Skip to content

Add pose space visualisation #243

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
jolaem merged 5 commits into from
Oct 30, 2023
Merged

Add pose space visualisation #243

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
9161a7f
Add pose space visualisation
jolaem Oct 27, 2023
4daf358
Make sure pose also visualised when no tests present
jolaem Oct 27, 2023
78d3ce5
Add legend to pose embedding visualisation
jolaem Oct 27, 2023
cb476f5
Make sure xs and ys are np arrays
jolaem Oct 27, 2023
63b81ac
Update number of plots for tests
jolaem Oct 27, 2023
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
16 changes: 13 additions & 3 deletions avae/train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -572,15 +572,25 @@ def train(
y_val,
np.full(shape=len(x_test), fill_value="test"),
]
if pose:
ps = np.r_[p_train, p_val, p_test]
else:
xs = np.r_[x_train, x_val]
ys = np.r_[y_train, y_val]
vis.latent_embed_plot_tsne(
xs, ys, classes_list, epoch=epoch, writer=writer
)
if pose:
ps = np.r_[p_train, p_val]

vis.latent_embed_plot_tsne(xs, ys, epoch=epoch, writer=writer)
vis.latent_embed_plot_umap(
xs, ys, classes_list, epoch=epoch, writer=writer
)
if pose:
vis.latent_embed_plot_tsne(
ps, ys, epoch=epoch, writer=writer, mode="pose"
)
vis.latent_embed_plot_umap(
ps, ys, epoch=epoch, writer=writer, mode="pose"
)

if config.VIS_DYN:
# merge img and rec into one image for display in altair
Expand Down
170 changes: 133 additions & 37 deletions avae/vis.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -171,7 +171,10 @@ def latent_embed_plot_tsne(
logging.info(
"################################################################",
)
logging.info("Visualising static TSNE embedding...\n")
if not mode:
logging.info("Visualising static TSNE embedding...\n")
else:
logging.info("Visualising static TSNE embedding " + mode + "...\n")

xs = np.asarray(xs)
ys = np.asarray(ys)
Expand All @@ -180,9 +183,27 @@ def latent_embed_plot_tsne(
if len(ys) < perplexity:
perplexity = len(ys) - 1

lats = TSNE(
n_components=2, perplexity=perplexity, random_state=42
).fit_transform(xs)
if len(xs.shape) != 2:
logging.error("Embedding only accepts 2D arrays.")
exit(1)
if xs.shape[-1] == 1:
logging.warning(
"Data contains 1 dimension, cannot create embedding,"
" plotting histogram instead...\n"
)
if xs.shape[-1] == 2:
logging.warning(
"Data already contains 2 dimensions, cannot create"
" embedding, plotting scatter of original data...\n"
)

if xs.shape[-1] > 2:
lats = TSNE(
n_components=2, perplexity=perplexity, random_state=42
).fit_transform(xs)
elif xs.shape[-1] == 2 or xs.shape[-1] == 1:
lats = xs

if classes is None:
classes = sorted(list(np.unique(ys)))
else:
Expand All @@ -203,27 +224,52 @@ def latent_embed_plot_tsne(
figsize=(int(n_classes / 2) + 4, int(n_classes / 2) + 2)
)
# When the number of classes is less than 3 the image becomes two small

colours = colour_per_class(classes)

for mol_id, mol in enumerate(set(ys.tolist())):
idx = np.where(np.array(ys.tolist()) == mol)[0]
if xs.shape[-1] != 1:

for mol_id, mol in enumerate(set(ys.tolist())):
idx = np.where(np.array(ys.tolist()) == mol)[0]

color = colours[classes.index(mol)]
color = colours[classes.index(mol)]

plt.scatter(
lats[idx, 0],
lats[idx, 1],
s=24,
label=mol[:4],
facecolor=color,
edgecolor=color,
alpha=0.2,
)

plt.scatter(
lats[idx, 0],
lats[idx, 1],
s=24,
label=mol[:4],
facecolor=color,
edgecolor=color,
alpha=0.2,
ax.legend(bbox_to_anchor=(1.05, 1), loc="upper left", fontsize=16)
plt.xlabel("TSNE-1")
plt.ylabel("TSNE-2")

if xs.shape[-1] == 1:

for mol_id, mol in enumerate(set(ys.tolist())):
idx = np.where(np.array(ys.tolist()) == mol)[0]
cols = colours[classes.index(mol)]
plt.hist(
lats[idx],
100,
color=cols,
histtype="step",
stacked=True,
fill=False,
label=mol[:4],
)
plt.legend(
prop={"size": 10},
bbox_to_anchor=(1.05, 1),
loc="upper left",
fontsize=16,
)
plt.xlabel("dim 1")
plt.ylabel("freq")

ax.legend(bbox_to_anchor=(1.05, 1), loc="upper left", fontsize=16)
plt.xlabel("TSNE-1")
plt.ylabel("TSNE-2")
plt.tight_layout()
plt.savefig(f"plots/embedding_TSNE{mode}.png")

Expand Down Expand Up @@ -255,9 +301,33 @@ def latent_embed_plot_umap(
"################################################################",
)

logging.info("Visualising static UMAP embedding...\n")
reducer = umap.UMAP(random_state=42)
embedding = reducer.fit_transform(xs)
if not mode:
logging.info("Visualising static UMAP embedding...\n")
else:
logging.info("Visualising static UMAP embedding " + mode + "...\n")

xs = np.asarray(xs)
ys = np.asarray(ys)

if len(xs.shape) != 2:
logging.error("Embedding only accepts 2D arrays.")
exit(1)
if xs.shape[-1] == 1:
logging.warning(
"Data contains 1 dimension, cannot create embedding,"
" plotting histogram instead...\n"
)
if xs.shape[-1] == 2:
logging.warning(
"Data already contains 2 dimensions, cannot create"
" embedding, plotting scatter of original data...\n"
)

if xs.shape[-1] > 2:
reducer = umap.UMAP(random_state=42)
embedding = reducer.fit_transform(xs)
elif xs.shape[-1] == 2 or xs.shape[-1] == 1:
embedding = xs

if classes is None:
classes = sorted(list(np.unique(ys)))
Expand All @@ -278,23 +348,49 @@ def latent_embed_plot_umap(

colours = colour_per_class(classes)

for mol_id, mol in enumerate(set(ys.tolist())):
idx = np.where(np.array(ys.tolist()) == mol)[0]
color = colours[classes.index(mol)]

ax.scatter(
embedding[idx, 0],
embedding[idx, 1],
s=24,
label=mol[:4],
facecolor=color,
edgecolor=color,
alpha=0.2,
if xs.shape[-1] != 1:

for mol_id, mol in enumerate(set(ys.tolist())):
idx = np.where(np.array(ys.tolist()) == mol)[0]
color = colours[classes.index(mol)]

ax.scatter(
embedding[idx, 0],
embedding[idx, 1],
s=24,
label=mol[:4],
facecolor=color,
edgecolor=color,
alpha=0.2,
)

ax.legend(bbox_to_anchor=(1.05, 1), loc="upper left", fontsize=16)
plt.xlabel("UMAP-1")
plt.ylabel("UMAP-2")

if xs.shape[-1] == 1:

for mol_id, mol in enumerate(set(ys.tolist())):
idx = np.where(np.array(ys.tolist()) == mol)[0]
cols = colours[classes.index(mol)]
plt.hist(
embedding[idx],
100,
color=cols,
histtype="step",
stacked=True,
fill=False,
label=mol[:4],
)
plt.legend(
prop={"size": 10},
bbox_to_anchor=(1.05, 1),
loc="upper left",
fontsize=16,
)
plt.xlabel("dim 1")
plt.ylabel("freq")

ax.legend(bbox_to_anchor=(1.05, 1), loc="upper left", fontsize=16)
plt.xlabel("UMAP-1")
plt.ylabel("UMAP-2")
plt.tight_layout()
plt.savefig(f"plots/embedding_UMAP{mode}.png")

Expand Down
16 changes: 8 additions & 8 deletions tests/test_train_eval_pipeline.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,11 +106,11 @@ def test_model_a_mrc(self):
) = helper_train_eval(self.data)

self.assertEqual(n_dir_train, 4)
self.assertEqual(n_plots_train, 32)
self.assertEqual(n_plots_train, 34)
self.assertEqual(n_latent_train, 2)
self.assertEqual(n_states_train, 2)

self.assertEqual(n_plots_eval, 50)
self.assertEqual(n_plots_eval, 52)
self.assertEqual(n_latent_eval, 4)
self.assertEqual(n_states_eval, 3)

Expand All @@ -129,10 +129,10 @@ def test_model_b_mrc(self):
) = helper_train_eval(self.data)

self.assertEqual(n_dir_train, 4)
self.assertEqual(n_plots_train, 32)
self.assertEqual(n_plots_train, 34)
self.assertEqual(n_latent_train, 2)
self.assertEqual(n_states_train, 2)
self.assertEqual(n_plots_eval, 50)
self.assertEqual(n_plots_eval, 52)
self.assertEqual(n_latent_eval, 4)
self.assertEqual(n_states_eval, 3)

Expand All @@ -157,10 +157,10 @@ def test_model_a_npy(self):
) = helper_train_eval(self.data)

self.assertEqual(n_dir_train, 4)
self.assertEqual(n_plots_train, 30)
self.assertEqual(n_plots_train, 32)
self.assertEqual(n_latent_train, 2)
self.assertEqual(n_states_train, 2)
self.assertEqual(n_plots_eval, 47)
self.assertEqual(n_plots_eval, 49)
self.assertEqual(n_latent_eval, 4)
self.assertEqual(n_states_eval, 3)

Expand All @@ -185,10 +185,10 @@ def test_model_b_npy(self):
) = helper_train_eval(self.data)

self.assertEqual(n_dir_train, 4)
self.assertEqual(n_plots_train, 30)
self.assertEqual(n_plots_train, 32)
self.assertEqual(n_latent_train, 2)
self.assertEqual(n_states_train, 2)
self.assertEqual(n_plots_eval, 47)
self.assertEqual(n_plots_eval, 49)
self.assertEqual(n_latent_eval, 4)
self.assertEqual(n_states_eval, 3)

Expand Down