import gudhi
from scripts.ssc.persistence_pairings_visualization.utils_definitions import make_plot
from src.datasets.datasets import SwissRoll
if __name__ == "__main__":
dataset_sampler = SwissRoll()
n_points = 2048
seed = 13
samples, color = dataset_sampler.sample(n_points, seed=seed)
tc = gudhi.TangentialComplex(intrisic_dim=1, points=samples)
tc.compute_tangential_complex()
simplex_tree = tc.create_simplex_tree()
print(simplex_tree.get_skeleton(1))
skeleton_sorted = sorted(simplex_tree.get_skeleton(1), key=lambda t: t[1])
pairings = []
for element in skeleton_sorted:
pair = element[0]
if len(pair) == 2 and element[1] == 0:
print(pair)
pairings.append(pair)
make_plot(samples, pairings, color, name='witness_TEST')
dataset_sampler = SwissRoll()
N_WITNESSES = 2048
n_samples = 128
path_to_save = '/Users/simons/PycharmProjects/MT-VAEs-TDA/output/visualisation_nnsys/wc{}_w{}/'.format(
n_samples, N_WITNESSES)
N_sim = 100
ks = [1, 2, 3, 4, 6, 8, 12, 16]
ntot = int(len(ks) * N_sim)
counter = 1
for seed in list(set(np.random.randint(1, 100000, N_sim))):
witnesses, color_ = dataset_sampler.sample(N_WITNESSES, seed=seed)
ind = random.sample(range(N_WITNESSES), n_samples)
landmarks, color = witnesses[ind, :], color_[ind]
witnesses_tensor = torch.from_numpy(witnesses)
landmarks_tensor = torch.from_numpy(landmarks)
witness_complex = WitnessComplex(landmarks_tensor, witnesses_tensor)
witness_complex.compute_simplicial_complex(d_max=1,
r_max=10,
create_simplex_tree=False,
create_metric=True)
for k in ks:
import time
from sklearn.manifold import SpectralEmbedding
from src.datasets.datasets import SwissRoll, Spheres
from src.utils.plots import plot_2Dscatter
if __name__ == "__main__":
dataset = SwissRoll()
data, color = dataset.sample(n_samples=2560)
start = time.time()
embedding = SpectralEmbedding(n_components=2,n_jobs=1, n_neighbors=90)
X_transformed = embedding.fit_transform(data)
end = time.time()
print('It took: {}'.format(end - start))
plot_2Dscatter(data = X_transformed, labels=color, path_to_save= None, title = None, show = True)
#
# count_pairings(n_samples, pairs_filtered)
# make_plot(landmarks, pairs_filtered, color, name=name)
n_samples_array = [32,48,64,96,128]
n_witnesses_array = [256,512,1024]
seeds = [10,13,20]
n_samples_array = [64]
n_witnesses_array = [512]
seeds = [27]
for n_witnesses in n_witnesses_array:
for seed in seeds:
for n_samples in n_samples_array:
name = 'witness_ssc_corrected_nl{}_nw{}_seed{}'.format(n_samples, n_witnesses, seed)
dataset_sampler = SwissRoll()
n_landmarks = n_samples
seed = seed
landmarks, color = dataset_sampler.sample(n_landmarks, seed = seed)
witnesses, _ = dataset_sampler.sample(n_witnesses, seed=(seed+17))
distances = wl_table(witnesses,landmarks)
pairs = get_pairs_0(distances)
pairs_filtered = get_persistence_pairs(pairs, n_samples)
count_pairings(n_samples, pairs_filtered)
make_plot(landmarks, pairs_filtered, color, name=name)
# 'WCTopoAE128': wctopoae_128,
#
# }
# eval_models_dict = {
# 'TopoAE64' : topoae_64
# }
# set metrices to evaluate
#metrics = ['K_min', 'K_max','K_avg','llrme','continuity','trustworthiness']
#metrics = ['Trust', 'Cont', 'LGD', 'K_min', 'K_max', 'K_avg']
metrics = ['RRE', 'Trust', 'Cont', 'IsoX', 'IsoZ', 'IsoXlist', 'IsoZlist']
# sample data
n_samples = 2560
dataset = SwissRoll()
data, labels = dataset.sample(n_samples=n_samples, seed=1)
for model_name, path in eval_models_dict.items():
# load WC-AE
print('START: {}'.format(model_name))
model_kwargs = dict(input_dim=3,
latent_dim=2,
size_hidden_layers=[32, 32])
autoencoder = Autoencoder_MLP_topoae(**model_kwargs)
model = WitnessComplexAutoencoder(autoencoder)
state_dict = torch.load(os.path.join(path, 'model_state.pth'))
model.load_state_dict(state_dict)
model.eval()
dataset_test = TensorDataset(torch.Tensor(data), torch.Tensor(labels))
dataloader_eval = torch.utils.data.DataLoader(dataset_test,
import torch
from sklearn.metrics import pairwise_distances
from src.datasets.datasets import SwissRoll
from src.topology.witness_complex import WitnessComplex
if __name__ == "__main__":
N_WITNESSES = 512
N_LANDMARKS = 32
landmark_dist = torch.ones(N_LANDMARKS, N_LANDMARKS) * 1000000
dataset = SwissRoll()
X_witnesses, _ = dataset.sample(n_samples=N_WITNESSES)
ind_l = random.sample(range(N_WITNESSES), N_LANDMARKS)
X_landmarks = X_witnesses[ind_l, :]
witness_complex1 = WitnessComplex(X_landmarks, X_witnesses)
for n_jobs in [1]:
witness_complex1 = WitnessComplex(X_landmarks, X_witnesses)
start = time.time()
witness_complex1.compute_metric_optimized(n_jobs=-1)
end = time.time()
print('{} jobs --- Time needed: {}'.format(n_jobs, end - start))
witness_complex2 = WitnessComplex(X_landmarks, X_witnesses)
start = time.time()
if len(element[0]) == 2:
if element[1] == 0:
pairs.append(element[0])
print('Skeleton W-LIST: {}'.format(skeleton))
print('Pairs W-LIST: {}'.format(pairs))
dataset_sampler = SwissRoll()
n_landmarks = (64+64)
n_total = 4048*2*2*2*2
seed = 9
#landmarks_, color_ = dataset_sampler.sample(n_landmarks, seed = 9)
witnesses_, color_ = dataset_sampler.sample(n_total, seed = seed)
landmark_indices = random.sample(list(range(0,n_total)), n_landmarks)
#landmarks_ = gudhi.pick_n_random_points(points=witnesses_, nb_points=n_landmarks)
landmarks, color_landmarks = witnesses_[landmark_indices,:], color_[landmark_indices]
witness_complex = gudhi.EuclideanWitnessComplex(witnesses=witnesses_, landmarks=landmarks)
simplex_tree = witness_complex.create_simplex_tree(max_alpha_square = 100,
limit_dimension=1)
skeleton = simplex_tree.get_skeleton(1)
pairs = []
for element in skeleton:
if len(element[0]) == 2:
from src.datasets.datasets import SwissRoll
from src.evaluation.eval import Multi_Evaluation
dataset_sampler = SwissRoll()
data, label = dataset_sampler.sample(1000, seed=1)
data2, label = dataset_sampler.sample(1000, seed=2)
evaluator = Multi_Evaluation(seed=1)
ev_result = evaluator.get_multi_evals(
data, data, label, ks=[5,10,15])
print(ev_result)
data_obs[:, 1],
data_obs[:, 2],
c='tab:cyan')
plt_out = ax.scatter(X_out, Y_out, Z_out, c='tab:orange')
ax.set_title(
str(4 * N_obs) + '-sampling of the necklace with ' + str(N_out) +
' outliers')
ax.legend((plt_obs, plt_out), ('data', 'outliers'), loc='lower left')
return data
if __name__ == "__main__":
dataset_sampler = SwissRoll()
name = '128_noise4_DTM'
data, color = dataset_sampler.sample(128, noise=0.4)
st = DTMFiltration(data, m=0.01, p=10, dimension_max=1)
st.persistence()
pers_pairs = st.persistence_pairs()
print(pers_pairs)
pairings = np.array([[pers_pairs[0][1][0], pers_pairs[0][1][1]]])
for pair in pers_pairs[1:-1]:
pairings = np.vstack((pairings, np.array([[pair[1][0], pair[1][1]]])))
make_plot(data, pairings, color, name=name)
# name = '128_noise4_reg'
# make_data(data, color, name = name)
#
# path_pairings = '{}pairings_{}.npy'.format(PATH_ROOT, name)
dataset_sampler = SwissRoll()
N_WITNESSES = 2048
n_samples = 128
path_to_save = '/Users/simons/PycharmProjects/MT-VAEs-TDA/output/visualisation_nnsys/wc{}_w{}/'.format(
n_samples, N_WITNESSES)
N_sim = 100
ks = [1]
ntot = int(len(ks) * N_sim)
counter = 1
for seed in [30]:
witnesses, color_ = dataset_sampler.sample((N_WITNESSES - 128),
seed=seed)
landmarks, color = dataset_sampler.sample(128, seed=30)
witnesses_all = np.vstack((witnesses, landmarks))
witnesses_tensor = torch.from_numpy(witnesses_all)
landmarks_tensor = torch.from_numpy(landmarks)
witness_complex = WitnessComplex(landmarks_tensor, witnesses_tensor)
# witness_complex.compute_simplicial_complex(d_max=1,
# r_max=10,
# create_simplex_tree=False,
# create_metric=True)
witness_complex.compute_metric_optimized(n_jobs=4)
for k in ks:
tSNE_path = '/Users/simons/MT_data/sync/euler_sync_scratch/schsimo/output/tsne_swissroll_2/SwissRoll-n_samples2560-tSNE--n_jobs1-perplexity50-seed672-017a5cba/train_latents.npz'
eval_models_dict = {'UMAP': UMAP_path, 'tSNE': tSNE_path}
eval_seeds = {'UMAP': UMAP_seed, 'tSNE': tSNE_see}
metrics = ['RRE', 'Trust', 'Cont', 'IsoX', 'IsoZ', 'IsoXlist', 'IsoZlist']
# sample data
for model_name, path in eval_models_dict.items():
# load WC-AE
print('START: {}'.format(model_name))
n_samples = 2560
dataset = SwissRoll()
X_eval, labels = dataset.sample(n_samples=n_samples,
seed=eval_seeds[model_name])
X_eval, X_val, y_train, y_val, = train_test_split(
X_eval, X_eval, test_size=0.2, random_state=eval_seeds[model_name])
Z_eval = np.load(path)['latents']
# evaluate for multiple ks, what? -> Cont, Trust, ll-RMSE, K
ks = [15, 30, 45]
#ks = [int(k) for k in np.linspace(15,150,10)]
# eval = Multi_Evaluation(model=model)
# ev_result = eval.get_multi_evals(data=X_eval, latent=Z_eval, labels=Y_eval, ks=ks)
ev_result = CompPerformMetrics(X_eval, Z_eval, ks=ks, dataset='norm')
print('Done')
# collect results and save in df.
n_w = 512
#path = '/Users/simons/PycharmProjects/MT-VAEs-TDA/output/TDA/DoubleAnnulus'
dataset = SwissRoll()
small = [0.5,0.8]
large = [1, 1.3]
area_l = math.pi*(large[1]**2-large[0]**2)
area_s = math.pi*(small[1]**2-small[0]**2)
sample_ratio = area_l/area_s
for seed in [22]:
for n_l in [64,128]:
X_w, w_ = dataset.sample(n_w, seed=seed)
ind = random.sample(range(n_w), n_l)
X_l, l_ = X_w[ind, :], w_[ind]
# n_l = int(n_l * (1+sample_ratio))
#
# df_an1 = annulus(int(sample_ratio*n_w), large[0], large[1], seed=seed)
# df_an2 = annulus(n_w, small[0], small[1], label=1, seed=(seed+12))
# df_an = df_an1.append(df_an2, ignore_index=True)
#
# X_w = df_an[['x','y']].to_numpy()
# ind = random.sample(range(n_w), n_l)
# X_l = X_w[ind, :]
witness_complex = WitnessComplex(X_l, X_w)
if __name__ == "__main__":
dataset_sampler = SwissRoll()
n_samples_array = [128]
tot_count = len(n_samples_array) * 100
progress_count = 1
for n_samples in n_samples_array:
path_to_save = '/Users/simons/PycharmProjects/MT-VAEs-TDA/output/visualisation_nnsys/final_pretty/vr/'.format(
n_samples)
for seed in [30]:
print('{} out of {}'.format(progress_count, tot_count))
progress_count += 1
name = 'vr_ns{}_seed{}'.format(n_samples, seed)
data, color = dataset_sampler.sample(n_samples, seed=seed)
data, pairings, color = make_data(data, color, name=name)
# path_pairings = '{}pairings_{}.npy'.format(PATH_ROOT_SWISSROLL, name)
# path_data = '{}data_{}.npy'.format(PATH_ROOT_SWISSROLL, name)
# path_color = '{}color_{}.npy'.format(PATH_ROOT_SWISSROLL, name)
# pairings, data, color = np.load(path_pairings), np.load(path_data), np.load(path_color)
#
make_plot(data,
pairings,
color,
name=name,
path_root=path_to_save,
cmap=plt.cm.viridis)
# name = '512_1'
from sklearn.manifold import TSNE
from src.datasets.datasets import SwissRoll
if __name__ == "__main__":
dataset = SwissRoll()
data, labels = dataset.sample(n_samples=100)
model = TSNE()
from src.datasets.datasets import SwissRoll
from src.topology.witness_complex import WitnessComplex
if __name__ == "__main__":
unique_id = str(uuid.uuid4())[:4]
df_timing = pd.DataFrame()
dataset = SwissRoll()
n_witnesses = [512, 1024, 2048]
#n_landmarks = [int(i) for i in np.logspace(5, 9, num=5, base=2.0)]
n_landmarks = [8, 16, 32, 64]
data, _ = dataset.sample(n_samples=max(n_witnesses))
df_timing_data = []
for n_w in n_witnesses:
ind_w = random.sample(range(max(n_witnesses)), n_w)
X_witnesses = data[ind_w, :]
for n_l in n_landmarks:
ind_l = random.sample(range(max(n_witnesses)), n_l)
X_landmarks = data[ind_l, :]
distance_matrix_not_computed = True
r_max = 15
while (distance_matrix_not_computed):
witness_complex1 = WitnessComplex(X_landmarks, X_witnesses)
start = time.time()
