data = data.T
sd = np.array([[i for i in range(1,201)]], dtype='float64')
sd = sd.T
with model:
#The prior is a drawn from a uniform distribution from -5 to 15 for each of the 200 dimensions.
params = pm.Uniform('params', lower=-5.0, upper=15.0, shape=(200,1), dtype='float64')
#I can't get the multivariate distribution to run for some reason. I think a bunch of normal distributions with the correct mu and sds should work...
pm.Normal('true_dist', mu=params, sd=sd, observed=data)
#pm.MvNormal('true_dist', mu=params, tau=prec_matrix, observed=data)
#Algorithmic values given in paper (eps= b* in paper)
step = pm.Dream(DEpairs=1, snooker=.1, nseedchains=2000, eps=10e-6, nCR=3, multitry=5)
#njobs = number of chains
trace = pm.sample(400000, step, njobs=3)
dictionary_to_pickle = {}
for dictionary in trace:
for var in dictionary:
dictionary_to_pickle[var] = trace[var]
pickle.dump(dictionary_to_pickle, open('2015_03_17_mv_test_mtdreamzs.p', 'wb'))
print 'Saving trace to text...'
text.dump('2015_03_17_mv_test_mtdreamzs', trace)
print 'Trace saved successfully.'
td = (t10 + t90) / 2
ts = t90 - t10
yfinal = ysim_momp[-1]
momp_sim = [td, ts, yfinal]
e3 = np.sum(np.exp(-1*(momp_data - momp_sim) ** 2 / (2 * momp_var)))
error = -1.*np.log(e1) + -1.*np.log(e2) + -1.*np.log(e3)
#print error
return -1*error,
#return (e1, e2, e3,)
if "__main__" == __name__:
nm = Nightmare(EARM,likelihood,xnominal,'test')
nm.run_pso(4, 25,200)
ranked = nm.pso.return_ranked_populations()
savename = 'unweighted_cost_function_100k_std_1'
traces = nm.run_DREAM(nsamples=100000)
from pymc.backends import text
text.dump('traces_%s'%savename, traces)
dictionary_to_pickle = {}
for dictionary in traces:
for var in dictionary:
dictionary_to_pickle[var] = traces[var]
pickle.dump(dictionary_to_pickle, open('pickled_traces_%s.p'%savename, 'wb'))
pm.Normal('thermodynamic_box2', mu=box2, sd=1e-2, observed=1)
box3 = likelihood_thermobox3(model.KD_AA_cat1, model.KD_AA_cat2, model.KD_AG_allo1, model.KD_AG_allo2)
pm.Normal('thermodynamic_box3', mu=box3, sd=1e-2, observed=1)
box4 = likelihood_thermobox4(model.KD_AG_cat1, model.KD_AG_cat2, model.KD_AG_allo1, model.KD_AG_allo3)
pm.Normal('thermodynamic_box4', mu=box4, sd=1e-2, observed=1)
#Start from end of last trace
#start2 = {('KD_AA_cat1',np.log10(cox2_model.parameters['kr_AA_cat1'].value/cox2_model.parameters['kf_AA_cat1'].value)), ('kcat_AA1', np.log10(cox2_model.parameters['kcat_AA1'].value)), ('KD_AA_cat2',-2.76), ('kcat_AA2',.235), ('KD_AA_cat3',-2.265), ('kcat_AA3',.1760), ('KD_AG_cat1', np.log10(cox2_model.parameters['kr_AG_cat1'].value/cox2_model.parameters['kf_AG_cat1'].value)), ('kcat_AG1', np.log10(cox2_model.parameters['kcat_AG1'].value)), ('KD_AG_cat2',-.7495), ('KD_AG_cat3',-2.793), ('kcat_AG3',-.5095), ('KD_AA_allo1',2.290), ('KD_AA_allo2',.1001), ('KD_AA_allo3',-.3808), ('KD_AG_allo1',2.280), ('KD_AG_allo2', np.log10(cox2_model.parameters['kr_AG_allo2'].value/cox2_model.parameters['kf_AG_allo2'].value)), ('KD_AG_allo3',1.650)}
#Select MCMC stepping method
step = pm.Dream(nseedchains=120, blocked=True, start_random=True, save_history=True, parallel=False, multitry=5, adapt_crossover=True, model_name='CORM_mtdreamzs_5chain_profiling')
#old_trace = text.load('test')
#print 'old trace: ',old_trace['KD_AA_cat2']
trace = pm.sample(500, step, start=start, njobs=5)
dictionary_to_pickle = {}
for dictionary in trace:
for var in dictionary:
dictionary_to_pickle[var] = trace[var]
text.dump('CORM_mtdreamzs_5chain_profiling', trace)
pickle.dump(dictionary_to_pickle, open('CORM_mtdreamzs_5chain_profiling.p', 'wb'))
icrp = pm.Deterministic('icrp', icrp)
ecrp = pm.Deterministic('ecrp', ecrp)
momp = pm.Deterministic('momp', momp)
#error_like = pm.ArbLikelihood('like', error)
#Select point in parameter space to start
#start = pm.find_MAP()
#Select stepping method
nseedchains = 10*len(earm.parameters_rules())
step = pm.Dream(variables=[model.params], nseedchains=nseedchains, blocked=True, multitry=5, start_random=False, save_history=True, parallel=False, adapt_crossover=False, history_file='2015_04_30_earm_embedded_mtdreamzs_normal_prior_history.npy', crossover_file='2015_04_18_earm_embedded_mtdreamzs_normal_prior_crossovervals.npy')
old_trace = text.load('2015_04_30_earm_embedded_mtdreamzs_normal_prior')
trace = pm.sample(15000, step, njobs=3, trace=old_trace, use_mpi=False) #pass njobs=None to start multiple chains on different cpus
text.dump('2015_05_01_earm_embedded_mtdreamzs_normal_prior', trace)
dictionary_to_pickle = {}
for dictionary in trace:
for var in dictionary:
dictionary_to_pickle[var] = trace[var]
pickle.dump(dictionary_to_pickle, open('2015_05_01_earm_embedded_mtdreamzs_normal_prior.p', 'wb'))
from helper_fxns import convert_param_vec_dict_to_param_dict
from helper_fxns import merge_traces
from helper_fxns import print_convergence_summary
old_traces = pickle.load(open('2015_04_30_earm_embedded_mtdreamzs_normal_prior_merged_traces_80000.p'))
trace_list = [old_traces, dictionary_to_pickle]
return log_L
with pm.Model() as model:
params = pm.Flat('params', shape=(2))
#log_like = likelihood(model.x, model.y)
like = pm.ArbLikelihood('like', likelihood(model.params))
#like = pm.Potential('like', likelihood(model.params))
step = pm.Dream_mpi(blocked=True, start_random=False, save_history=True, parallel=True, history_file='ndim_banana_seed.npy', multitry=5)
start = [{'x':m[chain][0], 'y':m[chain][1]} for chain in range(3)]
trace = pm.sample(25000, step, start=start, njobs=3, use_mpi=True)
dictionary_to_pickle = {}
for dictionary in trace:
for var in dictionary:
dictionary_to_pickle[var] = trace[var]
text.dump('2015_04_22_ndim_banana_mtdreamzs', trace)
pickle.dump(dictionary_to_pickle, open('2015_04_22_ndim_banana_mtdreamzs.p', 'wb'))
erl_like = pm.ArbLikelihood('erl_output', erl)
no_erl_like = pm.ArbLikelihood('no_erl_output', no_erl)
erl = pm.Deterministic('erl', erl)
no_erl = pm.Deterministic('no_erl', no_erl)
nseedchains = 10*len(egfr.parameters_rules())
step = pm.Dream_mpi(variables=[model.params], nseedchains=nseedchains, blocked=True, multitry=5, start_random=True, save_history=True, parallel=False, adapt_crossover=True)
#old_trace = text.load('2015_04_29_earm_direct_mtdreamzs_normal_prior')
trace = pm.sample(100, step, njobs=5, use_mpi=True) #pass njobs=None to start multiple chains on different cpus
if rank == 0:
text.dump('2015_06_07_egfr_qualitative_calibration', trace)
dictionary_to_pickle = {}
for dictionary in trace:
for var in dictionary:
dictionary_to_pickle[var] = trace[var]
pickle.dump(dictionary_to_pickle, open('2015_06_07_egfr_qualitative_calibration.p', 'wb'))
from helper_fxns import convert_param_vec_dict_to_param_dict
from helper_fxns import merge_traces
from helper_fxns import print_convergence_summary
#old_traces = pickle.load(open('2015_04_29_earm_direct_mtdreamzs_normal_prior_merged_traces_80000.p'))
#trace_list = [old_traces, dictionary_to_pickle]
icrp = pm.Deterministic('icrp', icrp)
ecrp = pm.Deterministic('ecrp', ecrp)
momp = pm.Deterministic('momp', momp)
#error_like = pm.ArbLikelihood('like', error)
#Select point in parameter space to start
#start = pm.find_MAP()
#Select stepping method
nseedchains = 10*len(earm.parameters_rules())
step = pm.Dream(variables=[model.params], nseedchains=nseedchains, blocked=True, multitry=5, start_random=False, save_history=True, parallel=False, adapt_crossover=False, history_file='2015_04_30_earm_direct_mtdreamzs_normal_prior_history.npy', crossover_file='2015_04_18_earm_direct_mtdreamzs_normal_prior_crossovervals.npy')
old_trace = text.load('2015_04_30_earm_direct_mtdreamzs_normal_prior')
trace = pm.sample(15000, step, njobs=3, trace=old_trace, use_mpi=False) #pass njobs=None to start multiple chains on different cpus
text.dump('2015_05_01_earm_direct_mtdreamzs_normal_prior', trace)
#text.dump('test', trace)
dictionary_to_pickle = {}
for dictionary in trace:
for var in dictionary:
dictionary_to_pickle[var] = trace[var]
pickle.dump(dictionary_to_pickle, open('2015_05_01_earm_direct_mtdreamzs_normal_prior.p', 'wb'))
#pickle.dump(dictionary_to_pickle, open('test.p', 'wb'))
from helper_fxns import convert_param_vec_dict_to_param_dict
from helper_fxns import merge_traces
from helper_fxns import print_convergence_summary
#Select point in parameter space to start
starting_pts = np.ones(len(starting_vals)) * np.random.randn(3, len(starting_vals)) + starting_vals
for npt, pt in enumerate(starting_pts):
while np.any(pt < lower_limits) or np.any(pt > upper_limits):
starting_pts[npt] = np.ones(len(starting_vals)) * np.random.randn(1, len(starting_vals)) + starting_vals
starts = [{'params':starting_pts[chain]} for chain in range(3)]
starts[0]['params'] = starting_vals
#print 'starts: ',starts
#Select stepping method
nseedchains = 10*len(earm.parameters_rules())
step = pm.Dream(variables=[model.params], nseedchains=nseedchains, verbose=True, snooker = 0, blocked=True, start_random=True, save_history=True, parallel=False, model_name='earm_syn_dreamz_3chain_a')
#old_trace = text.load('2015_04_30_earm_direct_mtdreamzs_normal_prior')
trace = pm.sample(100000, step, njobs=3, use_mpi=False) #pass njobs=None to start multiple chains on different cpus
text.dump('earm_syn_dreamz_3chain_a', trace)
#text.dump('test', trace)
dictionary_to_pickle = {}
for dictionary in trace:
for var in dictionary:
dictionary_to_pickle[var] = trace[var]
pickle.dump(dictionary_to_pickle, open('earm_syn_dreamz_3chain_a.p', 'wb'))
#pickle.dump(dictionary_to_pickle, open('test.p', 'wb'))
# from helper_fxns import convert_param_vec_dict_to_param_dict
# from helper_fxns import merge_traces
# from helper_fxns import print_convergence_summary
#
def setUpClass(cls):
super(TestTextDumpFunction, cls).setUpClass()
text.dump(cls.name1, cls.mtrace1)
with cls.model:
cls.mtrace1 = text.load(cls.name1)
icrp = pm.Deterministic('icrp', icrp)
ecrp = pm.Deterministic('ecrp', ecrp)
momp = pm.Deterministic('momp', momp)
#Select point in parameter space to start
#start = pm.find_MAP()
#Select stepping method
nseedchains = 10*len(earm.parameters_rules())
step = pm.Dream(variables=[model.params], nseedchains=nseedchains, blocked=True, multitry=5, start_random=False, save_history=True, parallel=False, adapt_crossover=False)
#old_trace = text.load('2015_04_28_earm_embedded_mtdreamzs_normal_prior')
trace = pm.sample(150000, step, njobs=3, use_mpi=False) #pass njobs=None to start multiple chains on different cpus
text.dump('2015_04_11_earm_indirect_mtdreamzs_normal_prior', trace)
dictionary_to_pickle = {}
for dictionary in trace:
for var in dictionary:
dictionary_to_pickle[var] = trace[var]
pickle.dump(dictionary_to_pickle, open('2015_04_11_earm_indirect_mtdreamzs_normal_prior.p', 'wb'))
from helper_fxns import convert_param_vec_dict_to_param_dict
from helper_fxns import merge_traces
from helper_fxns import print_convergence_summary
#old_traces = pickle.load(open('2015_04_28_earm_embedded_mtdreamzs_normal_prior_merged_traces_50000.p'))
#trace_list = [old_traces, dictionary_to_pickle]
#Select point in parameter space to start
starting_pts = np.ones(len(starting_vals)) * np.random.randn(3, len(starting_vals)) + starting_vals
for npt, pt in enumerate(starting_pts):
while np.any(pt < lower_limits) or np.any(pt > upper_limits):
starting_pts[npt] = np.ones(len(starting_vals)) * np.random.randn(1, len(starting_vals)) + starting_vals
starts = [{'params':starting_pts[chain]} for chain in range(3)]
starts[0]['params'] = starting_vals
#print 'starts: ',starts
#Select stepping method
nseedchains = 10*len(earm.parameters_rules())
step = pm.Dream(variables=[model.params], nseedchains=nseedchains, verbose=True, blocked=True, start_random=False, save_history=True, parallel=False, model_name='earm_mtdreamzs_3chain_uni')
#old_trace = text.load('2015_04_30_earm_direct_mtdreamzs_normal_prior')
trace = pm.sample(30000, step, start=starts, njobs=3, use_mpi=False) #pass njobs=None to start multiple chains on different cpus
text.dump('earm_mtdreamzs_3chain_uni', trace)
#text.dump('test', trace)
dictionary_to_pickle = {}
for dictionary in trace:
for var in dictionary:
dictionary_to_pickle[var] = trace[var]
pickle.dump(dictionary_to_pickle, open('earm_mtdreamzs_3chain_uni.p', 'wb'))
#pickle.dump(dictionary_to_pickle, open('test.p', 'wb'))
# from helper_fxns import convert_param_vec_dict_to_param_dict
# from helper_fxns import merge_traces
# from helper_fxns import print_convergence_summary
#
return log_L
with pm.Model() as model:
params = pm.Flat('params', shape=(100))
#log_like = likelihood(model.x, model.y)
like = pm.ArbLikelihood('like', likelihood(model.params))
#like = pm.Potential('like', likelihood(model.params))
step = pm.Dream_mpi(blocked=True, start_random=False, save_history=True, history_file='ndim_gaussian_seed.npy', multitry=5, parallel=True)
start = [{'params':m[chain]} for chain in range(3)]
trace = pm.sample(50000, step, start=start, njobs=3, use_mpi=True)
dictionary_to_pickle = {}
for dictionary in trace:
for var in dictionary:
dictionary_to_pickle[var] = trace[var]
text.dump('2015_04_22_ndim_gaussian_mtdreamzs', trace)
pickle.dump(dictionary_to_pickle, open('2015_04_22_ndim_gaussian_mtdreamzs.p', 'wb'))
#error_like = pm.ArbLikelihood('like', error)
#Select point in parameter space to start
#start = pm.find_MAP()
#Select stepping method
nseedchains = 10*len(earm.parameters_rules())
step = pm.Dream_mpi(variables=[model.params], nseedchains=nseedchains, blocked=True, multitry=5, start_random=False, save_history=True, parallel=False, adapt_crossover=False)
#old_trace = text.load('2015_04_29_earm_direct_mtdreamzs_normal_prior')
trace = pm.sample(150, step, njobs=3, use_mpi=True) #pass njobs=None to start multiple chains on different cpus
if rank == 0:
#text.dump('2015_04_30_earm_direct_mtdreamzs_normal_prior', trace)
text.dump('test', trace)
dictionary_to_pickle = {}
for dictionary in trace:
for var in dictionary:
dictionary_to_pickle[var] = trace[var]
#pickle.dump(dictionary_to_pickle, open('2015_04_30_earm_direct_mtdreamzs_normal_prior.p', 'wb'))
pickle.dump(dictionary_to_pickle, open('test.p', 'wb'))
from helper_fxns import convert_param_vec_dict_to_param_dict
from helper_fxns import merge_traces
from helper_fxns import print_convergence_summary
#old_traces = pickle.load(open('2015_04_29_earm_direct_mtdreamzs_normal_prior_merged_traces_80000.p'))
