def gpusim(td, param):
# parameters from DVParam
gamma = param[0]
a = param[1]
K = param[2]
nu = param[3]
r = param[4]
theta = param[5]
Vmax = param[6]
inittrait = param[7]
initpop = param[8]
initpop_sigma = param[9]
break_on_mu = bool(param[10])
sim_evo_time = td.sim_evo_time
events = td.sim_events
# Initialize trait evolution and population evolution matrices
trait_RI_dr = np.zeros((sim_evo_time + 1, td.total_species)) # trait
population_RI_dr = np.zeros(
(sim_evo_time + 1, td.total_species)).astype(np.int32) # population
V = np.zeros((sim_evo_time + 1, td.total_species)) # trait variance
# initialize condition for species trait and population
trait_RI_dr[0, (0, 1)] = inittrait # trait for species
population_RI_dr[0, (0, 1)] = np.random.normal(initpop, initpop_sigma,
2).astype(np.int32)
V[0] = (1 / td.total_species) # <----- why?
existing_species = td.traittable
node = 0
next_event = events[node]
idx = np.where(existing_species[node] == 1)[0] # existing species
# trait-population coevolution model
for i in range(sim_evo_time):
# pull current state
Ni = population_RI_dr[i, idx]
Vi = V[i, idx]
zi = trait_RI_dr[i, idx]
Ki = K
dtz = theta - zi
time_competition = time.time()
beta, sigma, sigmasqr = competition_functions(a, zi, Ni)
timeend_com = time.time()
if (i % 10000 == 0):
print("competition function costs %f s" %
(timeend_com - time_competition))
# update
update_time1 = time.time()
var_trait = Vi / (2.0 * Ni)
trait_RI_dr[i + 1,
idx] = zi + Vi * (2.0 * gamma * dtz +
1 / Ki * sigma) + np.random.normal(
0.0, var_trait)
mu = Ni * r * np.exp(-gamma * dtz**2 +
(1 - beta / Ki)) # un-truncated mean
if np.any(mu <= 1.0): # mu < 1.0 + 1.11e-16
if (break_on_mu):
print(i, "invalid mean population size")
break
ztp_lambda = dvcpp.ztp_lambda_from_untruncated_mean(mu)
population_RI_dr[i + 1, idx] = dvcpp.ztpoisson(ztp_lambda)
V[i + 1, idx] = Vi / 2.0 + 2.0 * Ni * nu * Vmax / (1.0 + 4.0 * Ni * nu) \
+ Vi**2 * (
-2.0 * gamma + 4.0 * gamma**2 * dtz**2 +
1.0 / Ki * (2.0 * a * beta - sigmasqr) + 4.0 * gamma / Ki *
dtz * sigma + sigma**2 / Ki**2
)
update_time2 = time.time()
if (i % 10000 == 0):
print("update function costs %f s" % (update_time2 - update_time1))
# events
while (i + 1) == next_event[0]:
event_time1 = time.time()
daughter = next_event[2]
if (daughter == -1):
# extinction
extinct_species = next_event[1]
V[i + 1, extinct_species] = None
trait_RI_dr[i + 1, extinct_species] = None
population_RI_dr[i + 1, extinct_species] = 0
else:
# speciation
parent = next_event[1]
parentN = population_RI_dr[i + 1, parent]
if parentN <= 1:
print(i, "attempt to split singleton")
# results in split <- 0, will be trapped by sanity check below
split = dvcpp.split_binomial50(parentN)
population_RI_dr[i + 1, daughter] = parentN - split
population_RI_dr[i + 1, parent] = split
V[i + 1, parent] *= 0.5
V[i + 1, daughter] = V[i + 1, parent]
trait_RI_dr[i + 1, daughter] = trait_RI_dr[i + 1, parent]
# advance to next event/node
node = node + 1
next_event = events[node]
idx = np.where(existing_species[node] == 1)[0]
event_time2 = time.time()
print("node event costs %f s" % (event_time2 - event_time1))
# sanity check
if np.any(population_RI_dr[i + 1, idx] < 1):
print(i, 'Inconsistent extinction')
break
if np.any(V[i + 1, idx] < 0.0) or np.any(V[i + 1, idx] > 100000.0):
print(i, 'runaway variance')
break
row_ext = np.where(population_RI_dr == 0)[0]
col_ext = np.where(population_RI_dr == 0)[1]
V[row_ext, col_ext] = None
trait_RI_dr[row_ext, col_ext] = None
return {'sim_time': i + 1, 'N': population_RI_dr, 'Z': trait_RI_dr, 'V': V}
def DVSimTVM(td, param):
# parameters from DVParamLiang
gamma = param[0]
a = param[1]
K = param[2]
h2 = param[3] * param[3]
nu = param[4]
r = param[5]
theta = param[6]
V00 = param[7]
V01 = param[8]
Vmax = param[9]
inittrait = param[10]
initpop = param[11]
initpop_sigma = param[12]
break_on_mu = bool(param[13])
sim_evo_time = td.sim_evo_time
events = td.sim_events
# Initialize trait evolution and population evolution matrices
trait_RI_dr = np.zeros((sim_evo_time + 1, td.total_species)) # trait
population_RI_dr = np.zeros(
(sim_evo_time + 1, td.total_species)).astype(np.int32) # population
existing_species = td.traittable
idx = np.where(existing_species[0] == 1)[0] # existing species
assert (idx.size == 2
), "number of existing species shall be 2 at simulation start"
# Initialize trait variances
V = np.zeros((sim_evo_time + 1, td.total_species))
V[0, idx] = [V00, V01]
# initialize condition for species trait and population
trait_RI_dr[0, idx] = inittrait # trait for species
population_RI_dr[0, idx] = np.random.normal(initpop, initpop_sigma,
2).astype(np.int32)
node = 0
next_event = events[node]
# trait-population co-evolution model, Liang
for i in range(sim_evo_time):
# pull current state
Ni = population_RI_dr[i, idx]
Vi = V[i, idx]
zi = trait_RI_dr[i, idx]
Ki = K
dtz = theta - zi
beta, sigma, sigmasqr = competition_functions_metabolism(a, zi, Ni)
# update
var_trait = Vi / (2.0 * Ni)
trait_RI_dr[i + 1,
idx] = zi + h2 * Vi * (2.0 * gamma * dtz +
1 / Ki * sigma) + np.random.normal(
0.0, var_trait)
mu = Ni * r * np.exp(-gamma * dtz**2 +
(1 - beta / Ki)) # un-truncated mean
if np.any(mu <= 1.0): # mu < 1.0 + 1.11e-16
if (break_on_mu):
# print(i, "invalid mean population size")
break
if np.any(mu > 2**28):
print(i, 'run-away lambda')
break
ztp_lambda = dvcpp.ztp_lambda_from_untruncated_mean(mu)
population_RI_dr[i + 1, idx] = dvcpp.ztpoisson(ztp_lambda)
V[i + 1, idx] = (1-h2/2.0)*Vi + 2.0*h2 * Ni * nu * Vmax / (1.0 + 4.0 * Ni * nu) \
+ h2/2.0 * Vi**2 * (
-2.0 * gamma + 4.0 * gamma**2 * dtz**2 +
1.0 / Ki * (2.0 * a * beta - sigmasqr) + 4.0 * gamma / Ki *
dtz * sigma + sigma**2 / Ki**2
)
# events
while (i + 1) == next_event[0]:
daughter = next_event[2]
if (daughter == -1):
# extinction
extinct_species = next_event[1]
V[i + 1, extinct_species] = None
trait_RI_dr[i + 1, extinct_species] = None
population_RI_dr[i + 1, extinct_species] = 0
else:
# speciation
parent = next_event[1]
parentN = population_RI_dr[i + 1, parent]
# if parentN <= 1:
# print(i, "attempt to split singleton")
# results in split <- 0, will be trapped by sanity check below
split = dvcpp.split_binomial50(parentN)
population_RI_dr[i + 1, daughter] = parentN - split
population_RI_dr[i + 1, parent] = split
V[i + 1, parent] *= 0.5
V[i + 1, daughter] = V[i + 1, parent]
trait_RI_dr[i + 1, daughter] = trait_RI_dr[i + 1, parent]
# advance to next event/node
node = node + 1
next_event = events[node]
idx = np.where(existing_species[node] == 1)[0]
# sanity check
if np.any(population_RI_dr[i + 1, idx] < 1):
# print(i, 'Inconsistent extinction')
break
if np.any(V[i + 1, idx] < 0.0) or np.any(V[i + 1, idx] > 100000.0):
print(i, 'runaway variance')
break
row_ext = np.where(population_RI_dr == 0)[0]
col_ext = np.where(population_RI_dr == 0)[1]
V[row_ext, col_ext] = None
trait_RI_dr[row_ext, col_ext] = None
return {
'sim_time': i + 1,
'N': population_RI_dr[-1],
'Z': trait_RI_dr[-1],
'V': V[-1]
}