# In[4]:
with h5py.File(glob.glob('output/'+model_name+'*')[0] + '/result.h5','r') as results:
L, mu, W, sigma, pi = [results[key][()] for key in ['L',
'mu',
'W',
'sigma',
'pi']]
# In[15]:
# ## $H\pi$ (sparsity) over iterations
bsc = BSC_ET(*[model_specifications[key] for key in ['D','H','Hprime','gamma']])
data = {'y':data}
model_params = {'W':W[-1],'pi':pi[-1],'sigma':sigma[-1]}
anneal = LinearAnnealing(1)
anneal['T'] = [(0,1.)]
inferred_data = bsc.inference(anneal,model_params,data,no_maps=1)
#test_patches = np.reshape(data['y'],(data['y'].shape[0],)+patchsize)
#inferred_data = {key:np.squeeze(value) for key,value in inferred_data.iteritems()}
# In[10]:
#Increases variance by a muliplicative factor that slowly goes down to 1
anneal['T'] = T # [(iteration, value),... ]
#Reduces truncation rate so as not to prematurely exclude data
anneal['Ncut_factor'] = Ncut
#Simulated annealing of parameters
#anneal['W_noise'] = Wnoise
#Include prior parameters in the annealing schedule
anneal['anneal_prior'] = False
output_path = create_output_path(basename = job_name)
model = BSC_ET(D, H, Hprime, gamma, to_learn=['W','sigma','pi','mu'])
data = {'y':patches}
out_fname = output_path + "/data.h5"
#setting up logging/output
print_list = ('T', 'Q', 'pi', 'sigma', 'N', 'MAE', 'L')
dlog.set_handler(print_list, TextPrinter)
h5store_list = ('W', 'pi', 'sigma', 'y', 'MAE', 'N','L','Q','mu')
dlog.set_handler(h5store_list, StoreToH5, output_path +'/result.h5')
###### Initialize model #######
# Initialize (Random) model parameters
dlog.set_handler(["pi"], YTPlotter)
dlog.set_handler(["sigma"], YTPlotter)
dlog.set_handler("y", RFViewer, rf_shape=(D2, D2))
# Prepare ground-truth GFs (bars)
W_gt = np.zeros((H, D2, D2))
for i in xrange(D2):
W_gt[i, i, :] = bar_value
W_gt[D2 + i, :, i] = bar_value
if neg_bars > 0.0:
W_gt[sample(range(H), np.int(H * neg_bars))] *= -1
W_gt = W_gt.reshape((H, D))
W_gt += np.random.normal(size=(H, D), scale=0.5)
# Prepare model...
model = BSC_ET(D, H, Hprime, gamma, to_learn)
mparams = {"W": W_gt, "pi": pi_gt, "sigma": sigma_gt, "mu": mu_gt}
mparams = comm.bcast(mparams)
pprint("Generating Model Parameters:")
pprint("pi = " + np.str(mparams["pi"]) + "; sigma = " + np.str(mparams["sigma"]))
# Generate trainig data
my_N = N // comm.size
my_data = model.generate_data(mparams, my_N)
dlog.append("y", my_data["y"][0:20])
# Choose annealing schedule
anneal = LinearAnnealing(anneal_steps)
anneal["T"] = [(15, start_temp), (-10, end_temp)]
anneal["Ncut_factor"] = [(0, 0.0), (2.0 / 3, 1.0)]
#patches_flat = patches_flat - intercept
# In[6]:
output_path = create_output_path()
N = patches_flat.shape[0]
D = patches_flat.shape[1]
H = 40
Hprime = 6
gamma = 5
model = BSC_ET(D, H, Hprime, gamma, to_learn=['W','sigma','pi'])
data = {'y':patches_flat}
out_fname = output_path + "/data.h5"
# In[7]:
#setting up logging/output
print_list = ('T', 'Q', 'pi', 'sigma', 'N', 'MAE', 'L')
dlog.set_handler(print_list, TextPrinter)
h5store_list = ('W', 'pi', 'sigma', 'y', 'MAE', 'N','L','Q')
dlog.set_handler(['pi'], YTPlotter)
dlog.set_handler(['sigma'], YTPlotter)
dlog.set_handler('y', RFViewer, rf_shape=(D2, D2))
# Prepare ground-truth GFs (bars)
W_gt = np.zeros((H, D2, D2))
for i in xrange(D2):
W_gt[i, i, :] = bar_value
W_gt[D2 + i, :, i] = bar_value
if neg_bars > 0.0:
W_gt[sample(range(H), np.int(H * neg_bars))] *= -1
W_gt = W_gt.reshape((H, D))
W_gt += np.random.normal(size=(H, D), scale=0.5)
# Prepare model...
model = BSC_ET(D, H, Hprime, gamma, to_learn)
mparams = {'W': W_gt, 'pi': pi_gt, 'sigma': sigma_gt, 'mu': mu_gt}
mparams = comm.bcast(mparams)
pprint("Generating Model Parameters:")
pprint("pi = " + np.str(mparams['pi']) + "; sigma = " +
np.str(mparams['sigma']))
# Generate trainig data
my_N = N // comm.size
my_data = model.generate_data(mparams, my_N)
dlog.append('y', my_data['y'][0:20])
# Choose annealing schedule
anneal = LinearAnnealing(anneal_steps)
anneal['T'] = [(15, start_temp), (-10, end_temp)]
