Skip to content
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

format: minor format fixes for multi modality modules and example scripts #190

Merged
merged 1 commit into from
Feb 15, 2023
Merged

format: minor format fixes for multi modality modules and example scripts #190

Show file tree
Hide file tree
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
48 changes: 21 additions & 27 deletions dance/modules/multi_modality/match_modality/cmae.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,12 +4,12 @@

Reference
---------
Yang, Karren Dai, et al. "Multi-domain translation between single-cell imaging and sequencing data using autoencoders." Nature communications 12.1 (2021): 1-10.
Yang, Karren Dai, et al. "Multi-domain translation between single-cell imaging and sequencing data using autoencoders."
Nature communications 12.1 (2021): 1-10.

"""
import math
import os
import sys

import torch
import torch.nn as nn
Expand Down Expand Up @@ -159,9 +159,10 @@ def __init__(self, input_dim, params, shared_layer=False):
self.dec = nn.Sequential(*decoder_layers)

def forward(self, images):
# This is a reduced VAE implementation where we assume the outputs are multivariate Gaussian distribution with mean = hiddens and std_dev = all ones.
# This is a reduced VAE implementation where we assume the outputs are multivariate Gaussian distribution
# with mean = hiddens and std_dev = all ones.
hiddens = self.encode(images)
if self.training == True:
if self.training:
noise = Variable(torch.randn(hiddens.size()).cuda(hiddens.data.get_device()))
images_recon = self.decode(hiddens + noise)
else:
Expand Down Expand Up @@ -259,7 +260,8 @@ class CMAE(nn.Module):
Parameters
----------
hyperparameters : dictionary
A dictionary that contains arguments of CMAE. For details of parameters in parser args, please refer to link (parser help document).
A dictionary that contains arguments of CMAE. For details of parameters in parser args, please refer to link
(parser help document).

"""

Expand Down Expand Up @@ -319,7 +321,6 @@ def predict(self, mod1, mod2, metric='l1'):
Joint embedding of input modalities.

"""

with torch.no_grad():
emb1, _ = self.gen_a.encode(mod1)
emb2, _ = self.gen_b.encode(mod2)
Expand Down Expand Up @@ -348,7 +349,6 @@ def score(self, mod1, mod2, labels):
Matching accuracy.

"""

with torch.no_grad():
pred = self.predict(mod1, mod2)
return (pred[torch.arange(pred.shape[0]).long(), labels.long()].mean()).item()
Expand Down Expand Up @@ -419,17 +419,21 @@ def _gen_update(self, x_a, x_b, super_a, super_b, hyperparameters, a_labels=None
class_weight = hyperparameters['gan_w'] if "class_w" not in hyperparameters else hyperparameters["class_w"]

# total loss
self.loss_gen_total = hyperparameters['gan_w'] * self.loss_latent_a + \
hyperparameters['gan_w'] * self.loss_latent_b + \
class_weight * self.loss_class_a + \
class_weight * self.loss_class_b + \
hyperparameters['recon_x_w'] * self.loss_gen_recon_x_a + \
hyperparameters['recon_x_w'] * self.loss_gen_recon_x_b + \
hyperparameters['super_w'] * self.loss_supervision
self.loss_gen_total = (
hyperparameters['gan_w'] * self.loss_latent_a
+ hyperparameters['gan_w'] * self.loss_latent_b
+ class_weight * self.loss_class_a
+ class_weight * self.loss_class_b
+ hyperparameters['recon_x_w'] * self.loss_gen_recon_x_a
+ hyperparameters['recon_x_w'] * self.loss_gen_recon_x_b
+ hyperparameters['super_w'] * self.loss_supervision
) # yapf: disable

if variational:
self.loss_gen_total += hyperparameters['recon_kl_w'] * self.loss_gen_recon_kl_a + \
hyperparameters['recon_kl_w'] * self.loss_gen_recon_kl_b
self.loss_gen_total += (
hyperparameters['recon_kl_w'] * self.loss_gen_recon_kl_a
+ hyperparameters['recon_kl_w'] * self.loss_gen_recon_kl_b
) # yapf: disable

self.loss_gen_total.backward()
self.gen_opt.step()
Expand Down Expand Up @@ -496,7 +500,7 @@ def resume(self, checkpoint_dir):
state_dict = torch.load(os.path.join(checkpoint_dir, 'optimizer.pt'))
self.dis_opt.load_state_dict(state_dict['dis'])
self.gen_opt.load_state_dict(state_dict['gen'])
# Reinitilize schedulers
# Reinitialize schedulers
self.dis_scheduler = get_scheduler(self.dis_opt, hyperparameters, iterations)
self.gen_scheduler = get_scheduler(self.gen_opt, hyperparameters, iterations)
print('Resume from iteration %d' % iterations)
Expand All @@ -512,10 +516,6 @@ def save(self, snapshot_dir, iterations):
iterations : int
Current number of training iterations.

Returns
-------
None.

"""
# Save generators, discriminators, and optimizers
gen_name = os.path.join(snapshot_dir, 'gen_%08d.pt' % (iterations + 1))
Expand Down Expand Up @@ -546,10 +546,6 @@ def fit(self, train_mod1, train_mod2, aux_labels=None, checkpoint_directory='./c
val_ratio : float
Ratio for automatic train-validation split.

Returns
-------
None.

"""
hyperparameters = self.hyperparameters
idx = torch.randperm(train_mod1.shape[0])
Expand Down Expand Up @@ -586,8 +582,6 @@ def fit(self, train_mod1, train_mod2, aux_labels=None, checkpoint_directory='./c
self._gen_update(mod1, mod2, mod1, mod2, hyperparameters, variational=False)
print('Matching score:', self.score(train_mod1[val_idx], train_mod2[val_idx],
torch.arange(val_idx.shape[0])))
# print('Matching score:',
# self.evaluate(test_mod1, test_mod2, labels))

iterations += 1
if iterations >= hyperparameters['max_epochs']:
Expand Down
34 changes: 12 additions & 22 deletions dance/modules/multi_modality/match_modality/scmm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,20 +4,19 @@

Reference
---------
Minoura, Kodai, et al. "A mixture-of-experts deep generative model for integrated analysis of single-cell multiomics data." Cell reports methods 1.5 (2021): 100071.
Minoura, Kodai, et al. A mixture-of-experts deep generative model for integrated analysis of single-cell multiomics
data. Cell reports methods 1.5 (2021): 100071.

"""
import math
import os

import numpy as np
import pandas as pd
import scipy.sparse as sp
import torch
import torch.distributions as dist
import torch.nn as nn
import torch.nn.functional as F
from numpy import prod, sqrt
from numpy import prod
from pyro.distributions.zero_inflated import ZeroInflatedNegativeBinomial
from sklearn.cluster import DBSCAN, KMeans
from sklearn.neighbors import NearestNeighbors
Expand Down Expand Up @@ -131,7 +130,7 @@ def pz_params(self):
@property
def qz_x_params(self):
if self._qz_x_params is None:
raise NameError("qz_x params not initalised yet!")
raise NameError("qz_x params not initialized yet!")
return self._qz_x_params

@staticmethod
Expand Down Expand Up @@ -365,9 +364,11 @@ class MMVAE(nn.Module):
Parameters
----------
subtask : str
Name of the subtask which is composed of the name of two modality. This parameter will indicate some modality-specific features in the model.
Name of the subtask which is composed of the name of two modality. This parameter will indicate some
modality-specific features in the model.
params : argparse.Namespace
A Namespace object that contains arguments of MMVAE. For details of parameters in parser args, please refer to link (parser help document).
A Namespace object that contains arguments of MMVAE. For details of parameters in parser args, please refer to
link (parser help document).

"""

Expand Down Expand Up @@ -502,17 +503,11 @@ def fit(self, x_train, y_train, val_ratio=0.15):
----------
x_train : torch.Tensor
Input modality for training.

y_train : torch.Tensor
Target modality for training.

val_ratio : float
Ratio for automatic train-validation split.

Returns
-------
None.

"""

start_early_stop = self.params.deterministic_warmup
Expand All @@ -535,13 +530,8 @@ def fit(self, x_train, y_train, val_ratio=0.15):

train_mod1 = x_train.float().to(self.params.device)
train_mod2 = y_train.float().to(self.params.device)
# test_mod1 = x_test.float().to(self.params.device)
# test_mod2 = y_test.float().to(self.params.device)
# labels = labels.float().to(self.params.device)

vals = []
tr = []

tr, vals = [], []
for epoch in range(1, self.params.epochs + 1):
self.train()
b_loss = 0
Expand Down Expand Up @@ -576,7 +566,6 @@ def fit(self, x_train, y_train, val_ratio=0.15):
if epoch % 10 == 0:
print('Valid Matching score:',
self.score(train_mod1[val_idx], train_mod2[val_idx], torch.eye(val_idx.shape[0])))
# print('Test Matching score:', self.score(test_mod1, test_mod2, labels))

if epoch > start_early_stop and min(vals) != min(vals[-10:]):
print('Early stopped.')
Expand All @@ -593,7 +582,8 @@ def score(self, mod1, mod2, labels=None, metric='minkowski'):
mod2 : torch.Tensor
Features of modality 2.
labels : torch.Tensor optional
Labels of matching modality, i.e. cell correspondence between two modalities. Required when metric is not 'loss'.
Labels of matching modality, i.e. cell correspondence between two modalities. Required when metric is not
'loss'.
metric : str optional
Metric of the score function, by default to be 'minkowski'.

Expand Down Expand Up @@ -662,7 +652,7 @@ def predict(self, mod1, mod2, metric='minkowski'):
pred = []
with torch.no_grad():
for i, batch_idx in enumerate(data_loader):
dataT = [mod1[batch_idx].float().to(), mod2[batch_idx].float().to()]
dataT = [mod1[batch_idx], mod2[batch_idx]]
lats = self._get_latents(dataT, sampling=False)
if i == 0:
pred = lats
Expand Down