def draw_sample(output_dist, sample_temperature, sample_argmax, _rnd):
if sample_argmax:
output_dist = np.eye(256)[np.argmax(output_dist, axis=-1)]
else:
if sample_temperature is not None:
output_dist = softmax(output_dist, sample_temperature)
output_dist = output_dist / np.sum(output_dist + 1e-7)
output_dist = random.multinomial(1, output_dist)
return output_dist
def experiment(mean_b, n_max=POP_MAX, n0=INITIAL_CELLS):
"""
Simulate the fate of the cell population. Random variables are drawn from a multinomial law to determine the fate of the cells at each step, which is equivalent to drawing a random variable for each cell independently.
Args:
mean_b (float): value of the adaptation timing.
n_max (int, optional): maximum population allowed, defaults to POP_MAX.
n_0 (int, optional): initial population, defaults to INITIAL_CELLS.
Returns:
int : number of steps needed to fill the medium with healthy (repaired) cells
"""
damaged_cells = n0
repaired_cells = 0
adapted_cells = 0
dead_cells = 0
n = 0
while repaired_cells < n_max:
a = alpha_sto(n)
b = beta_sto(n, mean_b)
# Fate of damaged cells
draw = rd.multinomial(damaged_cells, [max(0,1-a-b-GAMMA_DAM), a, b, min(1-a-b,GAMMA_DAM)]) # Draw a multinomial law
damaged_cells = draw[0]
repaired_cells += draw[1]
adapted_cells += draw[2]
dead_cells += draw[3]
# Fate of adapted cells
draw = rd.multinomial(adapted_cells, [DELTA, GAMMA_AD, 1-DELTA-GAMMA_AD]) # Draw a multinomial law
repaired_cells = min(n_max, repaired_cells + draw[0])
dead_cells += draw[1]
adapted_cells = draw[2]
# Reproduce cells is space is available
if damaged_cells + repaired_cells + adapted_cells < n_max :
r = r = reproduce_sto(damaged_cells, adapted_cells, repaired_cells)
adapted_cells += r[0]
repaired_cells += r[1]
n+=1
return n
def generate_logistic_multinomial(X: spa.csr_matrix, W: np.array, b: np.array):
X1 = np.c_[np.ones(len(X)), X]
bW = np.c_[b, W]
nu = np.dot(X1, bW.T)
enu = np.exp(nu)
enu_sum = enu.sum(axis=1)
pi = enu / enu_sum[:, np.newaxis]
Z = np.empty_like(pi)
for i1 in range(len(pi)):
Z[i1] = random.multinomial(1, pi[i1], 1)
return Z
def generate_bsample(duplicate_counts,n): new_counts= []; new_counts.append(duplicate_counts[0]); new_counts.append(duplicate_counts[1]); for i in xrange(2,n): N = duplicate_counts[i][1]; if N < 2: new_counts.append(duplicate_counts[i]); continue; pvals = []; for j in xrange(2,len(duplicate_counts[i])): pvals.append(float(duplicate_counts[i][j])/N); #print >>sys.stderr, "PVALS",pvals,duplicate_counts[i] S = random.multinomial(N,pvals); new_counts.append(duplicate_counts[i]); for j in xrange(2,len(duplicate_counts[i])): new_counts[-1][j] = S[j-2]; print 'BOOTSTRAP-counts',i,new_counts[i]; return new_counts;
def generate_bsample(duplicate_counts, n):
new_counts = []
new_counts.append(duplicate_counts[0])
new_counts.append(duplicate_counts[1])
for i in xrange(2, n):
N = duplicate_counts[i][1]
if N < 2:
new_counts.append(duplicate_counts[i])
continue
pvals = []
for j in xrange(2, len(duplicate_counts[i])):
pvals.append(float(duplicate_counts[i][j]) / N)
#print >>sys.stderr, "PVALS",pvals,duplicate_counts[i]
S = random.multinomial(N, pvals)
new_counts.append(duplicate_counts[i])
for j in xrange(2, len(duplicate_counts[i])):
new_counts[-1][j] = S[j - 2]
print 'BOOTSTRAP-counts', i, new_counts[i]
return new_counts