Skip to content

Remove ** from file names and remove ' from commands #2

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

Open
banditbandito wants to merge 2 commits into
base: master
Choose a base branch
from
Open

Remove ** from file names and remove ' from commands #2

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
c2b222a
Update README.md
banditbandito Sep 21, 2021
690d845
Update train_loops.py
banditbandito Sep 23, 2021
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
34 changes: 17 additions & 17 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,33 +22,33 @@ For more information on making a Weight & Biases account see [(creating a weight

## Running an experiment
To rerun the experiment from Figure 3, you can run:
- `tvae --name 'tvae_2d_mnist'`
- `tvae --name tvae_2d_mnist`

To rerun the experiments from Figure 4, you can run:
- `tvae --name 'tvae_Lpartial_mnist'`
- `tvae --name 'tvae_Lpartial_dsprites'`
- `tvae --name tvae_Lpartial_mnist`
- `tvae --name tvae_Lpartial_dsprites`

To rerun the experiments from Tables 1, you can run:
- `tvae --name 'tvae_Lhalf_mnist'`
- `tvae --name 'tvae_Lshort_mnist'`
- `tvae --name 'bubbles_mnist'`
- `tvae --name 'tvae_L0_mnist'`
- `tvae --name 'nontvae_mnist'`
- `tvae --name tvae_Lhalf_mnist`
- `tvae --name tvae_Lshort_mnist`
- `tvae --name bubbles_mnist`
- `tvae --name tvae_L0_mnist`
- `tvae --name nontvae_mnist`

To rerun the experiments from Tables 2, you can run:
- `tvae --name 'tvae_Lhalf_dsprites'`
- `tvae --name 'tvae_Lpartial_dsprites'`
- `tvae --name 'tvae_Lshort_dsprites'`
- `tvae --name 'bubbles_dsprites'`
- `tvae --name 'tvae_L0_dsprites'`
- `tvae --name 'nontvae_dsprites'`
- `tvae --name tvae_Lhalf_dsprites`
- `tvae --name tvae_Lpartial_dsprites`
- `tvae --name tvae_Lshort_dsprites`
- `tvae --name bubbles_dsprites`
- `tvae --name tvae_L0_dsprites`
- `tvae --name nontvae_dsprites`

To rerun the generalization experiment described in Section B.4 (resulting in Figures 1 and 6), you can run:
- `tvae --name 'tvae_Lpartial_mnist_generalization'`
- `tvae --name tvae_Lpartial_mnist_generalization`

To rerun the experiments from Figures 22 and 23 (training on complex combined transformations), you can run:
- `tvae --name 'tvae_Lpartial_perspective_mnist'`
- `tvae --name 'tvae_Lpartial_rotcolor_mnist'`
- `tvae --name tvae_Lpartial_perspective_mnist`
- `tvae --name tvae_Lpartial_rotcolor_mnist`

## Basics of the framework
- All models are built using the `TVAE` module (see `tvae/containers/tvae.py`) which requires a z-encoder, a u-encoder, a decoder, and a 'grouper'. The grouper module defines the topographic structure of the latent space through a `model` (equivalent to W in the paper), and a `padder` which defines the boundary conditions.
Expand Down
8 changes: 4 additions & 4 deletions tvae/utils/train_loops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@ def train_epoch(model, optimizer, train_loader, log, savepath, epoch, eval_batch
model.plot_decoder_weights(wandb_on=wandb_on)
model.plot_encoder_weights(wandb_on=wandb_on)

Plot_Covariance_Matrix(s**2.0, s**2.0, name='Covariance_S**2_batch', wandb_on=wandb_on)
Plot_Covariance_Matrix(s**2.0, s**2.0, name='Covariance_Sxx2_batch', wandb_on=wandb_on)

if plot_fullcaptrav:
model.plot_capsule_traversal(x_batched.detach(),
Expand Down Expand Up @@ -122,7 +122,7 @@ def eval_epoch(model, val_loader, log, savepath, epoch, n_is_samples=100,
all_labels = torch.cat(all_labels, 0)
if plot_cov:
Plot_Covariance_Matrix(all_s, all_s, name='Covariance_S_Full', wandb_on=wandb_on)
Plot_Covariance_Matrix(all_s**2.0, all_s**2.0, name='Covariance_S**2_Full', wandb_on=wandb_on)
Plot_Covariance_Matrix(all_s**2.0, all_s**2.0, name='Covariance_Sxx2_Full', wandb_on=wandb_on)
if plot_maxact:
Plot_MaxActImg(all_s, all_x, os.path.join(savepath, 'samples'), epoch, wandb_on=wandb_on)
if plot_class_selectivity:
Expand Down Expand Up @@ -184,7 +184,7 @@ def train_epoch_dsprites(model, optimizer, train_loader, log, savepath, epoch, e
model.plot_decoder_weights(wandb_on=wandb_on)
model.plot_encoder_weights(wandb_on=wandb_on)

Plot_Covariance_Matrix(s**2.0, s**2.0, name='Covariance_S**2_batch', wandb_on=wandb_on)
Plot_Covariance_Matrix(s**2.0, s**2.0, name='Covariance_Sxx2_batch', wandb_on=wandb_on)

if plot_fullcaptrav:
model.plot_capsule_traversal(x_batched.detach(),
Expand All @@ -205,4 +205,4 @@ def train_epoch_dsprites(model, optimizer, train_loader, log, savepath, epoch, e
for t in cap_offsets:
total_cap_corr[t] = np.corrcoef(cap_offsets[t], true_offsets[t])[0,1]

return total_loss, total_neg_logpx_z, total_kl, total_eq_loss, total_cap_corr, num_batches
return total_loss, total_neg_logpx_z, total_kl, total_eq_loss, total_cap_corr, num_batches