def generate_data(t, y0, func, func_name, data_type, num_traj):
'''
NOTES: Generates multiple data trajectories (number specified by num_traj) starting near y0.
Saves this data as 3D array with axes
- 0 = ith trajectory
- 1 = point at time t_i
- 2 = spatial dimension y_i
INPUT:
t = vector where elements are times at which to store solution (t subset of [args.T0,args.Tend], containing endpoints)
y0 = initial position data
func = function defining ODE used to generate data
func_name = string with name of ODE
data_type = string with label for data (e.g. training, validation, testing)
num_traj = number of trajectories to generate
OUTPUT:
None
'''
assert (len(t.shape) == 1), 't must be a 1D array.'
assert ((args.T0 <= min(t)) &
(max(t) <= args.Tend)), 't must be subset of [T0,Tend].'
assert (len(y0.shape) == 1), 'y0 must be 1D array.'
assert (type(func_name) == str), 'func_name must be string.'
assert (type(data_type) == str), 'data_type must be string.'
assert (type(num_traj) == int), 'num_traj must be integer.'
data_y = []
for _ in range(num_traj):
y = solve_ODE(t, y0, func)
data_y.append(y)
data_y = np.stack(data_y, axis=0)
#data_v = func(t=None,y=data_y) # exact velocity calculated
#data_v = fwd_euler(t=None,y=data_y) # velocity calculated using forward euler
data_v = central_diff(
t=None, y=data_y) #velocity calculated using central difference
# uncomment this part if using fwd_euler or central_diff
if len(data_y.shape) == 1:
data_y = data_y[1:-1]
else:
data_y = data_y[:, 1:-1, :]
if data_y.shape[2] == 3:
plot_3D(data_y, func_name, args.data_dir,
data_type) # visualize solution
# save data
np.save(args.data_dir + '/' + data_type + '_y', data_y)
np.save(args.data_dir + '/' + data_type + '_v', data_v)
def generate_data(t, y0, func, func_name, data_type, num_traj):
'''
NOTES: Generates multiple data trajectories (number specified by num_traj) starting near y0.
Saves this data as 3D array with axes
- 0 = ith trajectory
- 1 = point at time t_i
- 2 = spatial dimension y_i
INPUT:
t = vector where elements are times at which to store solution (t subset of [args.T0,args.Tend], containing endpoints)
y0 = initial position data
func = function defining ODE used to generate data
func_name = string with name of ODE
num_traj = number of trajectories to generate
OUTPUT:
None
'''
assert (len(t.shape) == 1), 't must be a 1D array.'
assert ((args.T0 <= min(t)) &
(max(t) <= args.Tend)), 't must be subset of [T0,Tend].'
assert (type(func_name) == str), 'func_name must be string.'
assert (type(num_traj) == int), 'num_traj must be integer.'
data_y = []
for _ in range(num_traj):
if args.ODE_system == 'Glycolytic': y0 = initial(y0)
y = solve_ODE(t, y0, func)
data_y.append(y)
data_y = np.stack(data_y, axis=0)
# add noise
if args.noise != None:
tot_num_traj = data_y.shape[0]
for traj in range(tot_num_traj):
y_vals = data_y[traj, :, 2]
noiseSigma = args.noise * y_vals
mu, sigma = 0, 1
noise = noiseSigma * np.random.normal(mu, sigma, args.num_point)
data_y[traj, :, 2] += noise
# save data for model 1
if args.split_method == 1:
np.save(args.data_dir + '/data_y', data_y)
if data_y.shape[2] == 3:
plot_3D(data_y, func_name, args.data_dir,
data_type) # visualize solution
# split and save train/val/test based on model 2
elif args.split_method == 2:
train_y = data_y[:, :args.T_ss, :]
val_y = data_y[:args.V_ss, args.T_ss:, :]
test_y = data_y[args.V_ss:, args.T_ss:, :]
np.save(args.data_dir + '/' + 'train' + '_y', train_y)
np.save(args.data_dir + '/' + 'val' + '_y', val_y)
np.save(args.data_dir + '/' + 'test' + '_y', test_y)
if data_y.shape[2] == 3:
plot_3D(train_y, func_name, args.data_dir,
'training') # visualize solution
plot_3D(val_y, func_name, args.data_dir, 'validation')
plot_3D(test_y, func_name, args.data_dir, 'testing')