mnist_test512_cuda = ConfigGrid_WCAE(
learning_rate=[1/100],
batch_size=[512],
n_epochs=[1],
weight_decay=[1e-6],
early_stopping=[50],
rec_loss_weight=[1],
top_loss_weight=[1,2],
match_edges=['push_active'],
k=[1],
r_max=[10],
model_class=[ConvAE_MNIST_SMALL],
model_kwargs=[dict()],
dataset=[MNIST_offline()],
sampling_kwargs=[dict(root_path = '/content/gdrive/My Drive/MT_projectfolder/MT/AEs-VAEs-TDA')],
eval=[ConfigEval(
active=False,
evaluate_on='test',
eval_manifold=False,
save_eval_latent=True,
save_train_latent=True,
online_visualization=False,
k_min=5,
k_max=45,
k_step=5,
)],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(n_jobs=[1], normalize=[True], mu_push=[1], online_wc=[True], wc_offline = [dict(path_to_data = '/content/gdrive/My Drive/MT_projectfolder/MT/AEs-VAEs-TDA/src/datasets/WitnessComplexes/mnist/MNIST_offline-bs512-seed838-noiseNone-ced06774')]),
experiment_dir='/output/WAE/mnist_precomputed_2',
seed=838,
device='cuda',
num_threads=1,
verbose=True,
)
unity_test = ConfigGrid_WCAE(
learning_rate=[1 / 100],
batch_size=[120],
n_epochs=[200],
weight_decay=[1e-6],
early_stopping=[20],
rec_loss_weight=[1],
top_loss_weight=[512, 1024],
match_edges=['push_active'],
k=[1],
r_max=[10],
model_class=[ConvAElarge_Unity],
model_kwargs=[dict()],
dataset=[Unity_Rotblock()],
sampling_kwargs=[dict()],
eval=[
ConfigEval(active=True,
evaluate_on='test',
eval_manifold=False,
save_eval_latent=True,
save_train_latent=True,
online_visualization=False,
k_min=5,
k_max=10,
k_step=5,
quant_eval=False)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(
n_jobs=[1],
normalize=[True],
mu_push=[1],
online_wc=[True],
dist_x_land=[True],
wc_offline=[
dict(
path_to_data=
'/Users/simons/PycharmProjects/MT-VAEs-TDA/src/datasets/simulated/block_rotation_1'
)
]),
experiment_dir=
'/Users/simons/PycharmProjects/MT-VAEs-TDA/output/WAE/testing_unity',
seed=1,
device='cpu',
num_threads=4,
verbose=True,
)
mnist_s838_1024_lw = ConfigGrid_WCAE(
learning_rate=[1 / 10, 1 / 100, 1 / 1000],
batch_size=[1024],
n_epochs=[1000],
weight_decay=[1e-6],
early_stopping=[30],
rec_loss_weight=[1],
top_loss_weight=[i for i in np.logspace(-2, -1, num=2, base=2.0)],
match_edges=['push_active'],
k=[1, 2, 3, 4, 5, 6],
r_max=[10],
model_class=[ConvAE_MNIST],
model_kwargs=[dict()],
dataset=[MNIST_offline()],
sampling_kwargs=[dict(root_path='/cluster/home/schsimo/MT/AEs-VAEs-TDA')],
eval=[
ConfigEval(
active=False,
evaluate_on='test',
eval_manifold=False,
save_eval_latent=True,
save_train_latent=False,
online_visualization=False,
k_min=5,
k_max=45,
k_step=5,
)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(n_jobs=[1],
normalize=[True],
mu_push=[1, 1.125],
online_wc=[True],
wc_offline=[dict(path_to_data=wcpath_mnist_s838_1024)]),
experiment_dir='/cluster/scratch/schsimo/output/mnist1024_2',
seed=838,
device='cuda',
num_threads=2,
verbose=False,
)
swissroll_visualize128 = ConfigGrid_WCAE(
learning_rate=[1 / 100],
batch_size=[128],
n_epochs=[1000],
weight_decay=[0],
early_stopping=[50],
rec_loss_weight=[1],
top_loss_weight=[8192],
match_edges=['push_active'],
k=[3],
r_max=[10],
model_class=[Autoencoder_MLP_topoae],
model_kwargs={
'input_dim': [3],
'latent_dim': [2],
'size_hidden_layers': [[32, 32]]
},
dataset=[SwissRoll()],
sampling_kwargs={
'n_samples': [2560] #2560
},
eval=[
ConfigEval(
active=False,
evaluate_on='test',
eval_manifold=False,
save_eval_latent=True,
save_train_latent=True,
online_visualization=True,
quant_eval=False,
k_min=5,
k_max=45,
k_step=5,
)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(
n_jobs=[1],
normalize=[True],
mu_push=[1.25],
online_wc=[True],
wc_offline=[
dict(
path_to_data=
'/Users/simons/MT_data/sync/euler_sync/schsimo/MT/output/WitnessComplexes/SwissRoll/nonoise/SwissRoll-bs128-seed5310-d39df50c'
)
]),
experiment_dir=
'/Users/simons/PycharmProjects/MT-VAEs-TDA/output/WAE/online_visualize2',
seed=5310,
device='cpu',
num_threads=1,
verbose=True,
)
rotopenai_1_local = ConfigGrid_WCAE(
learning_rate=[1 / 100],
batch_size=[180],
n_epochs=[1000],
weight_decay=[0],
early_stopping=[120],
rec_loss_weight=[1],
top_loss_weight=[1],
match_edges=['push_active'],
k=[2],
r_max=[10],
model_class=[ConvAE_Unity480320],
model_kwargs=[dict()],
dataset=[Unity_RotOpenAI()],
sampling_kwargs=[dict()],
eval=[
ConfigEval(active=True,
evaluate_on='test',
eval_manifold=False,
save_eval_latent=False,
save_train_latent=True,
online_visualization=False,
k_min=5,
k_max=10,
k_step=5,
quant_eval=False)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(
n_jobs=[1],
normalize=[True],
mu_push=[1.125],
online_wc=[True],
dist_x_land=[True],
lam_t_decay=[
dict([(i * 100, 1 / 2**ii)
for i, ii in enumerate([-2, -1, 0, 1, 2, 4, 8, 10, 12])])
],
wc_offline=[
dict(path_to_data='/src/datasets/simulated/openai_rotating')
],
pre_trained_model=[
'/Users/simons/PycharmProjects/MT-VAEs-TDA/output/WAE/openai/retrain_examples/1_/post/Unity_RotOpenAI-seed1-ConvAE_Unity480320-default-lr1_100-bs180-nep1000-rlw1-tlw1-mepush_active9_8-k3-rmax10-seed1-a31416d4'
]),
experiment_dir='/output/WAE/openai/retrain_examples/1_',
seed=1,
device='cpu',
num_threads=1,
verbose=True,
)
ConfigGrid_WCAE(
learning_rate=[1 / 10, 1 / 100, 1 / 1000],
batch_size=[int(bs)],
n_epochs=[1000],
weight_decay=[1e-6],
early_stopping=[50],
rec_loss_weight=[1],
top_loss_weight=[int(i) for i in np.logspace(9, 13, num=5, base=2.0)],
match_edges=['push_active'],
k=[1, 2, 3, 4, 5, 6],
r_max=[10],
model_class=[Autoencoder_MLP_topoae],
model_kwargs={
'input_dim': [3],
'latent_dim': [2],
'size_hidden_layers': [[32, 32]]
},
dataset=[SwissRoll()],
sampling_kwargs={'n_samples': [2560]},
eval=[
ConfigEval(
active=True,
evaluate_on='test',
eval_manifold=True,
save_eval_latent=True,
save_train_latent=True,
online_visualization=False,
k_min=15,
k_max=45,
k_step=15,
)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(n_jobs=[1],
normalize=[True],
mu_push=[1, 1.05, 1.1, 1.15, 1.2, 1.25],
online_wc=[True],
wc_offline=[dict_wc]),
experiment_dir=
'/cluster/scratch/schsimo/output/WCAE_swissroll_nonoise_FINAL',
seed=int(seed),
device='cpu',
num_threads=1,
verbose=False,
)
ConfigGrid_WCAE(
learning_rate=[lr],
batch_size=random.sample(
[int(i) for i in np.logspace(6, 9, num=4, base=2.0)], 4),
n_epochs=[1000],
weight_decay=[1e-6],
early_stopping=[32],
rec_loss_weight=[1],
top_loss_weight=[int(i) for i in np.logspace(9, 13, num=5, base=2.0)],
match_edges=['symmetric'],
k=[1, 2, 4, 8, 16],
r_max=[10],
model_class=[Autoencoder_MLP_topoae],
model_kwargs={
'input_dim': [3],
'latent_dim': [2],
'size_hidden_layers': [[32, 32]]
},
dataset=[SwissRoll()],
sampling_kwargs={'n_samples': [2560]},
eval=[
ConfigEval(
active=True,
evaluate_on='test',
save_eval_latent=True,
save_train_latent=True,
online_visualization=False,
k_min=5,
k_max=20,
k_step=5,
)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=[dict(n_jobs=1, normalize=True)],
experiment_dir=
'/cluster/scratch/schsimo/output/WCTopoAE_swissroll_symmetric',
seed=seed,
device='cpu',
num_threads=1,
verbose=False,
) for lr, seed in zip(list(np.repeat([1 / 10, 1 / 100, 1 / 1000], 5)),
ConfigGrid_WCAE(
learning_rate=[1 / 1000],
batch_size=[int(i) for i in np.logspace(3, 11, num=10, base=2.0)],
n_epochs=[1000],
weight_decay=[0],
early_stopping=[10],
rec_loss_weight=[1],
top_loss_weight=[j],
match_edges=['symmetric'],
k=[1],
r_max=[10],
model_class=[Autoencoder_MLP_topoae],
model_kwargs={
'input_dim': [101],
'latent_dim': [2],
'size_hidden_layers': [[32, 32]]
},
dataset=[Spheres()],
sampling_kwargs={'n_samples': [512]},
eval=[
ConfigEval(
active=True,
evaluate_on='test',
save_eval_latent=True,
save_train_latent=True,
online_visualization=False,
k_min=5,
k_max=80,
k_step=25,
)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=[dict()],
experiment_dir=
'/cluster/home/schsimo/MT/output/WCTopoAE/Spheres/testing',
seed=1,
device='cpu',
num_threads=1,
verbose=False,
) for j in [int(i) for i in np.logspace(0, 10, base=2, num=4)]
rotopenai_cluster = ConfigGrid_WCAE(
learning_rate=[1 / 10, 1 / 1000, 1 / 100],
batch_size=[180],
n_epochs=[1000],
weight_decay=[1e-6],
early_stopping=[50],
rec_loss_weight=[1],
top_loss_weight=[1],
match_edges=['push_active'],
k=[1, 2, 3, 4, 5],
r_max=[10],
model_class=[ConvAE_Unity480320],
model_kwargs=[dict()],
dataset=[Unity_RotOpenAI()],
sampling_kwargs=[dict(root_path='/cluster/home/schsimo/MT/AEs-VAEs-TDA')],
eval=[
ConfigEval(active=True,
evaluate_on='test',
eval_manifold=False,
save_eval_latent=False,
save_train_latent=True,
online_visualization=False,
k_min=5,
k_max=10,
k_step=5,
quant_eval=False)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(
n_jobs=[1],
normalize=[True],
mu_push=[1, 1.125, 1.25],
online_wc=[True],
dist_x_land=[True],
wc_offline=[
dict(
path_to_data=
'/cluster/home/schsimo/MT/AEs-VAEs-TDA/src/datasets/simulated/openai_rotating'
)
]),
experiment_dir='/cluster/scratch/schsimo/output/openai_rot',
seed=1,
device='cpu',
num_threads=1,
verbose=True,
)
corgi_1 = ConfigGrid_WCAE(
learning_rate=[1/10,1/100],
batch_size=[60],
n_epochs=[500],
weight_decay=[1e-6],
early_stopping=[30],
rec_loss_weight=[1],
top_loss_weight=[64,128],
match_edges=['push_active'],
k=[2],
r_max=[10],
model_class=[ConvAE_Unity480320],
model_kwargs=[dict()],
dataset=[Unity_RotCorgi()],
sampling_kwargs=[dict()],
eval=[ConfigEval(
active=True,
evaluate_on='test',
eval_manifold=False,
save_eval_latent=True,
save_train_latent=True,
online_visualization=False,
k_min=5,
k_max=10,
k_step=5,
quant_eval = False
)],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(n_jobs=[1], normalize=[True], mu_push=[1], online_wc=[True], dist_x_land = [True],
wc_offline=[dict(path_to_data='/Users/simons/PycharmProjects/MT-VAEs-TDA/src/datasets/simulated/corgi_rotation_1')]),
experiment_dir='/Users/simons/PycharmProjects/MT-VAEs-TDA/output/WAE/corgi/rotating',
seed=1,
device='cpu',
num_threads=4,
verbose=True,
)
rotopenai_test = ConfigGrid_WCAE(
learning_rate=[1 / 100],
batch_size=[180],
n_epochs=[1000],
weight_decay=[1e-6],
early_stopping=[30],
rec_loss_weight=[1],
top_loss_weight=[1 / 8, 1 / 4, 1 / 2, 1, 0],
match_edges=['push_active'],
k=[1],
r_max=[10],
model_class=[ConvAE_Unity480320],
model_kwargs=[dict()],
dataset=[Unity_RotOpenAI()],
sampling_kwargs=[dict()],
eval=[
ConfigEval(active=True,
evaluate_on='test',
eval_manifold=False,
save_eval_latent=True,
save_train_latent=True,
online_visualization=False,
k_min=5,
k_max=10,
k_step=5,
quant_eval=False)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(
n_jobs=[1],
normalize=[True],
mu_push=[1],
online_wc=[True],
dist_x_land=[True],
wc_offline=[
dict(path_to_data='/src/datasets/simulated/openai_rotating')
]),
experiment_dir='/output/WAE/openai/rotating',
seed=1,
device='cpu',
num_threads=1,
verbose=True,
)
corgi_30_decay = ConfigGrid_WCAE(
learning_rate=[1 / 100, 1 / 1000],
batch_size=[30],
n_epochs=[5000],
weight_decay=[1e-6, 1e-8, 0],
early_stopping=[50],
rec_loss_weight=[1],
top_loss_weight=[1024],
match_edges=['push_active'],
k=[2],
r_max=[10],
model_class=[ConvAE_Unity480320],
model_kwargs=[dict()],
dataset=[Unity_RotCorgi()],
sampling_kwargs=[
dict(root_path='/cluster/home/schsimo/MT/AEs-VAEs-TDA',
version=5,
landmarks=True)
],
eval=[
ConfigEval(active=True,
evaluate_on='test',
eval_manifold=False,
save_eval_latent=True,
save_train_latent=True,
online_visualization=False,
k_min=5,
k_max=10,
k_step=5,
quant_eval=False)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(
n_jobs=[1],
normalize=[True],
mu_push=[1],
online_wc=[True],
dist_x_land=[True],
lam_t_decay=[{
0: 1024,
25: 512,
50: 256,
75: 128,
100: 64,
125: 32,
150: 16,
150: 8,
500: 4,
1000: 2
}],
wc_offline=[
dict(
path_to_data=
'/cluster/home/schsimo/MT/AEs-VAEs-TDA/src/datasets/simulated/corgi_rotation_5_l'
)
]),
experiment_dir='/cluster/scratch/schsimo/output/corgi/corgi_30_std',
seed=1,
device='cpu',
num_threads=2,
verbose=False,
)
mnist_s838_64_fullk_fullnu = ConfigGrid_WCAE(
learning_rate=[1 / 10, 1 / 100, 1 / 1000],
batch_size=[64],
n_epochs=[1000],
weight_decay=[1e-6],
early_stopping=[30],
rec_loss_weight=[1],
top_loss_weight=[int(i) for i in np.logspace(0, 4, num=5, base=2.0)],
match_edges=['push_active'],
k=[1, 2, 3, 4, 5, 6],
r_max=[10],
model_class=[ConvAE_MNIST],
model_kwargs=[dict()],
dataset=[MNIST_offline()],
sampling_kwargs=[dict(root_path='/cluster/home/schsimo/MT/AEs-VAEs-TDA')],
eval=[
ConfigEval(
active=False,
evaluate_on='test',
eval_manifold=False,
save_eval_latent=True,
save_train_latent=False,
online_visualization=False,
k_min=5,
k_max=45,
k_step=5,
)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(n_jobs=[1],
normalize=[True],
mu_push=[1, 1.125, 1.25, 1.375],
online_wc=[True],
wc_offline=[dict(path_to_data=wcpath_mnist_s838_64)]),
experiment_dir='/cluster/scratch/schsimo/output/mnist64',
seed=838,
device='cpu',
num_threads=1,
verbose=False)
mnist_test_3d = ConfigGrid_WCAE(
learning_rate=[1 / 100],
batch_size=[1024],
n_epochs=[3],
weight_decay=[1e-6],
early_stopping=[50],
rec_loss_weight=[1],
top_loss_weight=[1],
match_edges=['push_active'],
k=[1],
r_max=[10],
model_class=[ConvAE_MNIST_3D],
model_kwargs=[dict()],
dataset=[MNIST_offline()],
sampling_kwargs=[dict()],
eval=[
ConfigEval(
active=False,
evaluate_on='test',
eval_manifold=False,
save_eval_latent=False,
save_train_latent=True,
online_visualization=False,
k_min=4,
k_max=5,
k_step=1,
)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(
n_jobs=[1],
normalize=[True],
mu_push=[1.05],
online_wc=[True],
wc_offline=[
dict(
path_to_data=
'/Users/simons/MT_data/sync/euler_sync/schsimo/MT/output/WitnessComplexes/mnist/MNIST_offline-bs1024-seed838-noiseNone-6f31dea2'
)
]),
experiment_dir=
'/Users/simons/PycharmProjects/MT-VAEs-TDA/scripts/ssc/output',
seed=838,
device='cpu',
num_threads=1,
verbose=False,
)
WCAE_sample_config = ConfigGrid_WCAE(
learning_rate=[1 / 1000],
batch_size=[64],
n_epochs=[5],
weight_decay=[0],
early_stopping=[35],
rec_loss_weight=[1],
top_loss_weight=[1024],
match_edges=['push_active'],
k=[1],
r_max=[10],
model_class=[Autoencoder_MLP_topoae],
model_kwargs={
'input_dim': [3],
'latent_dim': [2],
'size_hidden_layers': [[32, 32]]
},
dataset=[SwissRoll()],
sampling_kwargs={
'n_samples': [2560] #2560
},
eval=[
ConfigEval(
active=True,
evaluate_on='test',
save_eval_latent=True,
save_train_latent=True,
online_visualization=False,
k_min=5,
k_max=80,
k_step=25,
)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=[dict(n_jobs=1, normalize=True)],
experiment_dir='output/sample',
seed=1,
device='cpu',
num_threads=1,
verbose=True,
)
mnist_test = ConfigGrid_WCAE(
learning_rate=[1 / 100],
batch_size=[1024],
n_epochs=[3],
weight_decay=[1e-6],
early_stopping=[50],
rec_loss_weight=[1],
top_loss_weight=[1],
match_edges=['push_active'],
k=[1],
r_max=[10],
model_class=[ConvAE_MNIST],
model_kwargs=[dict()],
dataset=[MNIST_offline()],
sampling_kwargs=[dict()],
eval=[
ConfigEval(
active=False,
evaluate_on='test',
eval_manifold=False,
save_eval_latent=True,
save_train_latent=True,
online_visualization=False,
k_min=4,
k_max=5,
k_step=1,
)
],
uid=[''],
toposig_kwargs=[dict()],
method_args=dict(
n_jobs=[1],
normalize=[True],
mu_push=[1.05],
online_wc=[True],
wc_offline=[
dict(
path_to_data=
'/Users/simons/MT_data/sync/euler_sync/schsimo/MT/output/WitnessComplexes/mnist/MNIST_offline-bs1024-seed838-noiseNone-6f31dea2'
)
]),
experiment_dir='/output/WAE/mnist_precomputed',
seed=838,
device='cpu',
num_threads=1,
verbose=True,
)