def sample(preds, temperature = 1.0):
preds = np.asarry(preds).astype('float64')
preds = np.log(preds) / temperature
exp_preds = np.exp(preds)
preds = exp_preds / np.sum(exp_preds)
probas = np.random.multinomial(1, preds, 1)
return np.argmax(probas)
def d2v_infer_vecs(model, docs):
from gensim.models.doc2vec import Doc2Vec
import numpy as np
docs = spacy_tokenize(docs)
docvecs = np.asarry([model.infer_vector(docs[i]) for i in len(docs)])
return docvecs
def validation_loss(self,
overwrite_loss_func=False,
return_val_data=False):
self.model.eval()
if overwrite_loss_func:
loss_func = overwrite_loss_func
else:
loss_func = self.loss_func
if self.max_validation_steps == -1:
self.max_validation_steps = len(self.data_val)
if return_val_data:
return_val_data = []
val_loss = []
with torch.no_grad():
count = tqdm(
range(self.max_validation_steps),
desc="Validation",
disable=not self.verbose,
leave=True,
)
for data, i in zip(self.data_val, count):
data = data["data"].to(self.device)
pred = self.model(data)
val_loss.append(loss_func(pred, data).cpu().numpy())
if return_val_data:
return_val_data.append(data.cpu().numpy())
if return_val_data:
return np.asarray(val_loss), np.asarry(return_val_data)
return np.asarray(val_loss)
def __getitem__(self,
scan_index: int) -> Tuple[np.array, Tuple[int, int, int]]:
'''
returns:
- scan
- scan shape
'''
with h5py.File(self.scans[scan_index], mode='r') as f:
data = f[self.data_key]
return np.asarry(f), f.shape
def sort_hashes(query_hash, hashes):
"""
Given a dict of hogs:hashes, returns a sorted array of hogs and jaccard distances relative to query hog.
:param query hash: weighted minhash of the query
:param hashes: a dict of hogs:hashes
:return: sortedhogs, jaccard
"""
#sort the hashes by their jaccard relative to query hash
jaccard = [query_hash.jaccard(hashes[hog]) for hog in hashes]
index = np.argsort(jaccard)
sortedhogs = np.asarry(list(hashes.keys()))[index]
jaccard = jaccard[index]
return sortedhogs, jaccard
def load_next_image(self):
'''
Load the next image in a batch.
'''
if self._cur == len(self.indexlist):
self._cur =0
index = self.indexlist(self._cur)
#Choose the image type {e.g., .png, .jpg, .JPEG}.
im_file_name = index + '.JPEG'
im = np.asarry(Image)
def create_wrong(file_path):
folder = np.random.choice(glob.glob(file_path + "*"))
while floder == "fatalab":
folder = np.random.choice(glob.glob(file_path + "*"))
mat = np.zeros((480, 640), dtype='float32')
i = 0
j = 0
depth_file = np.random_choice(glob.glob(folder + "/*.dat"))
with open(depth_file) as file:
for line in file:
vals = line.split('\t')
for val in vals:
if val == "\n": continue
if int(val) > 1200 or int(val) == -1: val = 1200
mat[i][j] = float(int(val))
j += 1
j = j % 640
i += 1
mat = np.asarry(mat)
mat_small = mat[140:340, 220:420]
mat_small = (mat_small - np.mean(mat_small)) / np.max(mat_small)
plt.imshow(mat_small)
plt.show()
folder2 = np.random.choice(glob.glob(file_path + "*"))
while folder == folder2 or folder2 == "datalab": #it activates if it chose the same folder
folder2 = np.random.choice(glob.glob(file_path + "*"))
mat2 = np.zeros((480, 640), dtype='float32')
i = 0
j = 0
depth_file = np.random.choice(glob.glob(folder2 + "/*.dat"))
with open(depth_file) as file:
for line in file:
vals = line.split('\t')
for val in vals:
if val == "\n": continue
if int(val) > 1200 or int(val) == -1: val = 1200
mat2[i][j] = float(int(val))
j += 1
j = j % 640
i += 1
mat2 = np.asarray(mat2)
mat2_small = mat2[140:340, 220:420]
mat2_small = (mat2_small - np.mean(mat2_small)) / np.max(mat2_small)
plt.imshow(mat2_small)
plt.show()
return np.array([mat_small, mat2_small])
def bond_label_fraction(top_n=2):
"""
Get the fraction of each type of label 0 (low energy), 1 (high energy),
and 2 (unknown) in the dataset.
Note, this requires that when extracting the reactions, all the reactions
related to a reactant (ignore symmetry) should be extracted.
Note, here we analyze on the 0,0,0 type reactions.
"""
# filename = "~/Applications/db_access/mol_builder/reactions.pkl"
# filename = "~/Applications/db_access/mol_builder/reactions_n200.pkl"
# filename = "~/Applications/db_access/mol_builder/reactions_qc.pkl"
# filename = "~/Applications/db_access/mol_builder/reactions_qc_ws.pkl"
filename = "~/Applications/db_access/mol_builder/reactions_qc_ws_charge0.pkl"
extractor = ReactionCollection.from_file(filename)
groups = extractor.group_by_reactant_charge_0()
num_bonds = []
frac = defaultdict(list)
for ropb in groups:
counts = np.asarry([0, 0, 0])
all_none = True
for i, rxn in enumerate(ropb.order_reactions()):
energy = rxn.get_free_energy()
if energy is None:
counts[2] += 1
else:
all_none = False
if i < top_n:
counts[1] += 1
else:
counts[0] += 1
if all_none:
print(
"reactant {} {} has not broken bond reaction; should never happen"
.format(ropb.reactant.id, ropb.reactant.formula))
continue
n = len(ropb.order_reactions())
num_bonds.append(n)
frac = counts / n
print("### number of bonds in dataset (mean):", np.mean(num_bonds))
print("### number of bonds in dataset (median):", np.median(num_bonds))
for i, fr in enumerate(frac):
print(f"### label{i} bond ratio in dataset (mean): {np.mean(fr)}")
print(f"### label{i} bond ratio in dataset (mean): {np.median(fr)}")
def maloshell_cooling(A1, A2, Ea_1, Ea_2, c0, tau=132.5, T_room=21.1):
""" Approximates isomizeration rate during cooling, and assumes no
significant isomizeration after one e-fold (tau minutes) have past.
Default tau is taken from cooling one gallon of water
in an 8 quart stainless steel stock pot with a steel lid on.
Args:
A1 (float): Exponential prefactor for k1.
A2 (float): Exponential prefactor for k2.
Ea_1 (float): Activation energy for reaction 1.
Ea_2 (float): Activation energy for reaction 2.
c0 (np.ndarray): Initial condition vector.
tau (float): Cooling rate time scale in minutes. Default is 132.5 min.
T_room (float): Room temperature water is cooling in. Default is 21.1 C (70 F).
Returns:
function: Utilization function vector for c1, c2, c3
"""
# Make sure c0 is an array:
c0 = np.asarray(c0)
# Create time array with spacings of one minute.
t_arr = np.linspace(0., tau, np.ceil(tau) + 1)
# Grab the Newton Cooling function T(t) with T0=T_room and Ti=100 C
temp_func = NewtonCooling.T(T_room, 100., tau)
# Get rate functions:
k1_func = lambda t: reaction.arrhenius(Ea_1, A1)(temp_func(t))
k2_func = lambda t: reaction.arrhenius(Ea_2, A2)(temp_func(t))
def dcdt(c, t):
# Internal function to solve ODE. c = np.array([c1, c2, c3])
M = np.zeros((3, 3))
M[0, 0] = -reaction.RateEquations.dkn_dt(1, k1_func(t))(t)
M[1, 0] = reaction.RateEquations.dkn_dt(1, k1_func(t))(t)
M[1, 1] = -reaction.RateEquations.dkn_dt(1, k2_func(t))(t)
M[2, 1] = reaction.RateEquations.dkn_dt(1, k2_func(t))(t)
# print(M)
return np.dot(M, c)
c = odeint(dcdt, c0, t_arr)
# Extrapolation fill values for each component of vector c.
ext = [(c[0, i], c[-1, i]) for i in range(3)]
# Linear interpolation for each vector component. Assumes no utilization below t=0
# and fixed utilization above t value given as input.
return np.asarry(
[interp1d(t_arr, c[:, i], fill_value=ext[i]) for i in range(3)])
def getLayer(self, layer):
"""
Gets the given layer from the image
layer:- as string with one of 'rgb', 'l' or 'grad', selects which layer to retrieve
"""
if layer == 'rgb':
return self.rgbImage
elif layer == 'l':
return self.lumoImage
elif layer == 'grad':
return self.lumoGradient
else:
return np.asarry([[0]], dtype=np.int32)
def getAfPointTile(self, afPointIndex, layer):
"""
Returns the tile corresponding to the primary AF Point for the image, from the selected layer
afPointIndex:- the index of the AF point to retrieve the tile for
layer:- as string with one of 'rgb', 'l' or 'grad', selects which layer the tile comes from
returns:- a numpy array with the requested tile
"""
top, bottom, left, right = determineAfPointBox(afPointIndex)
if layer == 'rgb':
return self.rgbImage[top:bottom, left:right, :]
elif layer == 'l':
return self.lumoImage[top:bottom, left:right]
elif layer == 'grad':
return self.lumoGradient[top:bottom, left:right]
else:
return np.asarry([[0]], dtype=np.int32)
def create_couple_rgbd(file_path):
folder = np.random.choice(glob.glob(file_path + "*"))
while folder == "datalab":
folder = np.random.choice(glob.glob(file_path + "*"))
print(floder)
mat = np.zeros((480, 640), dtype='float32')
i = 0
j = 0
depth_file = np.random.choice(glob.glob(folder + "/*.dat"))
with open(depth_file) as file:
for line in file:
vals = line.split('\t')
for val in vals:
if val == "\n": continue
if int(val) > 1200 or int(val) == -1: val = 1200
mat[i][j] = float(int(val))
j += 1
j = j % 640
i += 1
mat = np.asarray(mat)
mat_small = mat[140:340, 220:420]
img = Image.open(depth_file[:-5] + "c.bmp")
img.thumbnail((640, 480))
img = np.asarray(img)
img = img[140:340, 220:420]
mat_small(mat_small - np.mean(mat_small)) / np.max(mat_small)
plt.imshow(mat_small)
plt.show()
plt.imshow(img)
plt.show()
mat2 = np.zeros((480, 640), dtype='float32')
i = 0
j = 0
depth_file = np.random.choice(glob.glob(floder + "/*.dat"))
with open(depth_file) as file:
for line in file:
vals = line.split('\t')
for val in vals:
if val == "\n": continue
if int(val) > 1200 or int(val) == -1: val = 1200
mat2[i][j] = float(int(val))
j += 1
j = j % 640
i += 1
mat2 = np.asarray(mat2)
mat2_small = mat2[140:340, 220:420]
img2 = Image.open(depth_file[:-5] + "c.bmp")
img2.thumbnail((640, 480))
img2 = np.asarry(img2)
img2 = img2[160:360, 240:440]
plt.imshow(img2)
plt.show()
mat2_small = (mat2_small - np.mean(mat2_small)) / np.max(mat2_small)
plt.imshow()
full1 = np.zeros((200, 200, 4))
full1[:, :, :3] = img[:, :, :3]
fill1[:, :, 3] = mat_small
full1 = np.zeros((200, 200, 4))
full1[:, :, :3] = img2[:, :, :3]
fill1[:, :, 3] = mat2_small
return np.array([fill1, full2])
# param_list = param_list[:11]
# labels
lbl_enc = preprocessing.LabelEncoder()
labels = lbl_enc.fit_transform(labels)
col_list = ['Class_%d'%i for i in xrange(1, 10)]
df = pd.DataFrame(index=train_ids)
i = 0
preds_train = []
for loss, ntree, param in param_list:
part_train = pd.read_csv("train_"+get_train_from_param(param, ntree), index_col=0)
#part_train.drop(['id'], axis=1)
#for col in col_list:
# df[col+'_%d'%i] = part_train[col]
preds_train.append(np.asarry(part_train))
i+=1
cv_search_param = False
if cv_search_param:
for j in xrange(100):
pass
else:
clf = linear_model.LogisticRegression('l1', C=0.1)
train = np.asarray(df)
clf.fit(train, labels)
df = pd.DataFrame(index=sample.id.values)
i = 0
preds_test = []
for loss, ntree, param in param_list:
def imp_solver(para, chain_mpo=None):
"""
:param para:
:param mps_init: init with this mps
:return:
"""
# --------------------------------------------------------
# initialization
# --------------------------------------------------------
el_list = para.el_list
vl_list = para.vl_list
# definte chain Ham
chain_ham = []
for i in range(0, 4):
ham = single_chain_ham(el_list[i], vl_list[i], 0.0)
chain_ham.append(ham)
int_gate = int_ham(para, para.tau / 2.0)
if (chain_mpo is None):
# chain mps
L_bath = len(el_list)
chain_mpo = []
for i in range(0, 4):
if (i % 2 == 0):
if_bath_sign = True
else:
if_bath_sign = False
mps_ = mps(para, L_bath + 1, if_bath_sign)
chain_mpo.append(mps_)
else:
for mpo in chain_mpo:
mpo.load_gnd() # load the gnd as init
mpo.if_find_gnd = False # set the gnd flag false
# TEBD solver
tebd_ = TEBD(para)
tebd_.init_state(chain_mpo, chain_ham, int_gate, para.tau / 2.0)
# --------------------------------------------------------
# imag time evolution to find gnd
# --------------------------------------------------------
Ntau = para.Ntau
max_en_diff = para.max_en_diff
En = 999999
i_check = 5 # check every i_check
for i in range(0, Ntau):
if (i % 20 == 0 and i != 0 and para.tau > 0.02):
para.tau = para.tau / 2.0
if (para.tau < 0.1):
i_check = 2
if (para.tau < 0.05):
i_check = 1
if (i % i_check == 0 or i == Ntau - 1):
En_new = tebd_.time_evolution(para.tau, if_calculate_gnd_en=True)
diff = np.abs(En - En_new)
En = En_new
if (diff < max_en_diff):
if (para.tau < 0.02):
print("Solved")
break
else:
para.tau = para.tau / 2.0
else:
tebd_.time_evolution(para.tau, if_calculate_gnd_en=False)
tebd_.save_gnd()
# get static quantitiy
tebd_.cal_static()
static_ = tebd_.static
En_gnd = En
En_diff = diff
if (diff > max_en_diff):
print("Not solved")
print("Energy diff is ", En_diff)
print("Gnd energy is ", En_gnd)
# --------------------------------------------------------
# real time evolution
# --------------------------------------------------------
gt = {}
for imp_ind in para.imp_list:
tebd_.load_gnd()
Nt = para.Nt
t_list = []
g_plus = [] # d_t d_dag
tebd_.act_d_dag(imp_ind)
for i in range(0, Nt):
if (i % para.i_meas == 0):
gt = tebd_.trace_with_d(imp_ind)
t_list.append(i * para.t)
g_plus.append(gt)
tebd_.time_evolution(-1j * para.t, if_calculate_gnd_en=False)
t_list = np.asarray(t_list)
g_plus = np.asarry(g_plus) * np.exp(1j * En_gnd * t_list)
tebd_.load_gnd()
Nt = para.Nt
t_list = []
g_minus = [] # d_t_dag d
tebd_.act_d(imp_ind)
for i in range(0, Nt):
if (i % para.i_meas == 0):
gt = tebd_.trace_with_d_dag(imp_ind)
t_list.append(i * para.t)
g_minus.append(gt)
tebd_.time_evolution(1j * para.t, if_calculate_gnd_en=False)
t_list = np.asarray(t_list)
g_minus = np.asarry(g_minus) * np.exp(-1j * En_gnd * t_list)
gt[imp_ind] = -1j * (g_plus + g_minus)
sol = {
't': t_list, # t list
'gt': gt, # gt dict
'static': static_, # static quantity (filling)
'mps': tebd_.chain_mpo, # mpo
'gnd_en': En_gnd, # Gnd energy
'en_diff': En_diff # energy diff
}
return sol
cfg.NET_G = net_G_name
algo.sample(dataloader, eval_name='eval',
eval_num=args.eval_num)
fid_score_now = \
fid_scores(output_dir, cfg, sample_num=args.sample_num,
gen_images_path=args.gen_paths, loop=True)
f.write('%s, %s, %.4f\n' % (date_str, net_G_name, fid_score_now))
f.close()
# Save the best FID score model
with open(os.path.join(output_dir, 'all_FID_eval.txt'), 'r') as f:
all_lines = f.readlines()
score_array =\
np.asarry([float(line.strip('\n').split(', ')[-1]) \
for line in all_lines])
if fid_score_now == score_array.min():
print("save the best FID score model, the score is %.4f" % \
fid_score_now)
net_G_name = all_lines[score_array.argmin()].split(', ')[1]
shutil.copy(os.path.join(output_dir, 'Model', net_G_name),
os.path.join(output_dir, 'model_reserve'))
elif args.inter_eval:
'''Choose two random variables and do interpolation'''
algo.sample(dataloader, eval_name='inter_eval', eval_num=args.eval_num)
elif args.LPIPS_eval:
'''Do LPIPS evaluation'''
f = open(os.path.join(output_dir, 'all_LPIPS_eval.txt'), 'a')
def fit_dist_extinct(wavelength,
observed_sed_nuFnu,
model,
error_observed_sed_nuFnu=None,
Rv=3.1,
modeldist=1.0,
additional_free=None,
logfit=False,
distance_range=[0.0, 1000.0],
model_noise_frac=0.1,
vary_av=True,
vary_distance=True,
av_range=[0.0, 10.0],
rv_range=[2.0, 20.0],
vary_rv=False,
**kwargs):
"""
Adapted from the fit_dist_extinct3.pro MCRE file designed to allow
distance and extinction to vary when computing the SED chsqr. For
a given inclination, this function returns the best fit distance,
extinction, Rv and chisqrd value.
Parameters
-----------
wavelength : Float Array
The wavelengths .... in microns
observed_sed_nuFnu: Quanity object in units W/m2
The nuFnu for the observed SED.
error_observed_sed_nuFnu : Float array
Error for the observed SED
model_noise_frac : Float
Noise fraction for the model SEDs
modeldist : Float
Distance to model disk in pc.
additional_free : Int
Number of additionaly free parameters to
include in the weighted chisqrd.
vary_distance : bool
Allow the distance to the target to vary?
Default is False
distance_range : 2 element iterable
Min and max allowable distance, in pc
vary_av : bool
Allow optical extinction (A_V) to vary?
Default is False
av_range : 2 element iterable
Min and max allowable A_V
vary_rv : bool
Allow rv to vary?
Default is False
Rv : float
Reddening law color parameter for extinction.
Default is 3.1
logfit : bool
Fit the SED on a logarithmic scale?
Default is False
"""
#wave_ang = [x*0.0001 for x in wavelength] #Convert from microns to angstroms
wave_ang = wavelength.to(units.Angstrom)
observed_sed_nuFnu = np.asarray(observed_sed_nuFnu)
# Scale model to 1pc
model_sed_1pc = np.asarray(model * modeldist**2)
if vary_distance:
a_distance = np.asarray([10] + distance_range)
else:
a_distance = np.asarray(modeldist)
if vary_av:
avsteps = (av_range[0] - av_range[1]) / 0.25
a_av = np.asarray([avsteps + 1] + av_range)
else:
a_av = np.asarray([0])
if vary_rv:
a_rv = np.asarray([10] + rv_range)
else:
a_rv = np.asarray([Rv])
if error_observed_sed_nuFnu:
err_obs = np.asarray(error_observed_sed_nuFnu)
else:
err_obs = np.asarray(0.1 * observed_sed_nuFnu)
if logfit:
ln_observed_sed_nuFnu = observed_sed_nuFnu
ln_err_obs = err_obs
subset = observed_sed_nuFnu != 0.0
ln_observed_sed_nuFnu[subset] = np.log(observed_sed_nuFnu[subset])
ln_err_obs[subset] = err_obs[subset] / observed_sed_nuFnu[subset]
# How many degrees of freedom?
dof = len(observed_sed_nuFnu)
chisqs = np.asarray([len(a_distance), len(a_av), len(a_rv)])
for i_r in range(0, len(a_rv) - 1):
ext = np.asarry(ccm_extinction(
a_rv[i_r], wave_ang)) # Use wavelength in Angstroms
for i_d in range(0, len(a_distance) - 1):
for i_a in range(0, len(a_av) - 1):
extinction = 10.0**((ext * a_av[i_a]) / (-2.5))
vout = (model_sed_1pc * extinction) / (a_distance[i_d])**2
if logfit:
ln_vout = vout
ln_vout[subset] = np.log(vout[subset])
chicomb = (ln_vout - ln_observed_sed_nuFnu)**2 / (
ln_err_obs**2 +
(ln_vout * np.log(model_noise_frac))**2)
else:
chicomb = (vout - observed_sed_nuFnu)**2 / (
err_obs**2 + (vout * model_noise_frac)**2)
chisqs[i_d, i_a, i_r] = np.asarray(
sum(chicomb) /
(dof - additonal_free)) #Normalize to Reduced chi square.
wmin = chisqs == min(chisqs)
sed_chisqr = min(chisqs)
best_distance = a_distance[wmin[0]]
best_av = a_av[wmin[1]]
best_rv = a_rv[wmin[2]]
return best_distance, best_av, best_rv, sed_chisqr
def fit_dist_extinct(wavelength, observed_sed_nuFnu, model, error_observed_sed_nuFnu = None, Rv=3.1, modeldist=1.0,
additional_free=None, logfit=False,
distance_range=[0.0,1000.0],model_noise_frac=0.1,
vary_av=True, vary_distance=True, av_range=[0.0,10.0],rv_range=[2.0,20.0],vary_rv=False, **kwargs):
"""
Adapted from the fit_dist_extinct3.pro MCRE file designed to allow
distance and extinction to vary when computing the SED chsqr. For
a given inclination, this function returns the best fit distance,
extinction, Rv and chisqrd value.
Parameters
-----------
wavelength : Float Array
The wavelengths .... in microns
observed_sed_nuFnu: Quanity object in units W/m2
The nuFnu for the observed SED.
error_observed_sed_nuFnu : Float array
Error for the observed SED
model_noise_frac : Float
Noise fraction for the model SEDs
modeldist : Float
Distance to model disk in pc.
additional_free : Int
Number of additionaly free parameters to
include in the weighted chisqrd.
vary_distance : bool
Allow the distance to the target to vary?
Default is False
distance_range : 2 element iterable
Min and max allowable distance, in pc
vary_av : bool
Allow optical extinction (A_V) to vary?
Default is False
av_range : 2 element iterable
Min and max allowable A_V
vary_rv : bool
Allow rv to vary?
Default is False
Rv : float
Reddening law color parameter for extinction.
Default is 3.1
logfit : bool
Fit the SED on a logarithmic scale?
Default is False
"""
#wave_ang = [x*0.0001 for x in wavelength] #Convert from microns to angstroms
wave_ang = wavelength.to(units.Angstrom)
observed_sed_nuFnu = np.asarray(observed_sed_nuFnu)
# Scale model to 1pc
model_sed_1pc = np.asarray(model * modeldist**2)
if vary_distance:
a_distance = np.asarray([10] + distance_range)
else:
a_distance = np.asarray(modeldist)
if vary_av:
avsteps = (av_range[0] - av_range[1])/0.25
a_av = np.asarray([avsteps+1]+av_range)
else:
a_av = np.asarray([0])
if vary_rv:
a_rv = np.asarray([10] + rv_range)
else:
a_rv = np.asarray([Rv])
if error_observed_sed_nuFnu:
err_obs = np.asarray(error_observed_sed_nuFnu)
else:
err_obs = np.asarray(0.1*observed_sed_nuFnu)
if logfit:
ln_observed_sed_nuFnu = observed_sed_nuFnu
ln_err_obs = err_obs
subset = observed_sed_nuFnu != 0.0
ln_observed_sed_nuFnu[subset] = np.log(observed_sed_nuFnu[subset])
ln_err_obs[subset] = err_obs[subset]/observed_sed_nuFnu[subset]
# How many degrees of freedom?
dof = len(observed_sed_nuFnu)
chisqs = np.asarray([len(a_distance),len(a_av),len(a_rv) ])
for i_r in range(0, len(a_rv)-1):
ext = np.asarry(ccm_extinction(a_rv[i_r],wave_ang)) # Use wavelength in Angstroms
for i_d in range(0, len(a_distance)-1):
for i_a in range(0, len(a_av)-1):
extinction = 10.0**((ext*a_av[i_a])/(-2.5))
vout = (model_sed_1pc*extinction)/(a_distance[i_d])**2
if logfit:
ln_vout = vout
ln_vout[subset] = np.log(vout[subset])
chicomb = (ln_vout-ln_observed_sed_nuFnu)**2/(ln_err_obs**2 + (ln_vout*np.log(model_noise_frac))**2)
else:
chicomb = (vout-observed_sed_nuFnu)**2/(err_obs**2 + (vout*model_noise_frac)**2)
chisqs[i_d, i_a, i_r] = np.asarray(sum(chicomb)/(dof-additonal_free)) #Normalize to Reduced chi square.
wmin = chisqs == min(chisqs)
sed_chisqr = min(chisqs)
best_distance = a_distance[wmin[0]]
best_av = a_av[wmin[1]]
best_rv = a_rv[wmin[2]]
return best_distance, best_av, best_rv, sed_chisqr
x_list = []
y_list = []
# 读取图片
image_path = ''
img = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
_, imgt = cv2.threshold(img, 140, 255, cv2.THRESH_BINARY_INV)
x_data = imgt / 255
y_data = ''
x_list.append(x_data)
y_list.append(y_data)
x_train, x_test, y_train, y_test = train_test_split(np.asarry(x_list),
np.asarray(y_list),
test_size=test_ratio,
random_state=42)
# 网络搭建
model = Sequential()
model.add(
Conv2D(16, kernel_size=(3, 3), activation='relu', input_shape=input_shape))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dense(64, activation='relu'))
model.add(Dense(code_classes, activation='softmax'))
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adadelta(),
def get_coefs(word, *arr) : return word, np.asarry(arr, dtype='float32')
embeddings_index = dict(get_coefs(*o.split(" ")) for o in open(EMBEDDING_FILE))
def get_coefs(word, *arr) : return word, np.asarry(arr, dtype='float32')
embedding_index = dict(get_coefs(*o.split(" ")) for o in open(EMBEDDING_FILE, encoding='utf8', erros='igonre') if len(o)>100)
def uniq_file_id(self):
return np.asarry(self.files['file_id'])
def _generate_bayes_net(self):
# 1. Start at any node (0)
# 2. At each node figure out the conditional prob
# 3. Add it to the new graph
# 4. Find unprocessed adjacent nodes
# 5. If any go to 2
# Else return the bayes net
# Will it be possible that zero is not root? If
# so, we need to pick one
root = 0
samples = np.asarry(self.samples)
self.bayes_net = nx.bfs_tree(self.spanning_graph, root)
for parent, child in self.bayes_net.edges():
parent_array = samples[:, parent]
# if node is not root, get probability of
# each gene appearing in parent
# Return an iterator over predecessor nodes of n
if not self.bayes_net.predecessors(parent):
freqs = np.histogram(parent_array, len(np.unique(parent_array)))[0]
# zip([a,b,c],[1,2,3]) -> [(a, 1), (...)]
parent_probs = dict(zip(np.unique(sub_child), freqs/(sum(freqs)*1.0)))
self.bayes_net.node[parent]["probabilities"] = {x:0 for x in range(len(self.samples))}
self.bayes_net.node[parent]["probabilities"].update(parent_nodes)
child_array = samples[:, child]
unique_parent = np.unique(parent_array)
for parent_val in unique_parents:
parent_inds = np.argwhere(parent_array == parent_val)
sub_child = child_array[parent_inds]
# Compute the histogram of a set of data.
freqs = np.histogram(sub_child, len(np.unique(sub_child)))[0]
child_probs = dict(zip(np.unique(sub_child), freqs/(sum(freqs)*1.0)))
self.bayes_net.node[child][parent_val] = {x:0 for x in range(len(self.samples))}
self.bayes_net.node[child][parent_val].update(parent_nodes)
self.bayes_net.node[child] = dict(probabilities=self.bayes_net.node[child])
# PRIM_MST Compute a minimum spanning with Prims's algorithm.
def _generate_spanning_graph(self):
return nx.prim_mst(self.complete_graph)
def _generate_mutual_information_graph(self):
samples = np.asarray(self.samples)
complete_graph = nx.complete_graph(samples.shape[1])
for edge in complete_graph.edges():
mutual_info = mutual_info_score(
samples[:, edge[0]],
samples[:, edge[1]]
)
complete_graph.edge[edge[0]][edge[1]]['weight'] = -mutual_info
return complete_graph