def test_simulate_sample(): dataDict = loadData.load_sim_data('tests/test_simdata/') x, y, dur_dat, im_dat, densities_dat, d_range = loadData.dataDict2vars(dataDict) kwargs_d = {"coupled_oms": True, "exponents":1} sw_args = ['subtractive', 'zero', 'off', "off", 'off', {'x': d_range[0], 'y': d_range[1]}, kwargs_d] sw_model = scenewalk_obj(*sw_args) chains_dict = np.load("tests/test_estimation/sample_dict.npy", allow_pickle=True).item() sim_id = sim.simulate_sample(dur_dat, im_dat, densities_dat, sw_model, chains_dict, "trial", start_loc="center", x_path=None, y_path=None, resample_durs=False) #sim_id = "sim_20191025-101400" sim_id = sim_id[4:] cwd = os.getcwd() folderPath = cwd + "/sim_%s" % sim_id simdataDict = loadData.load_sim_data(folderPath+'/') x_sim, y_sim, dur_sim, im_sim, _, _ = loadData.dataDict2vars(simdataDict) for s, o in zip([x_sim, y_sim, dur_sim, im_sim], [x, y, dur_dat, im_dat]): assert len(s) == len(o) assert type(s) == type(o) testsub = np.random.randint(len(o)) assert len(s[testsub]) == len(o[testsub]) assert type(s[testsub]) == type(o[testsub]) testtrial = np.random.randint(len(o[testsub])) assert len(s[testsub][testtrial]) == len(o[testsub][testtrial]) assert type(s[testsub][testtrial]) == type(o[testsub][testtrial]) shutil.rmtree(folderPath)
def test_contributing_fixs(inhib_method, att_map_init_type, shifts, exponents, locdep_decay_switch, n_hist): dataDict = loadData.load_sim_data('tests/test_simdata/') x_sim, y_sim, dur_sim, _, densities, rng = loadData.dataDict2vars(dataDict) rng2 = {'x': rng[0], 'y': rng[1]} fix_dens = densities[0] lsw = lsw_model(inhib_method, att_map_init_type, shifts, locdep_decay_switch, "off", rng2, { "n_history": n_hist, "exponents": exponents }) lsw.update_params(sw_params) _, hist_list = lsw.get_scanpath_likelihood(x_sim[0][0], y_sim[0][0], dur_sim[0][0], fix_dens, debug=True) n_normal = [i[0] for i in hist_list if i[0] == 'normal'] n_chopped = [i[0] for i in hist_list if i[0] == 'chopped'] assert len(n_normal) == n_hist assert len(n_chopped) == (len(x_sim[0][0]) - len(n_normal) - 1) triplets = [i[1] for i in hist_list if i[0] == 'chopped'] n_triplets = [len(i) for i in triplets] assert (np.asarray(n_triplets) == n_hist).all
def test_simulate_scanpath(inhib_method, att_map_init_type, shifts, exponents, locdep_decay_switch): """Tests that simulated scanpaths the right type, range and length""" sw = sw_model(inhib_method, att_map_init_type, shifts, locdep_decay_switch, "off", { 'x': (23, 100), 'y': (0, 90) }, {"exponents": exponents}) sw.update_params(sw_params) dataDict = loadData.load_sim_data('tests/test_simdata/') _, _, dur_sim, _, densities, _ = loadData.dataDict2vars(dataDict) fix_dens = densities[0] startpos = (np.sum((23, 100)) / 2, np.sum((0, 90)) / 2) x_path, y_path, avg_log_ll = sw.simulate_scanpath(dur_sim[0][0], fix_dens, startpos, get_LL=True) assert len(x_path) == len(y_path) == len(dur_sim[0][0]) assert (np.asarray(x_path) <= 100).all() assert (np.asarray(x_path) >= 23).all() assert (np.asarray(y_path) <= 90).all() assert (np.asarray(y_path) >= 0).all() assert x_path[0] == startpos[0] assert y_path[0] == startpos[1] assert isinstance(avg_log_ll, np.float128)
def test_save2dict_by_subj(): # setup a pretend data set dataDict = loadData.load_sim_data('tests/test_simdata/') x, y, dur, im, densities, _ = loadData.dataDict2vars(dataDict) def_args = OrderedDict({ 'omegaAttention': None, 'omfrac': 10, 'sigmaAttention': None, "sigmaInhib" : None, 'gamma': None, #"lamb":None, 'inhibStrength': 0.3, 'zeta': None, 'sigmaShift': None, 'shift_size':None, 'first_fix_OmegaAttention':1.5, 'cb_sd_x':4, 'cb_sd_y':3, 'omega_prevloc_frac':10, }) all_vp_list = list(range(len(im))) chains_list = [np.random.rand(3, 500, 7) for v in all_vp_list] utils.save2dict_by_subj(chains_list, all_vp_list, def_args, "tests/test_estimation/sample_dict.npy") vp_params = np.load("tests/test_estimation/sample_dict.npy", allow_pickle=True).item() assert isinstance(vp_params, dict) assert list(vp_params.keys()) == all_vp_list testsub = np.random.randint(0, len(all_vp_list)) assert vp_params[testsub].keys() == def_args.keys() print(vp_params[testsub].values()) assert any([isinstance(v, (list, np.ndarray)) for v in vp_params[testsub].values()]) assert any([isinstance(v, (int, float)) for v in vp_params[testsub].values()])
def test_para_vs_seq_trials(inhib_method, att_map_init_type, shifts, exponents, locdep_decay_switch): """ Checks parallized result is the same as non parallelized evaluation """ # Load Data dataDict = loadData.load_sim_data(os.path.join(THIS_DIR, 'test_simdata/')) x_dat, y_dat, dur_dat, im_dat, densities_dat, dat_range = loadData.dataDict2vars( dataDict) sw = sw_model(inhib_method, att_map_init_type, shifts, locdep_decay_switch, "off", dat_range, {"exponents": exponents}) sw.update_params(sw_params) # get one subject dataset x_dat_sub = x_dat[0] y_dat_sub = y_dat[0] dur_dat_sub = dur_dat[0] im_dat_sub = im_dat[0] sequential_list = evaluate_sw_parallel.get_total_list_LL_trials( sw, x_dat_sub, y_dat_sub, dur_dat_sub, im_dat_sub, densities_dat) nLL_parallel = evaluate_sw_parallel.get_neg_tot_like_trials_parallel( sw, x_dat_sub, y_dat_sub, dur_dat_sub, im_dat_sub, densities_dat, 5) nLL_sequential = np.negative(np.sum(sequential_list)) assert np.isclose(nLL_parallel, nLL_sequential)
def test_para_vs_seq_both(inhib_method, att_map_init_type, shifts, exponents, locdep_decay_switch): """ Checks parallized result is the same as non parallelized evaluation """ # Load Data dataDict = loadData.load_sim_data(os.path.join(THIS_DIR, 'test_simdata/')) x_dat, y_dat, dur_dat, im_dat, densities_dat, dat_range = loadData.dataDict2vars( dataDict) sw = sw_model(inhib_method, att_map_init_type, shifts, locdep_decay_switch, "off", dat_range, {"exponents": exponents}) sw.update_params(sw_params) nLL_sequential = evaluate_sw_parallel.get_total_neg_LL_subjs( sw, x_dat, y_dat, dur_dat, im_dat, densities_dat, 1) nLL_parallel = evaluate_sw_parallel.get_neg_tot_like_parallel( sw, x_dat, y_dat, dur_dat, im_dat, densities_dat, 5, 3) assert np.isclose(nLL_parallel, nLL_sequential)
def test_get_scanpath_likelihood(inhib_method, att_map_init_type, shifts, exponents, locdep_decay_switch): """Tests that returned scanpath likelihhods are not nan and have the right type""" sw = sw_model(inhib_method, att_map_init_type, shifts, locdep_decay_switch, "off", { 'x': (23, 100), 'y': (0, 90) }, {"exponents": exponents}) sw.update_params(sw_params) dataDict = loadData.load_sim_data('tests/test_simdata/') x_sim, y_sim, dur_sim, _, densities, _ = loadData.dataDict2vars(dataDict) fix_dens = densities[0] avg_log_ll = sw.get_scanpath_likelihood(x_sim[0][0], y_sim[0][0], dur_sim[0][0], fix_dens) assert isinstance(avg_log_ll, np.float128) assert not np.isnan(avg_log_ll)
def main(): datadict = loadData.load_data("corpus_training") datadict = loadData.chop_scanpaths(1, 10, datadict) x_dat, y_dat, dur_dat, im_dat, densities_dat, dat_range = loadData.dataDict2vars( datadict) dat_range_x = dat_range[0] dat_range_y = dat_range[1] prior_args = np.load('priors.npy', allow_pickle=True).item() def_dict = np.load('defaults.npy', allow_pickle=True).item() priors = OrderedDict() for p in prior_args.keys(): priors[p] = dream_estim.trpd(*prior_args[p]) sw_args = np.load('sw_args.npy', allow_pickle=True) sw_model = scenewalk_obj(*sw_args) vp = int(sys.argv[1]) - 1 # Ressource Allocation num_processes_subperdream = 1 num_processes_trials = 6 nchains = 3 niter = 10 dream_estim.dream_estim_and_save(sw_model, priors, def_dict, np.asarray([x_dat[vp]]), np.asarray([y_dat[vp]]), np.asarray([dur_dat[vp]]), np.asarray([im_dat[vp]]), densities_dat, num_processes_subperdream, num_processes_trials, nchains, niter, vp_nr=vp, destin_dir=".")
def bla(): inhib_method = "subtractive" att_map_init_type = "cb" shifts = "both" exponents = 1 locdep_decay_switch = "on" sw_params = OrderedDict({ "omegaAttention" : 1, "omegaInhib": 0.1, "omfrac": 2, "sigmaAttention" : 5, "sigmaInhib" : 4, "gamma" : 1, "lamb" : 1, "inhibStrength" : 0.01, "zeta" : 0.01, "sigmaShift" : 5, "shift_size" : 2, "phi" : 100, "first_fix_OmegaAttention" : 3, "cb_sd_x" : 5, "cb_sd_y" : 4, "omega_prevloc": 1, "chi": 0.5, "ompfactor":0.1 }) sw = sw_model(inhib_method, att_map_init_type, shifts, exponents, locdep_decay_switch, {'x': (23, 100), 'y': (0, 90)}, {"omp":"add"}) sw.update_params(sw_params) dataDict = loadData.load_sim_data('test_simdata/') x_sim, y_sim, dur_sim, _, densities, _ = loadData.dataDict2vars(dataDict) fix_dens = densities[0] avg_log_ll = sw.get_scanpath_likelihood(x_sim[0][0], y_sim[0][0], dur_sim[0][0], fix_dens)
def test_something(inhib_method, att_map_init_type, shifts, exponents, locdep_decay_switch): dataDict = loadData.load_sim_data('tests/test_simdata/') chop_dataDict = loadData.chop_scanpaths(0, 10, dataDict) x_sim, y_sim, dur_sim, _, densities, rng = loadData.dataDict2vars( chop_dataDict) rng2 = {'x': rng[0], 'y': rng[1]} fix_dens = densities[0] sw = sw_model(inhib_method, att_map_init_type, shifts, locdep_decay_switch, "off", rng2, {"exponents": exponents}) sw.update_params(sw_params) lsw = lsw_model(inhib_method, att_map_init_type, shifts, locdep_decay_switch, "off", rng2, { "n_history": 10, "exponents": exponents }) lsw.update_params(sw_params) sw_LL = sw.get_scanpath_likelihood(x_sim[0][0], y_sim[0][0], dur_sim[0][0], fix_dens) lsw_LL = lsw.get_scanpath_likelihood(x_sim[0][0], y_sim[0][0], dur_sim[0][0], fix_dens) assert sw_LL == lsw_LL