def create_mask(self,u0):
st=self.st
print('u0.shape',u0.shape)
rows=u0.shape[0]
cols=u0.shape[1]
kk = xrange(0,rows)
jj = xrange(0,cols)
kk = CsTransform.pynufft.appendmat(kk,cols)
jj = CsTransform.pynufft.appendmat(jj,rows).T
st['mask']=numpy.ones((rows,cols),dtype=numpy.float32)
#add circular mask
sp_rat=(rows**2+cols**2)*1.0
# for jj in xrange(0,cols):
# for kk in xrange(0,rows):
# if ( (kk-rows/2.0)**2+(jj-cols/2.0)**2 )/sp_rat > 1.0/8.0:
# st['mask'][kk,jj] = 0.0
if numpy.size(u0.shape) > 2:
for pp in range(2,numpy.size(u0.shape)):
st['mask'] = CsTransform.pynufft.appendmat(st['mask'],u0.shape[pp] )
return st
def simulate(self, prices, molecule, span):
"""
Derive labels from the sign of t-value of the linear trend
:param prices: Time series of {x_t}
:type prices: array like
:param molecule: the index of the observations we wish to label
:type molecule: array like
:param span: set of values of L, the look forward period
:type span: array like
"""
out = pd.DataFrame(index=molecule, columns=['tl', 'tval', 'bin'])
horizons = np.xrange(*span)
for dt in molecule:
df = pd.Series()
iloc = prices.index.get_loc(dt)
if iloc + max(horizons) > prices.shape[0]:
continue
for horizon in horizons:
dt1 = prices.index[iloc + horizon - 1]
df1 = prices.loc[dt:dt1]
df.loc[dt1] = t_value_lin(df1.values)
dt1 = df.replace([-np.inf, np.inf, np.nan], 0).abs().idxmax()
out.loc[dt,
['tl', 'tval', 'bin']] = df.index[-1], df[dt1], np.sign(
df[dt1])
out['tl'] = pd.to_datetime(out['tl'])
out['bin'] = pd.to_numeric(out['bin'], downcast='signed')
return out.dropna(subset=['bin'])
def compFq(rms, qs):
"""Compute scaling function F as:
F[scale] = pow(mean(RMS[scale]^q),1.0/q)
This function computes F for all qs at each scale.
The result is a 2d NxM array (N = rms.shape[0], M = len(qs))
Parameters
----------
rms: the RMS 2d array (RMS for scales in rows) computer by compRMS or fastRMS
qs: an array of q coefficients
Example
-------
# >>> X = cumsum(0.1*randn(8000))
# >>> scales = (2**arange(4,10)).astype('i4')
# >>> RMS = fastRMS(X,scales)
# >>> qs = arange(-5,5.1,1.0)
# >>> loglog(scales,compFq(RMS,qs),'.-')
"""
out = zeros((rms.shape[0], len(qs)), 'f8')
mRMS = np.ma.array(rms, mask=np.isnan(rms))
for qi in np.xrange(len(qs)):
p = qs[qi]
out[:, qi] = (mRMS**p).mean(1)**(1.0 / p)
out[:, qs == 0] = np.exp(0.5 * (np.log(mRMS**2.0)).mean(1))[:, None]
return out
def do_eval(sess,
eval_correct,
images_placeholder,
labels_placeholder,
data_set):
"""Runs one evaluation against the full epoch of data.
Args:
sess: The session in which the model has been trained.
eval_correct: The Tensor that returns the number of correct predictions.
images_placeholder: The images placeholder.
labels_placeholder: The labels placeholder.
data_set: The set of images and labels to evaluate, from
input_data.read_data_sets().
"""
# And run one epoch of eval.
true_count = 0 # Counts the number of correct predictions.
steps_per_epoch = int(data_set.num_examples / FLAGS.batch_size)
num_examples = steps_per_epoch * FLAGS.batch_size
for step in np.xrange(steps_per_epoch):
feed_dict = fill_feed_dict(data_set,
images_placeholder,
labels_placeholder)
true_count += sess.run(eval_correct, feed_dict=feed_dict)
precision = float(true_count) / float(num_examples)
print(' Num examples: %d Num correct: %d Precision @ 1: %0.04f' %
(num_examples, true_count, precision))
def download_all_shot_numbers(prepath, save_path, shot_list_files,
signals_full):
max_len = 30000
machine = shot_list_files.machine
signals = []
for sig in signals_full:
if not sig.is_defined_on_machine(machine):
print("Signal {} not defined on machine {}, omitting".format(
sig, machine))
else:
signals.append(sig)
save_prepath = prepath + save_path + "/"
shot_numbers, _ = shot_list_files.get_shot_numbers_and_disruption_times()
# can only use queue of max size 30000
shot_numbers_chunks = [
shot_numbers[i:i + max_len]
for i in np.xrange(0, len(shot_numbers), max_len)
]
start_time = time.time()
for shot_numbers_chunk in shot_numbers_chunks:
download_shot_numbers(shot_numbers_chunk, save_prepath, machine,
signals)
print("Finished downloading {} shots in {} seconds".format(
len(shot_numbers),
time.time() - start_time))
def assignCrowdingDist(self,individuals):
"""Assign a crowding distance to each individual's fitness. The
crowding distance can be retrieve via the :attr:`crowding_dist`
attribute of each individual's fitness.
"""
if len(individuals) == 0:
return
distances = [0.0] * len(individuals)
crowd = [(ind.fitness.values, i) for i, ind in enumerate(individuals)]
nobj = len(individuals[0].fitness.values)
for i in np.xrange(nobj):
crowd.sort(key=lambda element: element[0][i])
distances[crowd[0][1]] = float("inf")
distances[crowd[-1][1]] = float("inf")
if crowd[-1][0][i] == crowd[0][0][i]:
continue
norm = nobj * float(crowd[-1][0][i] - crowd[0][0][i])
for prev, cur, next in zip(crowd[:-2], crowd[1:-1], crowd[2:]):
distances[cur[1]] += (next[0][i] - prev[0][i]) / norm
for i, dist in enumerate(distances):
individuals[i].fitness.crowding_dist = dist
def aperture(startpx, startpy, radius, nRows, nCols):
r = radius
length = 2 * r
height = length
allx = np.xrange(startpx - int(np.ceil(length / 2.0)),
startpx + int(np.floor(length / 2.0)) + 1)
ally = np.xrange(startpy - int(np.ceil(height / 2.0)),
startpy + int(np.floor(height / 2.0)) + 1)
mask = np.zeros((nRows, nCols))
for x in allx:
for y in ally:
if (np.abs(x - startpx))**2 + (np.abs(y - startpy))**2 <= (
r)**2 and 0 <= y and y < nRows and 0 <= x and x < nCols:
mask[y, x] = 1.
return mask
def cal(row,trainLabel):
global w, b
res = 0
for i in np.xrange(len(row)):
res += row[i] * w[i]
res += b
res *= trainLabel
return res
def group_list(l, group_size):
"""
:param l: list or sequence
:param group_size:
:return: batch
"""
for i in np.xrange(0, len(l), group_size):
yield l[i:i + group_size]
def find(value, bin_edges):
"""
helper function for variogram.
"""
for k in np.xrange(len(bin_edges)):
if value<bin_edges[k]:
break
return k-1
def recurring_monthly(start_date, stop_date, base_string):
starting = dt.strptime(start_date, '%m/%d/%y')
dom = "{:02d}".format(int(start_date.split('/')[1]))
ending = dt.strptime(stop_date, '%m/%d/%y')
desc_list = [
base_string + ' ' + dt.strptime('%2.2d-%2.2d' %
(y, m), '%Y-%m').strftime('%b-%y')
for y in xrange(starting.year, ending.year + 1)
for m in xrange(starting.month if y == starting.year else 1,
ending.month + 1 if y == ending.year else 13)
]
time_list = [
dt.strptime('%2.2d-%2.2d' % (y, m), '%Y-%m').strftime('%y/%m/?')
for y in xrange(starting.year, ending.year + 1)
for m in xrange(starting.month if y == starting.year else 1,
ending.month + 1 if y == ending.year else 13)
]
time_list = [x.replace('?', dom) for x in time_list]
return time_list, desc_list
def import_raw_file(self, file):
reader = open(self.file)
row_offset = 2
data = []
with open(file, 'r') as txt_in:
for i in np.xrange(row_offset):
txt_in.next()
for line in txt_in:
data.append(line.split())
data = np.array(data, dtype=np.float64)
return data
def parent_path(depth=1):
"""
Return path to directory which is depth
levels above
"""
path = os.path.abspath(__file__)
n = 0
for j in np.xrange(1, len(path)+1):
if path[-j] == "/":
n += 1
if n == depth:
return path[:len(path)-j]
def next_batch(self, batch_size, fake_data=False):
"""Return the next `batch_size` examples from this data set."""
if fake_data:
fake_image = [1.0 for _ in np.xrange(784)]
fake_label = 0
return [fake_image for _ in np.xrange(batch_size)
], [fake_label for _ in np.xrange(batch_size)]
start = self._index_in_epoch
self._index_in_epoch += batch_size
if self._index_in_epoch > self._num_examples:
# Finished epoch
self._epochs_completed += 1
# Shuffle the data
perm = np.arange(self._num_examples)
np.random.shuffle(perm)
self._images = self._images[perm]
self._labels = self._labels[perm]
# Start next epoch
start = 0
self._index_in_epoch = batch_size
assert batch_size <= self._num_examples
end = self._index_in_epoch
return self._images[start:end], self._labels[start:end]
def perceptionClassify(trainGroup, trainLabels):
global w, b
isFind = False # the flag of find the best w and b
numSamples = trainGroup.shape[0]
mLength = trainGroup.shape[1]
w = [0]* mLength
b = 0
while(not isFind):
for i in np.xrange(numSamples):
if cal(trainGroup[i],trainLabels[i]) <= 0:
print("w: " + w + " b: " + b)
update(trainGroup[i],trainLabels[i])
break #end for loop
elif i == numSamples-1:
print("w: " + w + " b: " + b)
isFind = True #end while loop
def selTournamentDCD(self,individuals, k):
"""Tournament selection based on dominance (D) between two individuals, if
the two individuals do not interdominate the selection is made
based on crowding distance (CD). The *individuals* sequence length has to
be a multiple of 4. Starting from the beginning of the selected
individuals, two consecutive individuals will be different (assuming all
individuals in the input list are unique). Each individual from the input
list won't be selected more than twice.
This selection requires the individuals to have a :attr:`crowding_dist`
attribute, which can be set by the :func:`assignCrowdingDist` function.
:param individuals: A list of individuals to select from.
:param k: The number of individuals to select.
:returns: A list of selected individuals.
"""
if len(individuals) % 4 != 0:
raise ValueError("selTournamentDCD: individuals length must be a multiple of 4")
if k % 4 != 0:
raise ValueError("selTournamentDCD: number of individuals to select must be a multiple of 4")
def tourn(ind1, ind2):
if ind1.fitness.dominates(ind2.fitness):
return ind1
elif ind2.fitness.dominates(ind1.fitness):
return ind2
if ind1.fitness.crowding_dist < ind2.fitness.crowding_dist:
return ind2
elif ind1.fitness.crowding_dist > ind2.fitness.crowding_dist:
return ind1
if random.random() <= 0.5:
return ind1
return ind2
individuals_1 = random.sample(individuals, len(individuals))
individuals_2 = random.sample(individuals, len(individuals))
chosen = []
for i in np.xrange(0, k, 4):
chosen.append(tourn(individuals_1[i], individuals_1[i+1]))
chosen.append(tourn(individuals_1[i+2], individuals_1[i+3]))
chosen.append(tourn(individuals_2[i], individuals_2[i+1]))
chosen.append(tourn(individuals_2[i+2], individuals_2[i+3]))
return chosen
def predict(self, X):
""" X is N x D where each row is an example we wish to predict label for """
num_test = X.shape[0]
# let's make sure that the output type matches the input type
Ypred = np.zeros(num_test, dtype=self.ytr.dtype)
# loop over all test rows
for i in np.xrange(num_test):
# find the nearest training image to the i'th test image
# using the L1 distance (sum of absolute value difference)
distances = np.sum(np.abs(self.Xtr - X[i, :]), axis=1)
min_index = np.argmin(
distances) # get the index with smallest distance
Ypred[i] = self.ytr(
min_index) # predict the label of the nearest example
return Ypred
def compare_channel_replicates(data,
group=True,
title='',
col_groups=None,
cross=False):
"""
Plot (ncols x ncols) grid of scatterplots comparing the measures in each column
data must have 'accession_number' column to aggregate by if group is True
col_groups is a list of name-value pairs to split the plot into several
saved figures, where name is the label of the group and value is a
list of columns
OR one of 'Mar', 'Aug', 'Sep', 'Tyr'
cross [bool] - designates whether to include one panel with cross of all channels
Note that cross will not with with group
"""
# Default column groups for each dataset
if col_groups is None:
col_groups = [('all', np.xrange(data.shape[1]))]
elif isinstance(col_groups, basestring):
if col_groups == 'Aug':
col_groups = [
('Control', ['GFP_A1', 'GFP_A2', 'GFP_B1', 'GFP_B2']),
('KO93',
['KO93_A1', 'KO93_A2', 'KO93_B1', 'KO93_B2', 'KO93_B3']),
('KO95',
['KO95_A1', 'KO93_A2', 'KO95_A3', 'KO95_B1', 'KO95_B2']),
('DKO', ['DKO_A1', 'DKO_A2', 'DKO_B1', 'DKO_B2']),
]
elif col_groups == 'Sep':
col_groups = [
('P25_EE', ['P25_EE_A1', 'P25_EE_A2', 'P25_EE_A3']),
('EE', ['CT_EE_A1', 'CT_EE_A2', 'CT_EE_A3']),
('P25', ['P25_HC_A1', 'P25_HC_A2']),
('Control', ['CT_HC_A1', 'CT_HC_A2']),
]
# Obtain all named columns
all_cols = sum([g[1] for g in col_groups], [])
if group:
count = data.accession_number.groupby(data.accession_number).count()
aggregated = data[all_cols].groupby(data.accession_number).mean()
aggregated['n_pep'] = count
else:
aggregated = data
for name, cols in col_groups:
f = compare_measures(aggregated,
cols,
title=title,
corr=True,
count=False)
f.set_size_inches(10, 10)
f.tight_layout(rect=(0, 0, 1, 0.95))
f.savefig('figures/%s_channel_reps_%s.png' % (title, name), dpi=100)
# TODO separate aggregation code so cross_groups works with agg
if cross:
f = compare_measures(aggregated,
all_cols,
title="ALL",
corr=True,
count=False)
f.set_size_inches(2 * len(all_cols), 2 * len(all_cols))
f.tight_layout(rect=(0, 0, 1, 0.95))
f.savefig('figures/%s_channel_reps_ALL.png' % (title, ), dpi=100)
'''
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
# 构建数据：100个随机点
points_num = 100
# 之后要往vectors中填充100个点的值
vectors = []
# 用 Numpy 的正态随机分布函数生成 100 个点
# 这些点的(x, y)坐标值: 对应线性方程 y = 0.1 * x + 0.2
# 权重 (Weight) 为 0.1，偏差 (Bias)为 0.2
try:
# 运行100次
for i in np.xrange(points_num):
# 横坐标值，随机正态分布函数。区间0-0.66
x1 = np.random.normal(0.0, 0.66)
# 在真实值上加一些偏差
y1 = 0.1 * x1 + 0.2 + np.random.normal(0.0, 0.04)
# 将点list加入vectors列表中
vectors.append([x1, y1])
except:
for i in range(points_num):
x1 = np.random.normal(0.0, 0.66)
y1 = 0.1 * x1 + 0.2 + np.random.normal(0.0, 0.04)
vectors.append([x1, y1])
x_data = [v[0] for v in vectors] # 列表生成式取出真实的点的 x 坐标
y_data = [v[1] for v in vectors] # 真实的点的 y 坐标
def run_training():
"""Train MNIST for a number of steps."""
# Get the sets of images and labels for training, validation, and
# test on MNIST.
data_sets = input_data.read_data_sets(FLAGS.train_dir, FLAGS.fake_data)
# Tell TensorFlow that the model will be built into the default Graph.
with tf.Graph().as_default():
# Generate placeholders for the images and labels.
images_placeholder, labels_placeholder = placeholder_inputs(
FLAGS.batch_size)
# Build a Graph that computes predictions from the inference model.
logits = mnist.inference(images_placeholder,
FLAGS.hidden1,
FLAGS.hidden2)
# Add to the Graph the Ops for loss calculation.
loss = mnist.loss(logits, labels_placeholder)
# Add to the Graph the Ops that calculate and apply gradients.
train_op = mnist.training(loss, FLAGS.learning_rate)
# Add the Op to compare the logits to the labels during evaluation.
eval_correct = mnist.evaluation(logits, labels_placeholder)
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.merge_all_summaries()
# Create a saver for writing training checkpoints.
saver = tf.train.Saver()
# Create a session for running Ops on the Graph.
sess = tf.Session()
# Run the Op to initialize the variables.
init = tf.initialize_all_variables()
sess.run(init)
# Instantiate a SummaryWriter to output summaries and the Graph.
summary_writer = tf.train.SummaryWriter(FLAGS.train_dir,
graph_def=sess.graph_def)
# And then after everything is built, start the training loop.
for step in np.xrange(FLAGS.max_steps):
start_time = time.time()
# Fill a feed dictionary with the actual set of images and labels
# for this particular training step.
feed_dict = fill_feed_dict(data_sets.train,
images_placeholder,
labels_placeholder)
# Run one step of the model. The return values are the activations
# from the `train_op` (which is discarded) and the `loss` Op. To
# inspect the values of your Ops or variables, you may include them
# in the list passed to sess.run() and the value tensors will be
# returned in the tuple from the call.
_, loss_value = sess.run([train_op, loss],
feed_dict=feed_dict)
duration = time.time() - start_time
# Write the summaries and print an overview fairly often.
if step % 100 == 0:
# Print status to stdout.
print('Step %d: loss = %.2f (%.3f sec)' % (step, loss_value, duration))
# Update the events file.
summary_str = sess.run(summary_op, feed_dict=feed_dict)
summary_writer.add_summary(summary_str, step)
# Save a checkpoint and evaluate the model periodically.
if (step + 1) % 1000 == 0 or (step + 1) == FLAGS.max_steps:
saver.save(sess, FLAGS.train_dir, global_step=step)
# Evaluate against the training set.
print('Training Data Eval:')
do_eval(sess,
eval_correct,
images_placeholder,
labels_placeholder,
data_sets.train)
# Evaluate against the validation set.
print('Validation Data Eval:')
do_eval(sess,
eval_correct,
images_placeholder,
labels_placeholder,
data_sets.validation)
# Evaluate against the test set.
print('Test Data Eval:')
do_eval(sess,
eval_correct,
images_placeholder,
labels_placeholder,
data_sets.test)
def selSPEA2(individuals, k):
"""Apply SPEA-II selection operator on the *individuals*. Usually, the
size of *individuals* will be larger than *n* because any individual
present in *individuals* will appear in the returned list at most once.
Having the size of *individuals* equals to *n* will have no effect other
than sorting the population according to a strength Pareto scheme. The
list returned contains references to the input *individuals*. For more
details on the SPEA-II operator see [Zitzler2001]_.
:param individuals: A list of individuals to select from.
:param k: The number of individuals to select.
:returns: A list of selected individuals.
.. [Zitzler2001] Zitzler, Laumanns and Thiele, "SPEA 2: Improving the
strength Pareto evolutionary algorithm", 2001.
"""
N = len(individuals)
L = len(individuals[0].fitness.values)
K = math.sqrt(N)
strength_fits = [0] * N
fits = [0] * N
dominating_inds = [list() for i in np.xrange(N)]
for i, ind_i in enumerate(individuals):
for j, ind_j in enumerate(individuals[i+1:], i+1):
if ind_i.fitness.dominates(ind_j.fitness):
strength_fits[i] += 1
dominating_inds[j].append(i)
elif ind_j.fitness.dominates(ind_i.fitness):
strength_fits[j] += 1
dominating_inds[i].append(j)
for i in np.xrange(N):
for j in dominating_inds[i]:
fits[i] += strength_fits[j]
# Choose all non-dominated individuals
chosen_indices = [i for i in np.xrange(N) if fits[i] < 1]
if len(chosen_indices) < k: # The archive is too small
for i in np.xrange(N):
distances = [0.0] * N
for j in np.xrange(i + 1, N):
dist = 0.0
for l in np.xrange(L):
val = individuals[i].fitness.values[l] - \
individuals[j].fitness.values[l]
dist += val * val
distances[j] = dist
kth_dist = _randomizedSelect(distances, 0, N - 1, K)
density = 1.0 / (kth_dist + 2.0)
fits[i] += density
next_indices = [(fits[i], i) for i in np.xrange(N)
if not i in chosen_indices]
next_indices.sort()
#print next_indices
chosen_indices += [i for _, i in next_indices[:k - len(chosen_indices)]]
elif len(chosen_indices) > k: # The archive is too large
N = len(chosen_indices)
distances = [[0.0] * N for i in np.xrange(N)]
sorted_indices = [[0] * N for i in np.xrange(N)]
for i in np.xrange(N):
for j in np.xrange(i + 1, N):
dist = 0.0
for l in np.xrange(L):
val = individuals[chosen_indices[i]].fitness.values[l] - \
individuals[chosen_indices[j]].fitness.values[l]
dist += val * val
distances[i][j] = dist
distances[j][i] = dist
distances[i][i] = -1
# Insert sort is faster than quick sort for short arrays
for i in np.xrange(N):
for j in np.xrange(1, N):
l = j
while l > 0 and distances[i][j] < distances[i][sorted_indices[i][l - 1]]:
sorted_indices[i][l] = sorted_indices[i][l - 1]
l -= 1
sorted_indices[i][l] = j
size = N
to_remove = []
while size > k:
# Search for minimal distance
min_pos = 0
for i in np.xrange(1, N):
for j in np.xrange(1, size):
dist_i_sorted_j = distances[i][sorted_indices[i][j]]
dist_min_sorted_j = distances[min_pos][sorted_indices[min_pos][j]]
if dist_i_sorted_j < dist_min_sorted_j:
min_pos = i
break
elif dist_i_sorted_j > dist_min_sorted_j:
break
# Remove minimal distance from sorted_indices
for i in np.xrange(N):
distances[i][min_pos] = float("inf")
distances[min_pos][i] = float("inf")
for j in np.xrange(1, size - 1):
if sorted_indices[i][j] == min_pos:
sorted_indices[i][j] = sorted_indices[i][j + 1]
sorted_indices[i][j + 1] = min_pos
# Remove corresponding individual from chosen_indices
to_remove.append(min_pos)
size -= 1
for index in reversed(sorted(to_remove)):
del chosen_indices[index]
return [individuals[i] for i in chosen_indices]
def kernel(self, cineObj, st , mu, LMBD, gamma, nInner, nBreg):
self.st['sensemap']=self.st['sensemap']*self.st['mask']
orig_num_ky=numpy.shape(cineObj.tse)[1]
tse = cineObj.tse[:,orig_num_ky/2 - self.st['Nd'][0]/2 : orig_num_ky/2 + self.st['Nd'][0]/2,:]
# tse=cineObj.tse
# tse=numpy.abs(numpy.mean(self.st['sensemap'],-1))
tse=CsTransform.pynufft.appendmat(tse,self.st['Nd'][1])
#tse=Normalize(tse)
tse=numpy.transpose(tse,(0,1,3,2))
self.ttse=tse#CsTransform.pynufft.Normalize(tse)
self.tse0 = CsTransform.pynufft.CombineMulti(tse, -1)
print('line392, shape self.tse0',numpy.shape(self.tse0))
self.filter= numpy.ones(tse.shape)
dpss = numpy.kaiser(tse.shape[1], 1.0)*10.0
for ppp in range(0,tse.shape[1]):
self.filter[:,ppp,:,:]=self.filter[:,ppp,:,:]*dpss[ppp]
print('tse.shape',tse.shape)
# L= numpy.size(f)/st['M']
# image_dim=st['Nd']+(L,)
#
# if numpy.ndim(f) == 1:# preventing row vector
# f=numpy.reshape(f,(numpy.shape(f)[0],1),order='F')
# f0 = numpy.copy(f) # deep copy to prevent scope f0 to f
## u = numpy.zeros(image_dim,dtype=numpy.complex64)
f0=numpy.copy(cineObj.f)
f=numpy.copy(cineObj.f)
# u0=self.data2rho(f_internal,
# cineObj.dim_x,
# self.st['Nd'][0],
# self.st['Nd'][1],
# cineObj.ncoils,
# self.CsTransform
# ) # doing spatial transform
u0 = self.fun1(cineObj)
pdf = cineObj.pdf
pdf = CsTransform.pynufft.appendmat(pdf,self.st['Nd'][1])
pdf = numpy.transpose(pdf,(0,1,3,2))
# u0 = fftpack.fftn(u0,axes=(1,))
# u0 = fftpack.fftshift(u0,axes=(1,))
# #u0[:,:,u0.shape[2]/2,:] = u0[:,:,u0.shape[2]/2,:]/pdf[:,:,u0.shape[2]/2,:]
# u0 = u0#/pdf
# u0 = fftpack.ifftshift(u0,axes=(1,))
# u0 = fftpack.ifftn(u0,axes=(1,))
# print('cineObj.pdf.shape',cineObj.pdf.shape)
# for pj in range(0,4):
# matplotlib.pyplot.imshow(cineObj.pdf[:,:,pj].real)
# matplotlib.pyplot.show()
u0=self.fun2(u0)
u0=self.fun3(u0)
u0 = u0*self.st['sensemap'].conj()
u0 = CsTransform.pynufft.CombineMulti(u0,-1)
print('line443, shape u0',numpy.shape(u0))
#u0 = u0*self.filter
uker = self.create_laplacian_kernel(cineObj)
uker = CsTransform.pynufft.appendmat(uker,u0.shape[3])
self.u0 = u0
u = numpy.copy(self.tse0)
print('u0.shape',u0.shape)
(xx,bb,dd)=self.make_split_variables(u)
uf = numpy.copy(u) # only used for ISRA, written here for generality
murf = numpy.copy(u) # initial values
# #===============================================================================
#u_stack = numpy.empty(st['Nd']+(nBreg,),dtype=numpy.complex)
for outer in xrange(0,nBreg):
for inner in xrange(0,nInner):
# update u
print('iterating',[inner,outer])
#===============================================================
# update u # simple k-space deconvolution to guess initial u
u = self.update_u(murf, u, uker, xx, bb,cineObj)
c = numpy.max(numpy.abs(u[:])) # Rough coefficient
# to correct threshold of nonlinear shrink
#===================================================================
# # update d
#===================================================================
#===================================================================
# Shrinkage: remove tiny values "in somewhere sparse!"
# dx+bx should be sparse!
#===================================================================
# shrinkage
#===================================================================
dd=self.update_d(u,dd)
xx=self.shrink( dd, bb, c*1.0/LMBD/numpy.sqrt(numpy.prod(st['Nd'])))
#===============================================================
#===================================================================
# # update b
#===================================================================
bb=self._update_b(bb, dd, xx)
if outer < (nBreg-1): # do not update in the last loop
(f, uf, murf,u)=self.external_update(u, f, uf, f0, u0) # update outer Split_bregman
# u = CsTransform.pynufft.Normalize(u)
# for pp in range(0,u0.shape[2]):
# matplotlib.pyplot.subplot(numpy.sqrt(u0.shape[2])+1,numpy.sqrt(u0.shape[2])+1,pp)
# matplotlib.pyplot.imshow(numpy.sum(numpy.abs(u[...,pp,:]),-1),norm=norm,interpolation='nearest')
# matplotlib.pyplot.show()
#
return (u,uf)
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Mon Apr 23 18:08:05 2018
@author: osboxes
"""
import numpy as np
import cv2
img1 = cv2.imread('images/input/messi5.jpg')
e1 = cv2.getTickCount()
for i in np.xrange(5,49,2):
img1 = cv2.medianBlur(img1,i)
e2 = cv2.getTickCount()
t = (e2 - e1)/cv2.getTickFrequency()
print(t)
# Result I got is 0.521107655 seconds
def vibrate(self, length):
if length in np.xrange(1, 4):
# first byte tells it to vibrate; purpose of second byte is unknown
self.write_attr(0x19, pack('3B', 3, 1, length))
def ellipse_overlap_fast(f1, f2, options):
opts = {}
opts['normaliseFrames'] = True
opts['normalisedScale'] = 30
opts['minAreaRatio'] = 0.3
opts['frame2frame'] = False
opts['fix'] = False
parse_arg(opts, options)
N2 = f2.shape[0]
ellipsePairs = []
scores = []
if f1.shape[0] == 0 or f2.shape[0] == 0:
return ellipsePairs, scores
f1 = frame2ellipse(f1)
f2 = frame2ellipse(f2)
e1, eigVec1 = ellipse_eigen(f1)
e2, eigVec2 = ellipse_eigen(f2)
vggEll1 = ellipse2vggformat(f1, e1, eigVec1)
vggEll2 = ellipse2vggformat(f2, e2, eigVec2)
a1 = np.pi * np.sqrt(np.prod(e1, axis=1))
a2 = np.pi * np.sqrt(np.prod(e2, axis=1))
for i2 in range(N2):
if opts['normaliseFrames']:
s = opts['normalisedScale'] / np.sqrt(a2[i2] / np.pi)
else:
s = 1
if opts['frame2frame']:
ellipsePairs[i2] = np.hstack[i2 *
np.ones((f1.shape[0], 1), dtype=np.int), np.xrange(f1.shape[0])]
else:
thr = 4 * np.sqrt(a2[i2] / np.pi)
if opts['fix']:
thr = thr * s
canOverlap = scipy.spatial.distance.cdist(
f2[[i2], 0:2], f1[:, 0:2], 'euclidean') < thr
maxOverlap = np.minimum(a2[i2], a1) / \
np.maximum(a2[i2], a1) * canOverlap
_, pairs = np.where(maxOverlap > opts['minAreaRatio'])
ellipsePairs.extend(zip([i2] * pairs.shape[0], pairs.tolist()))
if len(pairs) == 0:
continue
if opts['normaliseFrames']:
vggS = np.array([1, 1, 1 / s**2, 1 / s**2, 1 / s**2, s, s, s, s])
lhsEllipse = vggS * vggEll2[[i2]]
rhsEllipse = vggEll1[pairs] * vggS
else:
lhsEllipse = vggEll2[[i2]]
rhsEllipse = vggEll1[pairs]
_, tw, _, _ = bench.vgg_compute_ellipse_overlap.vgg_compute_ellipse_overlap(
lhsEllipse, rhsEllipse, -1)
scores.extend((1 - tw / 100).tolist()[0])
return np.array(ellipsePairs), np.array(scores)
def update(row,trainLabel):
global w, b
for i in np.xrange(len(row)):
w[i] += trainLabel * row[i]
b += trainLabel
def chain(cosmo, data, command_line):
"""
Run a Markov chain of fixed length with a Metropolis Hastings algorithm.
Main function of this module, this is the actual Markov chain procedure.
After having selected a starting point in parameter space defining the
first **last accepted** one, it will, for a given amount of steps :
+ choose randomly a new point following the *proposal density*,
+ compute the cosmological *observables* through the cosmological module,
+ compute the value of the *likelihoods* of the desired experiments at this
point,
+ *accept/reject* this point given its likelihood compared to the one of
the last accepted one.
Every time the code accepts :code:`data.write_step` number of points
(quantity defined in the input parameter file), it will write the result to
disk (flushing the buffer by forcing to exit the output file, and reopen it
again.
.. note::
to use the code to set a fiducial file for certain fixed parameters,
you can use two solutions. The first one is to put all input 1-sigma
proposal density to zero (this method still works, but is not
recommended anymore). The second one consist in using the flag "-f 0",
to force a step of zero amplitude.
"""
## Initialisation
loglike = 0
# In case command_line.silent has been asked, outputs should only contain
# data.out. Otherwise, it will also contain sys.stdout
outputs = [data.out]
if not command_line.silent:
outputs.append(sys.stdout)
use_mpi = False
# check for MPI
try:
from mpi4py import MPI
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
# suppress duplicate output from slaves
if rank:
command_line.quiet = True
use_mpi = True
except ImportError:
# set all chains to master if no MPI
rank = 0
# Initialise master and slave chains for superupdate.
# Workaround in order to have one master chain and several slave chains even when
# communication fails between MPI chains. It could malfunction on some hardware.
# TODO: Would like to merge with MPI initialization above and make robust and logical
# TODO: Or if keeping current scheme, store value and delete jumping_factor.txt
# TODO: automatically if --parallel-chains is enabled
if command_line.superupdate and data.jumping_factor:
try:
jump_file = open(command_line.folder + '/jumping_factor.txt','r')
#if command_line.restart is None:
if not use_mpi and command_line.parallel_chains:
rank = 1
warnings.warn('MPI not in use, flag --parallel-chains enabled, '
'superupdate enabled, and a jumping_factor.txt file detected. '
'If relaunching in the same folder or restarting a run this '
'will cause all chains to be assigned as slaves. In this case '
'instead note the value in jumping_factor.txt, delete the '
'file, and pass the value with flag -f <value>. This warning '
'may then appear again, but you can safely disregard it.')
else:
# For restart runs we want to save the input jumping factor
# as starting jumping factor, but continue from the jumping
# factor stored in the file.
starting_jumping_factor = data.jumping_factor
# This will load the value irrespective of whether it starts
# with # (i.e. the jumping factor adaptation was started) or not.
jump_value = jump_file.read().replace('# ','')
data.jumping_factor = float(jump_value)
jump_file.close()
print('rank = ',rank)
except:
jump_file = open(command_line.folder + '/jumping_factor.txt','w')
jump_file.write(str(data.jumping_factor))
jump_file.close()
rank = 0
print('rank = ',rank)
starting_jumping_factor = data.jumping_factor
# Recover the covariance matrix according to the input, if the varying set
# of parameters is non-zero
if (data.get_mcmc_parameters(['varying']) != []):
# Read input covariance matrix
sigma_eig, U, C = sampler.get_covariance_matrix(cosmo, data, command_line)
# if we want to compute the starting point by minimising lnL (instead of taking it from input file or bestfit file)
minimum = 0
if command_line.minimize:
minimum = sampler.get_minimum(cosmo, data, command_line, C)
parameter_names = data.get_mcmc_parameters(['last_accepted'])
for index,elem in parameter_names:
data.mcmc_parameters[elem]['last_accepted'] = minimum[index]
# if we want to compute Fisher matrix and then stop
if command_line.fisher:
sampler.get_fisher_matrix(cosmo, data, command_line, C, minimum)
return
# warning if no jumps are requested
if data.jumping_factor == 0:
warnings.warn(
"The jumping factor has been set to 0. The above covariance " +
"matrix will not be used.")
# In case of a fiducial run (all parameters fixed), simply run once and
# print out the likelihood. This should not be used any more (one has to
# modify the log.param, which is never a good idea. Instead, force the code
# to use a jumping factor of 0 with the option "-f 0".
else:
warnings.warn(
"You are running with no varying parameters... I will compute " +
"only one point and exit")
data.update_cosmo_arguments() # this fills in the fixed parameters
loglike = sampler.compute_lkl(cosmo, data)
io_mp.print_vector(outputs, 1, loglike, data)
return 1, loglike
# In the fast-slow method, one need the Cholesky decomposition of the
# covariance matrix. Return the Cholesky decomposition as a lower
# triangular matrix
Cholesky = None
Rotation = None
if command_line.jumping == 'fast':
Cholesky = la.cholesky(C).T
Rotation = np.identity(len(sigma_eig))
# define path and covmat
input_covmat = command_line.cov
base = os.path.basename(command_line.folder)
# the previous line fails when "folder" is a string ending with a slash. This issue is cured by the next lines:
if base == '':
base = os.path.basename(command_line.folder[:-1])
command_line.cov = os.path.join(
command_line.folder, base+'.covmat')
# Fast Parameter Multiplier (fpm) for adjusting update and superupdate numbers.
# This is equal to N_slow + f_fast N_fast, where N_slow is the number of slow
# parameters, f_fast is the over sampling number for each fast block and f_fast
# is the number of parameters in each fast block.
for i in range(len(data.block_parameters)):
if i == 0:
fpm = data.over_sampling[i]*data.block_parameters[i]
else:
fpm += data.over_sampling[i]*(data.block_parameters[i] - data.block_parameters[i-1])
# If the update mode was selected, the previous (or original) matrix should be stored
if command_line.update:
if not rank and not command_line.silent:
print('Update routine is enabled with value %d (recommended: 50)' % command_line.update)
print('This number is rescaled by cycle length %d (N_slow + f_fast * N_fast) to %d' % (fpm,fpm*command_line.update))
# Rescale update number by cycle length N_slow + f_fast * N_fast to account for fast parameters
command_line.update *= fpm
previous = (sigma_eig, U, C, Cholesky)
# Initialise adaptive
if command_line.adaptive:
if not command_line.silent:
print('Adaptive routine is enabled with value %d (recommended: 10*dimension)' % command_line.adaptive)
print('and adaptive_ts = %d (recommended: 100*dimension)' % command_line.adaptive_ts)
print('Please note: current implementation not suitable for multiple chains')
if rank > 0:
raise io_mp.ConfigurationError('Adaptive routine not compatible with MPI')
if command_line.update:
warnings.warn('Adaptive routine not compatible with update, overwriting input update value')
if command_line.superupdate:
warnings.warn('Adaptive routine not compatible with superupdate, deactivating superupdate')
command_line.superupdate = 0
# Define needed parameters
parameter_names = data.get_mcmc_parameters(['varying'])
mean = np.zeros(len(parameter_names))
last_accepted = np.zeros(len(parameter_names),'float64')
ar = np.zeros(100)
if command_line.cov == None:
# If no input covmat was given, the starting jumping factor
# should be very small until a covmat is obtained and the
# original start jumping factor should be saved
start_jumping_factor = command_line.jumping_factor
data.jumping_factor = command_line.jumping_factor/100.
# Analyze module will be forced to compute one covmat,
# after which update flag will be set to False.
command_line.update = command_line.adaptive
else:
# If an input covmat was provided, take mean values from param file
# Question: is it better to always do this, rather than setting mean
# to last accepted after the initial update run?
for elem in parameter_names:
mean[parameter_names.index(elem)] = data.mcmc_parameters[elem]['initial'][0]
# Initialize superupdate
if command_line.superupdate:
if not rank and not command_line.silent:
print('Superupdate routine is enabled with value %d (recommended: 20)' % command_line.superupdate)
if command_line.superupdate < 20:
warnings.warn('Superupdate value lower than the recommended value. This '
'may increase the risk of poorly converged acceptance rate')
print('This number is rescaled by cycle length %d (N_slow + f_fast * N_fast) to %d' % (fpm,fpm*command_line.superupdate))
# Rescale superupdate number by cycle length N_slow + f_fast * N_fast to account for fast parameters
command_line.superupdate *= fpm
# Define needed parameters
parameter_names = data.get_mcmc_parameters(['varying'])
updated_steps = 0
stop_c = False
jumping_factor_rescale = 0
if command_line.restart:
try:
jump_file = open(command_line.cov,'r')
jumping_factor_rescale = 1
except:
jumping_factor_rescale = 0
c_array = np.zeros(command_line.superupdate) # Allows computation of mean of jumping factor
R_minus_one = np.array([100.,100.]) # 100 to make sure max(R-1) value is high if computation failed
# Local acceptance rate of last SU*(N_slow + f_fast * N_fast) steps
ar = np.zeros(command_line.superupdate)
# Store acceptance rate of last 5*SU*(N_slow + f_fast * N_fast) steps
backup_ar = np.zeros(5*command_line.superupdate)
# Make sure update is enabled
if command_line.update == 0:
if not rank and not command_line.silent:
print('Update routine required by superupdate. Setting --update 50')
print('This number is then rescaled by cycle length: %d (N_slow + f_fast * N_fast)' % fpm)
command_line.update = 50 * fpm
previous = (sigma_eig, U, C, Cholesky)
# If restart wanted, pick initial value for arguments
if command_line.restart is not None:
sampler.read_args_from_chain(data, command_line.restart)
# If restart from best fit file, read first point (overwrite settings of
# read_args_from_chain)
if command_line.bf is not None and not command_line.minimize:
sampler.read_args_from_bestfit(data, command_line.bf)
# Pick a position (from last accepted point if restart, from the mean value
# else), with a 100 tries.
for i in range(100):
if get_new_position(data, sigma_eig, U, i,
Cholesky, Rotation) is True:
break
if i == 99:
raise io_mp.ConfigurationError(
"You should probably check your prior boundaries... because " +
"no valid starting position was found after 100 tries")
# Compute the starting Likelihood
loglike = sampler.compute_lkl(cosmo, data)
# Choose this step as the last accepted value
# (accept_step), and modify accordingly the max_loglike
sampler.accept_step(data)
max_loglike = loglike
# If the jumping factor is 0, the likelihood associated with this point is
# displayed, and the code exits.
if data.jumping_factor == 0:
io_mp.print_vector(outputs, 1, loglike, data)
return 1, loglike
acc, rej = 0.0, 0.0 # acceptance and rejection number count
N = 1 # number of time the system stayed in the current position
# Print on screen the computed parameters
if not command_line.silent and not command_line.quiet:
io_mp.print_parameters(sys.stdout, data)
# Suppress non-informative output after initializing
command_line.quiet = True
k = 1
# Main loop, that goes on while the maximum number of failure is not
# reached, and while the expected amount of steps (N) is not taken.
while k <= command_line.N:
# If the number of steps reaches the number set in the adaptive method plus one,
# then the proposal distribution should be gradually adapted.
# If the number of steps also exceeds the number set in adaptive_ts,
# the jumping factor should be gradually adapted.
if command_line.adaptive and k>command_line.adaptive+1:
# Start of adaptive routine
# By B. Schroer and T. Brinckmann
# Modified version of the method outlined in the PhD thesis of Marta Spinelli
# Store last accepted step
for elem in parameter_names:
last_accepted[parameter_names.index(elem)] = data.mcmc_parameters[elem]['last_accepted']
# Recursion formula for mean and covmat (and jumping factor after ts steps)
# mean(k) = mean(k-1) + (last_accepted - mean(k-1))/k
mean += 1./k*(last_accepted-mean)
# C(k) = C(k-1) + [(last_accepted - mean(k))^T * (last_accepted - mean(k)) - C(k-1)]/k
C +=1./k*(np.dot(np.transpose(np.asmatrix(last_accepted-mean)),np.asmatrix(last_accepted-mean))-C)
sigma_eig, U = np.linalg.eig(np.linalg.inv(C))
if command_line.jumping == 'fast':
Cholesky = la.cholesky(C).T
if k>command_line.adaptive_ts:
# c = j^2/d
c = data.jumping_factor**2/len(parameter_names)
# c(k) = c(k-1) + [acceptance_rate(last 100 steps) - 0.25]/k
c +=(np.mean(ar)-0.25)/k
data.jumping_factor = np.sqrt(len(parameter_names)*c)
# Save the covariance matrix and the jumping factor in a file
# For a possible MPI implementation
#if not (k-command_line.adaptive) % 5:
# io_mp.write_covariance_matrix(C,parameter_names,str(command_line.cov))
# jump_file = open(command_line.folder + '/jumping_factor.txt','w')
# jump_file.write(str(data.jumping_factor))
# jump_file.close()
# End of adaptive routine
# If the number of steps reaches the number set in the update method,
# then the proposal distribution should be adapted.
if command_line.update:
# Start of update routine
# By M. Ballardini and T. Brinckmann
# Also used by superupdate and adaptive
# master chain behavior
if not rank:
# Add the folder to the list of files to analyze, and switch on the
# options for computing only the covmat
from parser_mp import parse
info_command_line = parse(
'info %s --minimal --noplot --keep-fraction 0.5 --keep-non-markovian --want-covmat' % command_line.folder)
info_command_line.update = command_line.update
if command_line.adaptive:
# Keep all points for covmat guess in adaptive
info_command_line = parse('info %s --minimal --noplot --keep-non-markovian --want-covmat' % command_line.folder)
# Tell the analysis to update the covmat after t0 steps if it is adaptive
info_command_line.adaptive = command_line.adaptive
# Only compute covmat if no input covmat was provided
if input_covmat != None:
info_command_line.want_covmat = False
# This is in order to allow for more frequent R-1 computation with superupdate
compute_R_minus_one = False
if command_line.superupdate:
if not (k+10) % command_line.superupdate:
compute_R_minus_one = True
# the +10 below is here to ensure that the first master update will take place before the first slave updates,
# but this is a detail, the code is robust against situations where updating is not possible, so +10 could be omitted
if (not (k+10) % command_line.update or compute_R_minus_one) and k > 10:
# Try to launch an analyze (computing a new covmat if successful)
try:
if not (k+10) % command_line.update:
from analyze import analyze
R_minus_one = analyze(info_command_line)
elif command_line.superupdate:
# Compute (only, i.e. no covmat) R-1 more often when using superupdate
info_command_line = parse(
'info %s --minimal --noplot --keep-fraction 0.5 --keep-non-markovian' % command_line.folder)
info_command_line.update = command_line.update
R_minus_one = analyze(info_command_line)
except:
if not command_line.silent:
print('Step ',k,' chain ', rank,': Failed to calculate covariance matrix')
if command_line.superupdate:
# Start of superupdate routine
# By B. Schroer and T. Brinckmann
c_array[(k-1)%(command_line.superupdate)] = data.jumping_factor
# If acceptance rate deviates too much from the target acceptance
# rate we want to resume adapting the jumping factor
# T. Brinckmann 02/2019: use mean a.r. over the last 5*len(ar) steps
# instead or the over last len(ar), which is more stable
if abs(np.mean(backup_ar) - command_line.superupdate_ar) > 5.*command_line.superupdate_ar_tol:
stop_c = False
# Start adapting the jumping factor after command_line.superupdate steps if R-1 < 10
# The lower R-1 criterium is an arbitrary choice to keep from updating when the R-1
# calculation fails (i.e. returns only zeros).
if (k > updated_steps + command_line.superupdate) and 0.01 < (max(R_minus_one) < 10.) and not stop_c:
c = data.jumping_factor**2/len(parameter_names)
# To avoid getting trapped in local minima, the jumping factor should
# not go below 0.1 (arbitrary) times the starting jumping factor.
if (c + (np.mean(ar) - command_line.superupdate_ar)/(k - updated_steps)) > (0.1*starting_jumping_factor)**2./len(parameter_names) or ((np.mean(ar) - command_line.superupdate_ar)/(k - updated_steps) > 0):
c += (np.mean(ar) - command_line.superupdate_ar)/(k - updated_steps)
data.jumping_factor = np.sqrt(len(parameter_names) * c)
if not (k-1) % 5:
# Check if the jumping factor adaptation should stop.
# An acceptance rate of 25% balances the wish for more accepted
# points, while ensuring the parameter space is properly sampled.
# The convergence criterium is by default (26+/-1)%, so the adaptation
# will stop when the code reaches an acceptance rate of at least 25%.
# T. Brinckmann 02/2019: use mean a.r. over the last 5*len(ar) steps
# instead or the over last len(ar), which is more stable
if (max(R_minus_one) < 0.4) and (abs(np.mean(backup_ar) - command_line.superupdate_ar) < command_line.superupdate_ar_tol) and (abs(np.mean(c_array)/c_array[(k-1) % (command_line.superupdate)] - 1) < 0.01):
stop_c = True
data.out.write('# After %d accepted steps: stop adapting the jumping factor at a value of %f with a local acceptance rate %f \n' % (int(acc),data.jumping_factor,np.mean(backup_ar)))
if not command_line.silent:
print('After %d accepted steps: stop adapting the jumping factor at a value of %f with a local acceptance rate of %f \n' % (int(acc), data.jumping_factor,np.mean(backup_ar)))
jump_file = open(command_line.folder + '/jumping_factor.txt','w')
jump_file.write('# '+str(data.jumping_factor))
jump_file.close()
else:
jump_file = open(command_line.folder + '/jumping_factor.txt','w')
jump_file.write(str(data.jumping_factor))
jump_file.close()
# Write the evolution of the jumping factor to a file
if not k % (command_line.superupdate):
jump_file = open(command_line.folder + '/jumping_factors.txt','a')
for i in np.xrange(command_line.superupdate):
jump_file.write(str(c_array[i])+'\n')
jump_file.close()
# End of main part of superupdate routine
if not (k-1) % (command_line.update/3):
try:
# Read the covmat
sigma_eig, U, C = sampler.get_covariance_matrix(
cosmo, data, command_line)
if command_line.jumping == 'fast':
Cholesky = la.cholesky(C).T
# Test here whether the covariance matrix has really changed
# We should in principle test all terms, but testing the first one should suffice
if not C[0,0] == previous[2][0,0]:
if k == 1:
if not command_line.silent:
if not input_covmat == None:
warnings.warn(
'Appending to an existing folder: using %s instead of %s. '
'If new input covmat is desired, please delete previous covmat.'
% (command_line.cov, input_covmat))
else:
warnings.warn(
'Appending to an existing folder: using %s. '
'If no starting covmat is desired, please delete previous covmat.'
% command_line.cov)
else:
# Start of second part of superupdate routine
if command_line.superupdate:
# Adaptation of jumping factor should start again after the covmat is updated
# Save the step number after it updated for superupdate and start adaption of c again
updated_steps = k
stop_c = False
cov_det = np.linalg.det(C)
prev_cov_det = np.linalg.det(previous[2])
# Rescale jumping factor in order to keep the magnitude of the jumps the same.
# Skip this update the first time the covmat is updated in order to prevent
# problems due to a poor initial covmat. Rescale the jumping factor after the
# first calculated covmat to the expected optimal one of 2.4.
if jumping_factor_rescale:
new_jumping_factor = data.jumping_factor * (prev_cov_det/cov_det)**(1./(2 * len(parameter_names)))
data.out.write('# After %d accepted steps: rescaled jumping factor from %f to %f, due to updated covariance matrix \n' % (int(acc), data.jumping_factor, new_jumping_factor))
if not command_line.silent:
print('After %d accepted steps: rescaled jumping factor from %f to %f, due to updated covariance matrix \n' % (int(acc), data.jumping_factor, new_jumping_factor))
data.jumping_factor = new_jumping_factor
else:
data.jumping_factor = starting_jumping_factor
jumping_factor_rescale += 1
# End of second part of superupdate routine
# Write to chains file when the covmat was updated
data.out.write('# After %d accepted steps: update proposal with max(R-1) = %f and jumping factor = %f \n' % (int(acc), max(R_minus_one), data.jumping_factor))
if not command_line.silent:
print('After %d accepted steps: update proposal with max(R-1) = %f and jumping factor = %f \n' % (int(acc), max(R_minus_one), data.jumping_factor))
try:
if stop_after_update:
k = command_line.N
print('Covariance matrix updated - stopping run')
except:
pass
previous = (sigma_eig, U, C, Cholesky)
except:
pass
command_line.quiet = True
# Start of second part of adaptive routine
# Stop updating the covmat after t0 steps in adaptive
if command_line.adaptive and k > 1:
command_line.update = 0
data.jumping_factor = start_jumping_factor
# Test if there are still enough steps left before the adaption of the jumping factor starts
if k > 0.5*command_line.adaptive_ts:
command_line.adaptive_ts += k
# Set the mean for the recursion formula to the last accepted point
for elem in parameter_names:
mean[parameter_names.index(elem)] = data.mcmc_parameters[elem]['last_accepted']
# End of second part of adaptive routine
# slave chain behavior
else:
# Start of slave superupdate routine
if command_line.superupdate:
# If acceptance rate deviates too much from the target acceptance
# rate we want to resume adapting the jumping factor. This line
# will force the slave chains to check if the jumping factor
# has been updated
if abs(np.mean(backup_ar) - command_line.superupdate_ar) > 5.*command_line.superupdate_ar_tol:
stop_c = False
# Update the jumping factor every 5 steps in superupdate
if not k % 5 and k > command_line.superupdate and command_line.superupdate and (not stop_c or (stop_c and k % command_line.update)):
try:
jump_file = open(command_line.folder + '/jumping_factor.txt','r')
# If there is a # in the file, the master has stopped adapting c
for line in jump_file:
if line.find('#') == -1:
jump_file.seek(0)
jump_value = jump_file.read()
data.jumping_factor = float(jump_value)
else:
jump_file.seek(0)
jump_value = jump_file.read().replace('# ','')
#if not stop_c or (stop_c and not float(jump_value) == data.jumping_factor):
if not float(jump_value) == data.jumping_factor:
data.jumping_factor = float(jump_value)
stop_c = True
data.out.write('# After %d accepted steps: stop adapting the jumping factor at a value of %f with a local acceptance rate %f \n' % (int(acc),data.jumping_factor,np.mean(backup_ar)))
if not command_line.silent:
print('After %d accepted steps: stop adapting the jumping factor at a value of %f with a local acceptance rate of %f \n' % (int(acc), data.jumping_factor,np.mean(backup_ar)))
jump_file.close()
except:
if not command_line.silent:
print('Reading jumping_factor file failed')
pass
# End of slave superupdate routine
# Start of slave update routine
if not (k-1) % (command_line.update/10):
try:
sigma_eig, U, C = sampler.get_covariance_matrix(
cosmo, data, command_line)
if command_line.jumping == 'fast':
Cholesky = la.cholesky(C).T
# Test here whether the covariance matrix has really changed
# We should in principle test all terms, but testing the first one should suffice
if not C[0,0] == previous[2][0,0] and not k == 1:
if command_line.superupdate:
# If the covmat was updated, the master has resumed adapting c
stop_c = False
data.out.write('# After %d accepted steps: update proposal \n' % int(acc))
if not command_line.silent:
print('After %d accepted steps: update proposal \n' % int(acc))
try:
if stop_after_update:
k = command_line.N
print('Covariance matrix updated - stopping run')
except:
pass
previous = (sigma_eig, U, C, Cholesky)
except:
pass
# End of slave update routine
# End of update routine
# Pick a new position ('current' flag in mcmc_parameters), and compute
# its likelihood. If get_new_position returns True, it means it did not
# encounter any boundary problem. Otherwise, just increase the
# multiplicity of the point and start the loop again
if get_new_position(
data, sigma_eig, U, k, Cholesky, Rotation) is True:
newloglike = sampler.compute_lkl(cosmo, data)
else: # reject step
rej += 1
if command_line.superupdate:
ar[k%len(ar)] = 0 # Local acceptance rate of last SU*(N_slow + f_fast * N_fast) steps
elif command_line.adaptive:
ar[k%len(ar)] = 0 # Local acceptance rate of last 100 steps
N += 1
k += 1
continue
# Harmless trick to avoid exponentiating large numbers. This decides
# whether or not the system should move.
if (newloglike != data.boundary_loglike):
if (newloglike >= loglike):
alpha = 1.
else:
alpha = np.exp(newloglike-loglike)
else:
alpha = -1
if ((alpha == 1.) or (rd.uniform(0, 1) < alpha)): # accept step
# Print out the last accepted step (WARNING: this is NOT the one we
# just computed ('current' flag), but really the previous one.)
# with its proper multiplicity (number of times the system stayed
# there).
io_mp.print_vector(outputs, N, loglike, data)
# Report the 'current' point to the 'last_accepted'
sampler.accept_step(data)
loglike = newloglike
if loglike > max_loglike:
max_loglike = loglike
acc += 1.0
N = 1 # Reset the multiplicity
if command_line.superupdate:
ar[k%len(ar)] = 1 # Local acceptance rate of last SU*(N_slow + f_fast * N_fast) steps
elif command_line.adaptive:
ar[k%len(ar)] = 1 # Local acceptance rate of last 100 steps
else: # reject step
rej += 1.0
N += 1 # Increase multiplicity of last accepted point
if command_line.superupdate:
ar[k%len(ar)] = 0 # Local acceptance rate of last SU*(N_slow + f_fast * N_fast) steps
elif command_line.adaptive:
ar[k%len(ar)] = 0 # Local acceptance rate of last 100 steps
# Store a.r. for last 5 x SU*(N_slow + f_fast * N_fast) steps
if command_line.superupdate:
backup_ar[k%len(backup_ar)] = ar[k%len(ar)]
# Regularly (option to set in parameter file), close and reopen the
# buffer to force to write on file.
if acc % data.write_step == 0:
io_mp.refresh_file(data)
# Update the outputs list
outputs[0] = data.out
k += 1 # One iteration done
# END OF WHILE LOOP
# If at this moment, the multiplicity is higher than 1, it means the
# current point is not yet accepted, but it also mean that we did not print
# out the last_accepted one yet. So we do.
if N > 1:
io_mp.print_vector(outputs, N-1, loglike, data)
# Print out some information on the finished chain
rate = acc / (acc + rej)
sys.stdout.write('\n# {0} steps done, acceptance rate: {1}\n'.
format(command_line.N, rate))
# In case the acceptance rate is too low, or too high, print a warning
if rate < 0.05:
warnings.warn("The acceptance rate is below 0.05. You might want to "
"set the jumping factor to a lower value than the "
"default (2.4), with the option `-f 1.5` for instance.")
elif rate > 0.6:
warnings.warn("The acceptance rate is above 0.6, which means you might"
" have difficulties exploring the entire parameter space"
". Try analysing these chains, and use the output "
"covariance matrix to decrease the acceptance rate to a "
"value between 0.2 and 0.4 (roughly).")
# For a restart, erase the starting point to keep only the new, longer
# chain.
if command_line.restart is not None:
os.remove(command_line.restart)
sys.stdout.write(' deleting starting point of the chain {0}\n'.
format(command_line.restart))
return
def _make_sense(self,u0):
st=self.st
L=numpy.shape(u0)[-1]
u0dims= numpy.ndim(u0)
print('in make_sense, u0.shape',u0.shape)
if u0dims-1 >0:
rows=numpy.shape(u0)[0]
# dpss_rows = numpy.kaiser(rows, 100)
# dpss_rows = numpy.fft.fftshift(dpss_rows)
# dpss_rows[3:-3] = 0.0
dpss_rows = numpy.ones(rows)
# replace above sensitivity because
# Frequency direction is not necessary
dpss_fil = dpss_rows
print('dpss shape',dpss_fil.shape)
if u0dims-1 > 1:
cols=numpy.shape(u0)[1]
dpss_cols = numpy.kaiser(cols, 100)
dpss_cols = numpy.fft.fftshift(dpss_cols)
dpss_cols[3:-3] = 0.0
dpss_fil = CsTransform.pynufft.appendmat(dpss_fil,cols)
dpss_cols = CsTransform.pynufft.appendmat(dpss_cols,rows)
dpss_fil=dpss_fil*numpy.transpose(dpss_cols,(1,0))
print('dpss shape',dpss_fil.shape)
if u0dims-1 > 2:
zag = numpy.shape(u0)[2]
dpss_zag = numpy.kaiser(zag, 100)
dpss_zag = numpy.fft.fftshift(dpss_zag)
dpss_zag[3:-3] = 0.0
dpss_fil = CsTransform.pynufft.appendmat(dpss_fil,zag)
dpss_zag = CsTransform.pynufft.appendmat(dpss_zag,rows)
dpss_zag = CsTransform.pynufft.appendmat(dpss_zag,cols)
dpss_fil=dpss_fil*numpy.transpose(dpss_zag,(1,2,0)) # low pass filter
print('dpss shape',dpss_fil.shape)
#dpss_fil=dpss_fil / 10.0
rms=numpy.sqrt(numpy.mean(u0*u0.conj(),-1)) # Root of sum square
st['sensemap']=numpy.ones(numpy.shape(u0),dtype=numpy.complex64)
print('sensemap shape',st['sensemap'].shape, L)
print('u0shape',u0.shape,rms.shape)
# print('L',L)
# print('rms',numpy.shape(rms))
for ll in xrange(0,L):
st['sensemap'][...,ll]=(u0[...,ll]+1e-16)/(rms+1e-16)
print('sensemap shape',st['sensemap'].shape, L)
print('rmsshape', rms.shape)
st['sensemap'][...,ll] = fftpack.fftn(st['sensemap'][...,ll],
st['sensemap'][...,ll].shape,
range(0,numpy.ndim(st['sensemap'][...,ll])))
st['sensemap'][...,ll] = st['sensemap'][...,ll] * dpss_fil
st['sensemap'][...,ll] = fftpack.ifftn(st['sensemap'][...,ll],
st['sensemap'][...,ll].shape,
range(0,numpy.ndim(st['sensemap'][...,ll])))
# st['sensemap'][...,ll]=fftpack.ifftn(fftpack.fftn(st['sensemap'][...,ll])*dpss_fil)
# st['sensemap'] = Normalize(st['sensemap'])
return st
