def get_trials(params, n_rep=100000):
'''
Generates n_rep number of facilitation curves for Go response for all simulated trials required
Parameters
-------------
params : sequence (4,) of float
k_facGo - scale of fac curve
pre_t_mean - average start time before target presentation
pre_t_sd - standard deviation of start time before target
Returns
--------
fac_i : array
facilitation curves for all simulated trials
t : array
sequence of time index
'''
k_facGo, pre_t_mean, pre_t_sd, tau_facGo, inhib, inhib_sd = params
t = np.linspace(-.4, .2, 600, endpoint=False)
# tau_facGo = 2 # Currently set, but will need to optomize
pre_t = np.random.normal(pre_t_mean, pre_t_sd, size=n_rep) # generates n_rep random numbers from a normal distribution of mean, sd that given into function
fac_i = np.zeros((n_rep, t.size)) # had to change from fac_i, t - why does this cause error now?!?! sets up empty array of zeros for all simulated trials
for i in range(n_rep): # for each simulated trial
myparams = k_facGo, tau_facGo, pre_t[i] # takes parameters passed into model plus pre_t number randomly generated for that simulated trial
fac_i[i] = fast.get_fac(t, myparams) # generates curve for that simulated trial
return fac_i, t
def get_trials(params, n_rep=100000):
'''
Generates n_rep number of facilitation curves for Go response for all simulated trials required
Parameters
-------------
params : sequence (4,) of float
k_facGo - scale of fac curve
pre_t_mean - average start time before target presentation
pre_t_sd - standard deviation of start time before target
Returns
--------
fac_i : array
facilitation curves for all simulated trials
t : array
sequence of time index
'''
k_facGo, pre_t_mean, pre_t_sd, tau_facGo, inhib = params
t = np.linspace(-.4, .2, 600, endpoint=False, dtype=np.float32)
# tau_facGo = 2 # Currently set, but will need to optomize
pre_t = np.array(np.random.normal(pre_t_mean, pre_t_sd, size=n_rep), dtype=np.float32)
fac_i_parallel = np.zeros((n_rep, t.size), dtype=np.float32)
if PAR_TEST:
fac_i = np.zeros((n_rep, t.size), dtype=np.float32)
t_start = time()
for i in range(n_rep): # for each simulated trial
myparams = k_facGo, tau_facGo, pre_t[i]
#fac_i[i] = get_fac(t, myparams)
fac_i[i] = fast.get_fac(t, myparams)
t_end = time()
s_time = t_end - t_start
print "Serial time: %.3f s" % s_time
# Used for testing get_fac_parallel, it will fill the array fac_i_parallel
#get_fac_parallel(fac_i_parallel, n_rep, t, len(t), k_facGo, tau_facGo, pre_t)
# Setup CUDA variables
tpb_x = 8 # threads per block in x dimension
tpb_y = 8 # threads per block in y dimension
block_dim = tpb_x, tpb_y
bpg_x = int(n_rep / tpb_x) + 1 # block grid x dimension
bpg_y = int(t.size / tpb_y) + 1 # block grid y dimension
grid_dim = bpg_x, bpg_y
t_start = time()
stream = cuda.stream()
with stream.auto_synchronize():
d_fac = cuda.to_device(fac_i_parallel, stream)
d_t = cuda.to_device(t, stream)
d_pre_t = cuda.to_device(pre_t, stream)
print "CUDA kernel: Block dim: ({tx}, {ty}), Grid dim: ({gx}, {gy})".format(tx=tpb_x, ty=tpb_y, gx=bpg_x, gy=bpg_y)
get_fac_cuda[grid_dim, block_dim](d_fac, n_rep, t, len(t), k_facGo, tau_facGo, pre_t)
d_fac.to_host(stream)
t_end = time()
c_time = t_end - t_start
print "CUDA time: %.3f s" % c_time
if PAR_TEST:
print "Difference betwwen fac_i and fac_i_parallel"
print (fac_i - fac_i_parallel)
print "Close enough? ", np.allclose(fac_i, fac_i_parallel, rtol=0, atol=1e-05)
print "Speed up: %.3f x" % (s_time / c_time)
return fac_i_parallel, t
