train],
Y[train],
m=pdeg,
n=qdeg,
trainingscale=trainingScale,
box=box,
strategy=2,
penaltyparam=l,
ppenaltybin=ppenaltybin.tolist(),
qpenaltybin=qpenaltybin.tolist()
)
error_l_k = np.sum([(rappsip(X[test])-Y[test])**2])
error_l += error_l_k
avgerror[index] = error_l / len(X)
avgerror_k[index] = error_l / len(X[test])
stderror = np.std(avgerror_k)/np.sqrt(k)
minIndex = np.argmin(avgerror)
minv = avgerror[minIndex]
minl = larr[minIndex]
currIndex = minIndex
while currIndex >= 0:
if(minv + stderror == avgerror[currIndex]):
break
elif(minv + stderror > avgerror[currIndex]):
currIndex -= 1
else:
currIndex += 1
break
if(currIndex == -1):
currIndex = 0
currl = larr[currIndex]
# print(minIndex)
# print(currIndex)
#
# print(avgerror)
# print(avgerror_k)
# print(stderror)
#
# print(minl)
# print(currl)
rappsip_min = RationalApproximationSIP(
X,
Y,
m=pdeg,
n=qdeg,
trainingscale=trainingScale,
box=box,
strategy=2,
penaltyparam=minl,
ppenaltybin=ppenaltybin.tolist(),
qpenaltybin=qpenaltybin.tolist()
)
rappsip_minpse = rappsip_min
if(currl != minl):
rappsip_minpse = RationalApproximationSIP(
X,
Y,
m=pdeg,
n=qdeg,
trainingscale=trainingScale,
box=box,
strategy=2,
penaltyparam=currl,
ppenaltybin=ppenaltybin.tolist(),
qpenaltybin=qpenaltybin.tolist()
)
rappsip = {"min":rappsip_min.asDict, "min plus SE":rappsip_minpse.asDict, "avgerror":avgerror.tolist(),
"avgerror_k":avgerror_k.tolist(), "stderror":stderror,"minIndex":minIndex,"minl":minl,
"minv":minv, "mpseIndex":currIndex, "mpsel":currl}
outJSON["p%s_q%s"%(str(pdeg),str(qdeg))] = rappsip
if(debug == 1):
import json
with open("/tmp/cv_latest.json", "w") as f:
json.dump(outJSON, f,indent=4, sort_keys=True)
# exit(1)
import json
with open(outfile, "w") as f:
json.dump(outJSON, f,indent=4, sort_keys=True)
def runRappsipBaseStrategy(infile,runs, box=np.array([[-1,1],[-1,1]]),trainingScale="1x", roboptstrategy="ms",outfile="out.json",debug=0,debugfile="/tmp/s0_latest.json"): X, Y = tools.readData(infile) return runRappsipBaseStrategyFromPoints(X,Y,runs, box,trainingScale, roboptstrategy,outfile,debug,debugfile)