added simple save/load pickle #2

thangbui · Mar 31, 2017 · c5ba7f3 · c5ba7f3
1 parent e56a31b
commit c5ba7f3
Show file tree

Hide file tree

Showing 4 changed files with 386 additions and 215 deletions.
diff --git a/examples/gpssm_hodgkin_huxley.py b/examples/gpssm_hodgkin_huxley.py
@@ -119,21 +119,22 @@ def model_V_n():
 	hypers = model_aep.init_hypers(y)
 	for key in params.keys():
 	    hypers[key] = params[key]
-	model_aep.update_hypers(hypers, alpha)
+	model_aep.update_hypers(hypers)
 	# optimise
 	# model_aep.set_fixed_params(['R', 'sn', 'sf'])
 	model_aep.set_fixed_params(['sf'])
 	model_aep.optimise(method='L-BFGS-B', alpha=alpha, maxiter=10000, reinit_hypers=False)
 	opt_hypers = model_aep.get_hypers()
 	plot_model(model_aep, 'AEP %.3f'%alpha)
 	plt.show()
+	model_aep.save_model('/tmp/gpssm_hh_VN.pickle')
 
 
 def model_all():
 	# load dataset
 	data = np.loadtxt('./sandbox/hh_data.txt')
 	# use the voltage and potasisum current
-	y = data
+	y = data[:, :4]
 	y = y / np.std(y, axis=0)
 	# init hypers
 	Dlatent = 2
@@ -152,7 +153,7 @@ def model_all():
 	hypers = model_aep.init_hypers(y)
 	for key in params.keys():
 	    hypers[key] = params[key]
-	model_aep.update_hypers(hypers, alpha)
+	model_aep.update_hypers(hypers)
 	# optimise
 	# model_aep.set_fixed_params(['R', 'sn', 'sf'])
 	model_aep.set_fixed_params(['sf'])
@@ -161,8 +162,46 @@ def model_all():
 	opt_hypers = model_aep.get_hypers()
 	plot_model(model_aep, 'AEP %.3f'%alpha)
 	plt.show()
+	model_aep.save_model('/tmp/gpssm_hh_all.pickle')
+
+
+def model_all_with_control():
+	# TODO: predict with control
+	# load dataset
+	data = np.loadtxt('./sandbox/hh_data.txt')
+	# use the voltage and potasisum current
+	y = data[:, :4]
+	xc = data[:, [-1]]
+	y = y / np.std(y, axis=0)
+	# init hypers
+	Dlatent = 2
+	Dobs = y.shape[1]
+	M = 30
+	T = y.shape[0]
+	R = np.ones(Dobs)*np.log(0.01)/2
+	lsn = np.log(0.01)/2
+	params = {'sn': lsn, 'R': R}
+
+	alpha = 0.4
+	print 'alpha = %.3f' % alpha
+	# create AEP model
+	model_aep = aep.SGPSSM(y, Dlatent, M, 
+	    lik='Gaussian', prior_mean=0, prior_var=1000)
+	hypers = model_aep.init_hypers(y)
+	for key in params.keys():
+	    hypers[key] = params[key]
+	model_aep.update_hypers(hypers)
+	# optimise
+	# model_aep.set_fixed_params(['R', 'sn', 'sf'])
+	model_aep.set_fixed_params(['sf'])
+	model_aep.optimise(method='L-BFGS-B', alpha=alpha, maxiter=30000, reinit_hypers=False)
+	# model_aep.optimise(method='adam', alpha=alpha, maxiter=10000, reinit_hypers=False, adam_lr=0.05)
+	opt_hypers = model_aep.get_hypers()
+	plot_model(model_aep, 'AEP %.3f'%alpha)
+	plt.show()
 
 
 if __name__ == '__main__':
 	model_all()
+	# model_all_with_control()
 	# model_V_n()
diff --git a/exps/gpssm/kink_exp.py b/exps/gpssm/kink_exp.py
@@ -63,7 +63,7 @@ def kink(T, process_noise, obs_noise, xprev=None):
 hypers = model.init_hypers(y_train)
 for key in params.keys():
     hypers[key] = params[key]
-model.update_hypers(hypers, alpha)
+model.update_hypers(hypers)
 model.set_fixed_params(['C'])
 model.optimise(method='L-BFGS-B', alpha=alpha, maxiter=np.inf, reinit_hypers=False)
 

diff --git a/exps/gpssm/lin_cos_exp.py b/exps/gpssm/lin_cos_exp.py
@@ -53,7 +53,7 @@ def func(T, process_noise, obs_noise, xprev=None):
 hypers = model.init_hypers(y_train)
 for key in params.keys():
     hypers[key] = params[key]
-model.update_hypers(hypers, alpha)
+model.update_hypers(hypers)
 model.set_fixed_params(['C', 'R'])
 model.optimise(method='L-BFGS-B', alpha=alpha, maxiter=10000, reinit_hypers=False)