Skip to content

Extend ablation #81

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
fariedabuzaid merged 2 commits into from
Oct 8, 2025
Merged

Extend ablation #81

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
57be8c9
extend ablations
fariedabuzaid Oct 8, 2025
91c6eca
Set device cuda
fariedabuzaid Oct 8, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
114 changes: 114 additions & 0 deletions experiments/cifar/cifar_complementary.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,114 @@
---
__object__: src.explib.base.ExperimentCollection
name: fashion_ablation
experiments:
- &exp_rad_logN
__object__: src.explib.hyperopt.HyperoptExperiment
name: cfair_full_radial_logN
skip: true
device: cuda
skip: true
scheduler: &scheduler
__object__: ray.tune.schedulers.ASHAScheduler
max_t: 1000000
grace_period: 1000000
reduction_factor: 2
num_hyperopt_samples: &num_hyperopt_samples 1
gpus_per_trial: &gpus_per_trial 1
cpus_per_trial: &cpus_per_trial 1
tuner_params: &tuner_params
metric: val_loss
mode: min
trial_config:
logging:
images: false
"image_shape": [28, 28]
dataset: &dataset
class:
__class__: src.explib.datasets.Cifar10Split
params:
space_to_depth_factor: 4
dataloc: /home/faried/Projects/USFlows/data/cifar10

epochs: &epochs 200000
patience: &patience 1
batch_size: &batch_size
__eval__: tune.choice([32])
optim_cfg: &optim
optimizer:
__class__: src.usflows.sophia.SophiaG
params:
lr:
__eval__: 1e-3
weight_decay: 0.0

model_cfg:
type:
__class__: src.usflows.flows.USFlow
params:
soft_training:
__eval__: tune.choice([False])
training_noise_prior:
__object__: pyro.distributions.Uniform
low:
__eval__: 1e-20
high: 0.01
prior_scale: 1.0
coupling_blocks:
__eval__: tune.choice([10])
lu_transform: 1
householder: 0
conditioner_cls:
__class__: src.usflows.networks.ConvNet2D
conditioner_args:
c_in: 48
c_hidden:
__eval__: tune.choice([32])
num_layers:
__eval__: tune.choice([3])
padding: same
kernel_size: 3
rescale_hidden: 1
normalize_layers:
__eval__: tune.choice([True])
gating:
__eval__: tune.choice([True])
in_dims: [48, 8, 8]
affine_conjugation: true
nonlinearity:
__eval__: tune.choice([torch.nn.ReLU()])
base_distribution:
__object__: pyro.distributions.Normal
loc:
__eval__: torch.zeros([48, 8, 8]).to("cuda")
scale:
__eval__: torch.ones([48, 8, 8]).to("cuda")
- &exp_normal
__overwrites__: *exp_rad_logN
name: fashion_full_laplace
skip: false
trial_config:
optim_cfg:
params:
lr:
__eval__: 1e-5
model_cfg:
params:
lu_transform: 0
affine_conjugation: false
base_distribution:
__object__: src.usflows.distributions.RadialDistribution
device: cuda
p:
__eval__: float("1")
loc:
__eval__: torch.zeros([48, 8, 8]).to("cuda")
norm_distribution:
__exact__:
__object__: src.usflows.distributions.LogNormal
loc:
__eval__: torch.ones([1]).to("cuda") * 6
scale:
__eval__: torch.ones([1]).to("cuda") * .35
device: cuda

114 changes: 114 additions & 0 deletions experiments/fashion/fashion_complementary.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,114 @@
---
__object__: src.explib.base.ExperimentCollection
name: fashion_ablation
experiments:
- &exp_rad_logN
__object__: src.explib.hyperopt.HyperoptExperiment
name: cfair_full_radial_logN
skip: true
device: cuda
skip: true
scheduler: &scheduler
__object__: ray.tune.schedulers.ASHAScheduler
max_t: 1000000
grace_period: 1000000
reduction_factor: 2
num_hyperopt_samples: &num_hyperopt_samples 1
gpus_per_trial: &gpus_per_trial 1
cpus_per_trial: &cpus_per_trial 1
tuner_params: &tuner_params
metric: val_loss
mode: min
trial_config:
logging:
images: false
"image_shape": [28, 28]
dataset: &dataset
class:
__class__: src.explib.datasets.FashionMnistSplit
params:
space_to_depth_factor: 4
dataloc: /home/faried/Projects/USFlows/data/fashion

epochs: &epochs 200000
patience: &patience 1
batch_size: &batch_size
__eval__: tune.choice([32])
optim_cfg: &optim
optimizer:
__class__: src.usflows.sophia.SophiaG
params:
lr:
__eval__: 1e-3
weight_decay: 0.0

model_cfg:
type:
__class__: src.usflows.flows.USFlow
params:
soft_training:
__eval__: tune.choice([False])
training_noise_prior:
__object__: pyro.distributions.Uniform
low:
__eval__: 1e-20
high: 0.01
prior_scale: 1.0
coupling_blocks:
__eval__: tune.choice([10])
lu_transform: 1
householder: 0
conditioner_cls:
__class__: src.usflows.networks.ConvNet2D
conditioner_args:
c_in: 48
c_hidden:
__eval__: tune.choice([32])
num_layers:
__eval__: tune.choice([3])
padding: same
kernel_size: 3
rescale_hidden: 1
normalize_layers:
__eval__: tune.choice([True])
gating:
__eval__: tune.choice([True])
in_dims: [48, 8, 8]
affine_conjugation: true
nonlinearity:
__eval__: tune.choice([torch.nn.ReLU()])
base_distribution:
__object__: pyro.distributions.Normal
loc:
__eval__: torch.zeros([48, 8, 8]).to("cuda")
scale:
__eval__: torch.ones([48, 8, 8]).to("cuda")

- &exp_normal
__overwrites__: *exp_rad_logN
name: fashion_full_laplace
skip: false
trial_config:
optim_cfg:
params:
lr:
__eval__: 1e-5
model_cfg:
params:
lu_transform: 0
affine_conjugation: false
base_distribution:
__object__: src.usflows.distributions.RadialDistribution
device: cuda
p:
__eval__: float("1")
loc:
__eval__: torch.zeros([48, 8, 8]).to("cuda")
norm_distribution:
__object__: src.usflows.distributions.LogNormal
loc:
__eval__: torch.ones([1]).to("cuda") * 6
scale:
__eval__: torch.ones([1]).to("cuda") * .35
device: cuda

149 changes: 149 additions & 0 deletions experiments/fashion/fashionclasses_macow_complementary.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,149 @@
---
__object__: src.usflows.explib.base.ExperimentCollection
name: fashion_ablation_macow
experiments:
- &exp_rad_logN
__object__: src.usflows.explib.hyperopt.HyperoptExperiment
name: cfair_full_normal
skip: false
device: cuda
scheduler: &scheduler
__object__: ray.tune.schedulers.ASHAScheduler
max_t: 1000000
grace_period: 1000000
reduction_factor: 2
num_hyperopt_samples: &num_hyperopt_samples 1
gpus_per_trial: &gpus_per_trial 1
cpus_per_trial: &cpus_per_trial 1
tuner_params: &tuner_params
metric: val_loss
mode: min
trial_config:
logging:
images: false
"image_shape": [28, 28]
dataset: &dataset
class:
__class__: src.usflow.explib.datasets.FashionMnistSplit
params:
space_to_depth_factor: 4
dataloc: /home/faried/Projects/USFlows/data/fashion
label: 0
epochs: &epochs 200000
patience: &patience 1
batch_size: &batch_size
__eval__: tune.choice([32])
optim_cfg: &optim
optimizer:
__class__: src.usflows.sophia.SophiaG
params:
lr:
__eval__: 1e-3
weight_decay: 0.0

model_cfg:
type:
__class__: src.usflows.flows.USFlow
params:
soft_training:
__eval__: tune.choice([False])
training_noise_prior:
__object__: pyro.distributions.Uniform
low:
__eval__: 1e-20
high: 0.01
prior_scale: 1.0
coupling_blocks:
__eval__: tune.choice([10])
lu_transform: 1
householder: 0
conditioner_cls:
__class__: src.usflows.networks.ConvNet2D
conditioner_args:
c_in: 16
c_hidden:
__eval__: tune.choice([32])
num_layers:
__eval__: tune.choice([3])
padding: same
kernel_size: 3
rescale_hidden: 1
normalize_layers:
__eval__: tune.choice([True])
gating:
__eval__: tune.choice([True])
in_dims: [16, 7, 7]
affine_conjugation: true
nonlinearity:
__eval__: tune.choice([torch.nn.ReLU()])
base_distribution:
__object__: src.usflows.distributions.RadialDistribution
device: cuda
p:
__eval__: float("1")
loc:
__eval__: torch.zeros([16, 7, 7]).to("cuda")
norm_distribution:
__object__: src.usflows.distributions.GammaMM
concentration:
__eval__: torch.rand([20]).to("cuda") * 75
rate:
__eval__: torch.rand([20]).to("cuda")
mixture_weights:
__eval__: torch.ones([20]).to("cuda") / 20
device: cuda
- __overwrites__: *exp_rad_logN
name: fashion1_radial_logN
trial_config:
dataset:
params:
label: 1
- __overwrites__: *exp_rad_logN
name: fashion2_radial_logN
trial_config:
dataset:
params:
label: 2
- __overwrites__: *exp_rad_logN
name: fashion3_radial_logN
trial_config:
dataset:
params:
label: 3
- __overwrites__: *exp_rad_logN
name: fashion4_radial_logN
trial_config:
dataset:
params:
label: 4
- __overwrites__: *exp_rad_logN
name: fashion5_radial_logN
trial_config:
dataset:
params:
label: 5
- __overwrites__: *exp_rad_logN
name: fashion6_radial_logN
trial_config:
dataset:
params:
label: 6
- __overwrites__: *exp_rad_logN
name: fashion7_radial_logN
trial_config:
dataset:
params:
label: 7
- __overwrites__: *exp_rad_logN
name: fashion8_radial_logN
trial_config:
dataset:
params:
label: 8
- __overwrites__: *exp_rad_logN
name: fashion9_radial_logN
trial_config:
dataset:
params:
label: 9

Loading
Loading