mult_and_shift = "post" arch_dict = { "D": system.D, "latent_dynamics": latent_dynamics, "mult_and_shift": mult_and_shift, "TIF_flow_type": TIF_flow_type, "repeats": nlayers, } k_max = 40 batch_size = 1000 lr_order = -3 min_iters = 1000 max_iters = 2000 train_dsn( system, batch_size, arch_dict, k_max=k_max, sigma_init=sigma_init, c_init_order=c_init_order, lr_order=lr_order, random_seed=random_seed, min_iters=min_iters, max_iters=max_iters, check_rate=100, dir_str="test", )
c_init_order, random_seed, dir_str, randsearch=randsearch) print('-- Hyperparameters --') print('batch_size:', batch_size) print('sigma_init:', sigma_init) print('c_init_order:', c_init_order) print('AL_fac:', AL_fac) print('epoch iters:', max_iters) train_dsn( system, arch_dict, batch_size, AL_it_max=AL_it_max, sigma_init=sigma_init, c_init_order=c_init_order, AL_fac=AL_fac, min_iters=min_iters, max_iters=max_iters, random_seed=random_seed, lr_order=lr_order, check_rate=100, dir_str='test', savedir=savedir, entropy=True, db=False, )
'repeats':repeats, 'nlayers':nlayers, 'upl':upl, 'sigma_init':sigma_init*np.ones((system.D,)) } param_dict.update(arch_params) arch_dict = get_arch_from_template(system, param_dict) AL_it_max = 20 batch_size = 1000 lr_order = -3 AL_fac = 4.0 iters = 5000 train_dsn( system, arch_dict, batch_size, AL_it_max=AL_it_max, c_init_order=c_init_order, AL_fac=AL_fac, min_iters=iters, max_iters=iters, random_seed=random_seed, lr_order=lr_order, check_rate=100, dir_str="V1Circuit_test", db=False )
"d_var": d_vars, } # set model options model_opts = {"g_FF": "c", "g_LAT": "linear", "g_RUN": "r"} T = 40 dt = 0.25 init_conds = np.expand_dims(np.array([1.0, 1.1, 1.2, 1.3]), 1) system = V1Circuit(fixed_params, behavior, model_opts, T, dt, init_conds) k_max = 20 batch_size = 1000 lr_order = -3 train_dsn( system, batch_size, arch_dict, k_max=k_max, sigma_init=sigma_init, c_init_order=c_init_order, lr_order=lr_order, random_seed=random_seed, min_iters=1000, max_iters=2000, check_rate=100, dir_str="V1Circuit_S_same_V_inc", )
mu = np.zeros((D, )) df_fac = 15 df = df_fac * D Sigma_dist = scipy.stats.invwishart(df=df, scale=df * np.eye(D)) Sigma = Sigma_dist.rvs(1) behavior = {"mu": mu, "Sigma": Sigma} print("Behavior:") print("mu") print(mu) print("Sigma") print(Sigma[0]) print(Sigma[1]) cost, phi, T_x = train_dsn( system, behavior, n, flow_dict, k_max=k_max, sigma_init=sigma_init, c_init_order=c_init_order, lr_order=lr_order, random_seed=random_seed, min_iters=min_iters, max_iters=max_iters, check_rate=check_rate, dir_str=dir_str, )
} param_dict.update(arch_params) arch_dict = get_arch_from_template(system, param_dict) n = 1000 AL_it_max = 20 AL_fac = 4.0 lr_order = -3 min_iters = 2000 max_iters = 2000 check_rate = 100 dist_seed = 0 dir_str = "2DLDS_test" np.random.seed(dist_seed) cost, z = train_dsn( system, arch_dict, n=n, AL_it_max=AL_it_max, c_init_order=c_init_order, AL_fac=AL_fac, min_iters=min_iters, max_iters=max_iters, random_seed=random_seed, lr_order=lr_order, check_rate=check_rate, dir_str=dir_str, db=False, )
"sigma_init": sigma_init, } param_dict.update(arch_params) arch_dict = get_arch_from_template(system, param_dict) batch_size = 300 AL_it_max = 40 AL_fac = 4.0 min_iters = 5000 max_iters = 5000 lr_order = -3 train_dsn( system, arch_dict, batch_size, AL_it_max=AL_it_max, c_init_order=c_init_order, AL_fac=AL_fac, min_iters=min_iters, max_iters=max_iters, random_seed=random_seed, lr_order=lr_order, check_rate=100, dir_str="SC_WTA_NI", savedir=None, entropy=True, db=False, )