def logpmf(X, N, P):
assert X.shape == N.shape == P.shape
# Limit to one-dimension arrays, as indexing an array using a boolean array
# of >1D (as I do with `results[idxs] = val`) is not supported by Numba.
assert X.ndim == N.ndim == P.ndim == 1
P_isclose_0 = util.isclose(0, P)
P_isclose_1 = util.isclose(1, P)
X_equals_0 = X == 0
X_equals_N = X == N
special = [
(np.logical_and(P_isclose_0, X_equals_0), 0),
(np.logical_and(P_isclose_0, np.logical_not(X_equals_0)), -np.inf),
(np.logical_and(P_isclose_1, X_equals_N), 0),
(np.logical_and(P_isclose_1, np.logical_not(X_equals_N)), -np.inf),
]
result = np.full_like(P, np.nan)
for idx in range(len(special)):
idxs = special[idx][0]
val = special[idx][1]
assert np.all(np.isnan(result[idxs]))
result[idxs] = val
unfilled = np.isnan(result)
Nu = N[unfilled]
Xu = X[unfilled]
Pu = P[unfilled]
result[unfilled] = util.log_N_choose_K(
Nu, Xu) + Xu * np.log(Pu) + (Nu - Xu) * np.log(1 - Pu)
assert not np.any(np.isnan(result))
return result
def done_command(label):
# TODO(alive): Rewrite with the paradigm used in timer_db.py.
# Move this logic into Entry.
if not os.path.isfile(_resource_path(label)):
util.tlog("No diary entry with label `%s` exists" % label)
return
with util.OpenAndLock(_resource_path(label), 'r') as f:
entry = json.load(f)
now = time.time()
span = now - entry['epoch']
effective = entry['effective']
# Handle orderings:
# 1. __enter__, new, done, __exit__.
# 2. new, done, __enter__, __exit__.
# 3. __enter__, __exit__, new, done.
#
# If we are in any of the above orderings AND effective is 0.0, then we
# simply set `effective` to `span`. In these cases, there is no interaction
# between diary and timer.
#
# If, however, the first condition is True, but the second is false, then
# we must be in case #1 above. The only way for `effective` to be non-zero
# here is for the user to have called timer.inc(). This is only possible
# if a timer is running, and therefore, cases #2 and #3 are ruled out. The
# else block handles this case.
if (util.isclose(entry['epoch'], entry['interval_start_time'])
and util.isclose(effective, 0.0)):
effective = span
else:
# Handle orderings:
# 1. __enter__, new, done, __exit__ (with call to timer.inc()).
# 5. new, __enter__, done, __exit__.
# Capture the amount of time elapsed after __enter__.
timer = timer_db.running_timer()
if timer:
with timer:
if timer.label == label:
effective += time.time() - entry['interval_start_time']
if util.isclose(span - effective, 0.0, abs_tol=_TIME_EPSILON):
overhead = 0.0
else:
overhead = (span - effective) / span
click.echo(" Start time: %s" % _format_timestamp(entry['epoch']))
click.echo(" End time: %s" % _format_timestamp(now))
click.echo(" Span (m): %.2f" % (span / 60.0))
click.echo(" Effective (m): %.2f" % (effective / 60.0))
click.echo(" Overhead (%%): %.1f%%" % (overhead * 100.0))
os.remove(_resource_path(label))
def calc_mut_p(A, Z, psi):
eta = softmax(psi) # Kx1
phi = np.dot(Z, eta) # Kx1
mut_phi = np.dot(A, phi) # Mx1
# TODO: should this use omega_v?
mut_p = 0.5 * mut_phi
# Avoid numerical issues.
delta = 1e-30
mut_p[util.isclose(mut_p, 0)] += delta
mut_p[util.isclose(mut_p, 0.5)] -= delta
return mut_p
def merge_one_station(self, window, strategy):
new_window = {}
for channel_id, channel_win in window.iteritems():
winnum = self.find_duplicate_channel(window, channel_id)
if len(winnum) == 1:
weighting = 1.0
if len(winnum) > 1:
weighting = self.calculate_weighting(
channel_id, winnum, strategy)
if self._verbose == 2:
print("%s" % str(channel_id)),
print(winnum),
print(" --> weighting: %.2f" % (weighting))
if isclose(weighting, 0.0):
continue
new_window[channel_id] = []
for win in channel_win:
newwin = {}
newwin["initial_weighting"] = weighting
channel_id = win["channel_id"]
content = channel_id.split(".")
newwin["obsd_id"] = channel_id
newwin["synt_id"] = "%s.%s.S3.MX%s" % (content[0], content[1],
content[3][-1])
newwin["relative_starttime"] = win["relative_starttime"]
newwin["relative_endtime"] = win["relative_endtime"]
new_window[channel_id].append(newwin)
return new_window
def merge_one_station(self, window, strategy):
new_window = {}
for channel_id, channel_win in window.iteritems():
winnum = self.find_duplicate_channel(window, channel_id)
if len(winnum) == 1:
weighting = 1.0
if len(winnum) > 1:
weighting = self.calculate_weighting(channel_id, winnum,
strategy)
if self._verbose == 2:
print("%s" % str(channel_id)),
print(winnum),
print(" --> weighting: %.2f" % (weighting))
if isclose(weighting, 0.0):
continue
new_window[channel_id] = []
for win in channel_win:
newwin = {}
newwin["initial_weighting"] = weighting
channel_id = win["channel_id"]
content = channel_id.split(".")
newwin["obsd_id"] = channel_id
newwin["synt_id"] = "%s.%s.S3.MX%s" % (content[0], content[1],
content[3][-1])
newwin["relative_starttime"] = win["relative_starttime"]
newwin["relative_endtime"] = win["relative_endtime"]
new_window[channel_id].append(newwin)
return new_window
def test_compute_significance_two_tails(self):
cases = [(5, 22, 0.0169), (16, 22, 0.0525), (100, 220, 0.2001),
(300, 640, 0.1231)]
for success, total, expected in cases:
assert util.isclose(compute_significance_two_tails(success, total),
expected,
rel_tol=1e-03)
def mkcells(gidinfo):
timeit()
for gid in gidinfo:
x, y, z = gidinfo[gid]
cell = h.Cell()
gidinfo[gid] = CellInfo(cell)
# cell shape is actually an arc and area is fastidious with respect
# to all 6 sides. But length
# treated as line distance between org points (interior corners
# in circumferential direction. Set diam so area is correct including
# end areas.
cell.soma.pt3dclear()
cell.soma.pt3dadd(x, y, z, 1.)
ilayer, icircle, ipt = gid2org(gid)
x1, y1, z1 = xyz(ilayer, icircle, ipt + 1)
cell.soma.pt3dadd(x1, y1, z1, 1.)
length = cell.soma.L
area = sum(mkgap.cell_side_areas(gid))
diam = area / pi / length
cell.soma.diam = diam
assert (isclose(cell.soma(.5).area(), area, abs_tol=area * 1e-5))
cell.position(x, y, z)
pc.set_gid2node(gid, rank)
nc = cell.connect2target(None)
pc.cell(gid, nc)
x = pc.allreduce(len(gidinfo), 1)
pr("Global number of real cells is %d" % x)
timeit("mkcells")
def update(self):
'''
This function updates the Button image with the reward,
and direction values
:return: None
'''
# If statement is to stop the termination squares from being updated
if len(self.children) > 0:
max_val = max(self.direction_values)
max_index = self.direction_values.index(max_val)
# If two direction_values are about the same as the max value,
# then don't display any arrow images
max_count = 0
for dir_val in self.direction_values:
if util.isclose(dir_val, max_val):
max_count += 1
if max_count is 1:
self.children[0].opacity = 1
self.children[0].source = self.background_images[max_index]
else:
self.children[0].opacity = 0
self.children[1].set_strength(max_val * 4)
self.children[1].color = self.colour
if self.colour.rgb != [1.0, 1.0, 1.0]:
self.children[1].draw()
def _do_gibbs_iter(V,
T_prime,
phi_alpha0,
phi_beta0,
logconc,
C,
Z,
check_full_llh=False):
N = len(Z)
Z = np.copy(Z)
for vidx in range(N):
old_cluster = Z[vidx]
Z[vidx] = -1
if not np.any(Z == old_cluster):
# If `vidx` was the only member, remove this cluster.
# Do so by moving the last cluster to its index.
# (These operations are valid even if `highest == old_cluster`.)
highest = C - 1
Z[Z == highest] = old_cluster
C -= 1
# `cweights`: LLHs of each cluster destination for `vidx`
# cweights[C] = LLH of adding new cluster
cweights = np.empty(C + 1)
# Consider every possible destination.
for cidx in range(C):
members = np.flatnonzero(Z == cidx)
cweights[cidx] = _calc_cweight(V, T_prime, phi_alpha0, phi_beta0,
vidx, members)
# Consider adding a new cluster.
cweights[C] = _calc_new_cluster_weight(V, T_prime, phi_alpha0,
phi_beta0, vidx, logconc)
cweights -= np.log(np.exp(logconc) + N - 1)
if check_full_llh:
cweights_full = _compute_cweights_full(V, T_prime, Z, phi_alpha0,
phi_beta0, vidx, logconc, C)
assert np.all(util.isclose(cweights, cweights_full))
cprobs = util.softmax(cweights)
new_cluster = util.sample_multinom(cprobs)
Z[vidx] = new_cluster
if new_cluster == C:
C += 1
llh = _calc_llh(V, T_prime, Z, phi_alpha0, phi_beta0, logconc)
return (C, Z, llh)
def _integral_separate_clusters(args):
phi1, V1_var_reads, V1_ref_reads, V1_omega, V2_var_reads, V2_ref_reads, V2_omega, midx, logsub = args
V1_total_reads = V1_var_reads + V1_ref_reads
logP = _binom_logpmf(
V1_var_reads,
V1_total_reads,
V1_omega * phi1,
)
upper = _make_upper(phi1, midx)
lower = _make_lower(phi1, midx)
A = V2_var_reads + 1
B = V2_ref_reads + 1
betainc_upper = betacdf(A, B, V2_omega * upper)
betainc_lower = betacdf(A, B, V2_omega * lower)
if util.isclose(betainc_upper, betainc_lower):
return 0
logP += np.log(betainc_upper - betainc_lower)
logP -= logsub
return np.exp(logP)
def is_running(self): return not util.isclose(self.endtime, 0, abs_tol=1e-3)
figure_index = 0
# First re-type distance and temperature arrays for pyplot:
distances = np.array([[distances[index]-distances[0] for index in range(0,len(distances))] for distances in distance_averages_for_each_num_states])
temperatures = np.array([[Temp[index] for index in range(0,len(Temp))] for Temp in temperatures_for_each_num_states])
distance_uncertainty = np.array([[uncertainty[index] for index in range(0,len(uncertainty))] for uncertainty in uncertainty_distances_for_each_num_states])
# Define temporary arrays to store the MBAR-evaluated, re-weighted properties at T = 400 K
states_temp = []
distances_temp = []
uncertainty_temp = []
for state_index in range(0,len(temperatures)):
# If the temperature distribution for 'num_states' includes a temperature within 'tol' (10.0 K) of 400 K, we will include it in our analysis, otherwise we omit the dataset
#
# For example, if num_states = 4, the temperatures we sample are: 300.0, 366.6, 433.3, and 500.0. Thus, we won't plot the results for 'num_states' = 4
if any(util.isclose(temperatures[state_index][temperature_index],400.0,tol=10.0) for temperature_index in range(0,len(temperatures[state_index]))):
for temperature_index in range(0,len(temperatures[state_index])):
if temperatures[state_index][temperature_index] == float(400.0):
states_temp.append(len(temperatures[state_index]))
distances_temp.append(distance_averages_for_each_num_states[state_index][temperature_index])
uncertainty_temp.append(uncertainty_distances_for_each_num_states[state_index][temperature_index])
break
# Define arrays with the datapoints we will be plotting
states_400 = np.array([states_temp[index] for index in range(0,len(states_temp))])
uncertainty_400 = np.array([uncertainty_temp[index] for index in range(0,len(uncertainty_temp))])
distances_400 = np.array([distances_temp[index] for index in range(0,len(distances_temp))])
#Plot <R_Na-Cl> @ 400 K vs. # of thermodynamic states
figure = pyplot.figure(figure_index)
pyplot.plot(states_400,distances_400,figure=figure)
pyplot.xlabel("Number of thermodynamic states")
temperatures = np.array([[Temp[index] for index in range(0, len(Temp))]
for Temp in temperatures_for_each_num_states])
distance_uncertainty = np.array(
[[uncertainty[index] for index in range(0, len(uncertainty))]
for uncertainty in uncertainty_distances_for_each_num_states])
# Define temporary arrays to store the MBAR-evaluated, re-weighted properties at T = 400 K
states_temp = []
distances_temp = []
uncertainty_temp = []
for state_index in range(0, len(temperatures)):
# If the temperature distribution for 'num_states' includes a temperature within 'tol' (10.0 K) of 400 K, we will include it in our analysis, otherwise we omit the dataset
#
# For example, if num_states = 4, the temperatures we sample are: 300.0, 366.6, 433.3, and 500.0. Thus, we won't plot the results for 'num_states' = 4
if any(
util.isclose(
temperatures[state_index][temperature_index], 400.0, tol=10.0)
for temperature_index in range(0, len(temperatures[state_index]))):
for temperature_index in range(0, len(temperatures[state_index])):
if temperatures[state_index][temperature_index] == float(400.0):
states_temp.append(len(temperatures[state_index]))
distances_temp.append(
distance_averages_for_each_num_states[state_index]
[temperature_index])
uncertainty_temp.append(
uncertainty_distances_for_each_num_states[state_index]
[temperature_index])
break
# Define arrays with the datapoints we will be plotting
states_400 = np.array(
[states_temp[index] for index in range(0, len(states_temp))])
uncertainty_400 = np.array(