rmsefname = "dump/RMSE_" + fname + "_" + str(total_states) + "states_" + str(num_iterations) + "iter" + ".json"
if os.path.exists(rmsefname):
rmsel = json.load(open(rmsefname, "r"))
else:
rmsel = []
rmsel.append(rmse)
json.dump(rmsel, open(rmsefname, "w"))
paramfname = "dump/PARAMS_" + fname + "_" + str(total_states) + "states_" + str(num_iterations) + "iter" + ".json"
if os.path.exists(paramfname):
paraml = json.load(open(paramfname, "r"))
else:
paraml = []
p = model.get_parameters()
pdict = {}
for id, param in p.iteritems():
if id == "T":
pdict[id] = param.get()[0, 1]
elif id == "L":
pdict[id] = [param.get()[0], param.get()[1]]
else:
pdict[id] = param.get()
pdict["Trans"] = list([list(x) for x in model.make_transitions()])
pdict["Pi"] = list(model.make_initial())
model.emission_mask[0] = False
pdict["Emit"] = list([list(x) for x in model.make_emissions(0, 0)])
paraml.append(pdict)
json.dump(paraml, open(paramfname, "w"))
def run_learned_model(skill, diff_params = None):
intermediate_states = 0
fname = skill.replace(" ","_")
fname = fname.replace("\"","")
X = np.loadtxt(open("dump/observations_" + fname + ".csv", "rb"), delimiter=",")
P = np.loadtxt(open("dump/problems_" + fname + ".csv","rb"),delimiter=",")
k = 5
#split 1/kth into test set
N = X.shape[0]
Xtest = []
Xnew = []
Ptest = []
Pnew = []
for c in range(N):
if c % k == 0:#random.random() < 1 / (k+0.0):
Xtest.append(X[c,:])
Ptest.append(P[c,:])
else:
Xnew.append(X[c,:])
Pnew.append(P[c,:])
X = Xnew
Xtest = np.array(Xtest)
P = Pnew
Ptest = np.array(Ptest)
model = MLFKTModel(X, P, 0, 0.1)
predl = []
errl = []
for c in range(3):
param_dict = json.load(open("feb20_exps/PARAMS_"+skill+"_2states_500iter.json","r"))
param_dict = param_dict[c]
params = model.get_parameters()
for k, v in param_dict.iteritems():
#print k, v
if k == "Pi":
val = np.array(v)
params["L"].set(val)
params["L"].save()
elif k == "Trans":
val = np.array(v)
params["T"].set(val)
params["T"].save()
elif k == "Emit":
G = scipy.special.logit(v[0][1])
S = scipy.special.logit(v[1][0])
params["G_0"].set(G)
params["S"].set(S)
params["G_0"].save()
params["S"].save()
else:
if diff_params is None:
params[k].set(v)
params[k].save()
else:
params[k].set(diff_params[k])
params[k].save()
params['Dsigma'].save()
model.load_test_split(Xtest, Ptest)
preds = model.get_predictions()
err = preds - Xtest
predl.append(preds)
errl.append(err)
return Xtest, Ptest, np.mean(predl,0), np.mean(errl,0)
