Skip to content

Fix exponent sign in distribution calculations and update training pa… #89

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
DoronHav merged 1 commit into from
Oct 9, 2025
Merged

Fix exponent sign in distribution calculations and update training pa… #89

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 8 additions & 8 deletions scenvi/ENVI.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -248,10 +248,10 @@ def mean_sc(self, sc_inp):
sc_inp = sc_inp[:, : self.dist_size_dict[self.sc_dist] * self.sc_data.shape[-1]]
if self.sc_dist == "zinb":
sc_r, sc_p, sc_d = jnp.split(sc_inp, 3, axis=-1)
return nn.softplus(sc_r) * jnp.exp(-sc_p) * (1 - nn.sigmoid(sc_d))
return nn.softplus(sc_r) * jnp.exp(sc_p) * (1 - nn.sigmoid(sc_d))
if self.sc_dist == "nb":
sc_r, sc_p = jnp.split(sc_inp, 2, axis=-1)
return nn.softplus(sc_r) * jnp.exp(-sc_p)
return nn.softplus(sc_r) * jnp.exp(sc_p)
if self.sc_dist == "pois":
sc_l = sc_inp
return sc_l
Expand Down Expand Up @@ -371,7 +371,7 @@ def grammian_cov(self, dec_cov):
dec_cov = jax_prob.math.fill_triangular(dec_cov)
return jnp.matmul(dec_cov, dec_cov.transpose([0, 2, 1]))

def create_train_state(self, key=random.key(0), init_lr=0.0001, decay_steps=4000):
def create_train_state(self, key=random.key(0), init_lr=3e-4, decay_steps=100):
"""
:meta private:
"""
Expand All @@ -384,7 +384,7 @@ def create_train_state(self, key=random.key(0), init_lr=0.0001, decay_steps=4000
key=subkey2,
)["params"]

lr_sched = optax.exponential_decay(init_lr, decay_steps, 0.5, staircase=True)
lr_sched = optax.exponential_decay(init_lr, decay_steps, 0.99, staircase=False)
tx = optax.adam(lr_sched) #

return TrainState.create(
Expand Down Expand Up @@ -443,15 +443,15 @@ def loss_fn(params):

def train(
self,
training_steps=16000,
training_steps=10000,
batch_size=128,
verbose=16,
init_lr=0.0001,
decay_steps=4000,
init_lr=3e-4,
decay_steps=100,
key=random.key(0),
):
"""
Set up optimization parameters and train the ENVI moodel
Set up optimization parameters and train the ENVI model


:param training_steps: (int) number of gradient descent steps to train ENVI (default 16000)
Expand Down
Loading