SURE-tools 2.4.7__py3-none-any.whl → 2.4.34__py3-none-any.whl

SURE/DensityFlow2.py

@@ -0,0 +1,1422 @@
+import pyro
+import pyro.distributions as dist
+from pyro.optim import ExponentialLR
+from pyro.infer import SVI, JitTraceEnum_ELBO, TraceEnum_ELBO, config_enumerate
+
+import torch
+import torch.nn as nn
+from torch.utils.data import DataLoader
+from torch.distributions.utils import logits_to_probs, probs_to_logits, clamp_probs
+from torch.distributions import constraints
+from torch.distributions.transforms import SoftmaxTransform
+
+from .utils.custom_mlp import MLP, Exp, ZeroBiasMLP
+from .utils.utils import CustomDataset, CustomDataset2, CustomDataset4, tensor_to_numpy, convert_to_tensor
+
+
+import os
+import argparse
+import random
+import numpy as np
+import datatable as dt
+from tqdm import tqdm
+from scipy import sparse
+
+import scanpy as sc
+from .atac import binarize
+
+from typing import Literal
+
+import warnings
+warnings.filterwarnings("ignore")
+
+import dill as pickle
+import gzip
+from packaging.version import Version
+torch_version = torch.__version__
+
+
+def set_random_seed(seed):
+    # Set seed for PyTorch
+    torch.manual_seed(seed)
+    
+    # If using CUDA, set the seed for CUDA
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed(seed)
+        torch.cuda.manual_seed_all(seed)  # For multi-GPU setups.
+    
+    # Set seed for NumPy
+    np.random.seed(seed)
+    
+    # Set seed for Python's random module
+    random.seed(seed)
+
+    # Set seed for Pyro
+    pyro.set_rng_seed(seed)
+
+class DensityFlow2(nn.Module):
+    def __init__(self,
+                 input_size: int,
+                 codebook_size: int = 200,
+                 cell_factor_size: int = 0,
+                 turn_off_cell_specific: bool = False,
+                 supervised_mode: bool = False,
+                 z_dim: int = 10,
+                 z_dist: Literal['normal','studentt','laplacian','cauchy','gumbel'] = 'gumbel',
+                 loss_func: Literal['negbinomial','poisson','multinomial','bernoulli'] = 'multinomial',
+                 inverse_dispersion: float = 10.0,
+                 use_zeroinflate: bool = False,
+                 hidden_layers: list = [500],
+                 hidden_layer_activation: Literal['relu','softplus','leakyrelu','linear'] = 'relu',
+                 nn_dropout: float = 0.1,
+                 post_layer_fct: list = ['layernorm'],
+                 post_act_fct: list = None,
+                 config_enum: str = 'parallel',
+                 use_cuda: bool = True,
+                 seed: int = 42,
+                 zero_bias: bool|list = True,
+                 dtype = torch.float32, # type: ignore
+                 ):
+        super().__init__()
+
+        self.input_size = input_size
+        self.cell_factor_size = cell_factor_size
+        self.inverse_dispersion = inverse_dispersion
+        self.latent_dim = z_dim
+        self.hidden_layers = hidden_layers
+        self.decoder_hidden_layers = hidden_layers[::-1]
+        self.allow_broadcast = config_enum == 'parallel'
+        self.use_cuda = use_cuda
+        self.loss_func = loss_func
+        self.options = None
+        self.code_size=codebook_size
+        self.supervised_mode=supervised_mode
+        self.latent_dist = z_dist
+        self.dtype = dtype
+        self.use_zeroinflate=use_zeroinflate
+        self.nn_dropout = nn_dropout
+        self.post_layer_fct = post_layer_fct
+        self.post_act_fct = post_act_fct
+        self.hidden_layer_activation = hidden_layer_activation
+        if type(zero_bias) == list:
+            self.use_bias = [not x for x in zero_bias]
+        else:
+            self.use_bias = [not zero_bias] * self.cell_factor_size
+        #self.use_bias = not zero_bias
+        self.turn_off_cell_specific = turn_off_cell_specific
+        
+        self.codebook_weights = None
+
+        set_random_seed(seed)
+        self.setup_networks()
+        
+        print(f"🧬 DensityFlow2 Initialized:")
+        print(f"   - Latent Dimension: {self.latent_dim}")
+        print(f"   - Gene Dimension: {self.input_size}")
+        print(f"   - Hidden Dimensions: {self.hidden_layers}")
+        print(f"   - Device: {self.get_device()}")
+        print(f"   - Parameters: {sum(p.numel() for p in self.parameters()):,}")
+
+    def setup_networks(self):
+        latent_dim = self.latent_dim
+        hidden_sizes = self.hidden_layers
+
+        nn_layer_norm, nn_batch_norm, nn_layer_dropout = False, False, False
+        na_layer_norm, na_batch_norm, na_layer_dropout = False, False, False
+
+        if self.post_layer_fct is not None:
+            nn_layer_norm=True if ('layernorm' in self.post_layer_fct) or ('layer_norm' in self.post_layer_fct) else False
+            nn_batch_norm=True if ('batchnorm' in self.post_layer_fct) or ('batch_norm' in self.post_layer_fct) else False
+            nn_layer_dropout=True if 'dropout' in self.post_layer_fct else False
+
+        if self.post_act_fct is not None:
+            na_layer_norm=True if ('layernorm' in self.post_act_fct) or ('layer_norm' in self.post_act_fct) else False
+            na_batch_norm=True if ('batchnorm' in self.post_act_fct) or ('batch_norm' in self.post_act_fct) else False
+            na_layer_dropout=True if 'dropout' in self.post_act_fct else False
+
+        if nn_layer_norm and nn_batch_norm and nn_layer_dropout:
+            post_layer_fct = lambda layer_ix, total_layers, layer: nn.Sequential(nn.Dropout(self.nn_dropout),nn.BatchNorm1d(layer.module.out_features), nn.LayerNorm(layer.module.out_features))
+        elif nn_layer_norm and nn_layer_dropout:
+            post_layer_fct = lambda layer_ix, total_layers, layer: nn.Sequential(nn.Dropout(self.nn_dropout), nn.LayerNorm(layer.module.out_features))
+        elif nn_batch_norm and nn_layer_dropout:
+            post_layer_fct = lambda layer_ix, total_layers, layer: nn.Sequential(nn.Dropout(self.nn_dropout), nn.BatchNorm1d(layer.module.out_features))
+        elif nn_layer_norm and nn_batch_norm:
+            post_layer_fct = lambda layer_ix, total_layers, layer: nn.Sequential(nn.BatchNorm1d(layer.module.out_features), nn.LayerNorm(layer.module.out_features))
+        elif nn_layer_norm:
+            post_layer_fct = lambda layer_ix, total_layers, layer: nn.LayerNorm(layer.module.out_features)
+        elif nn_batch_norm:
+            post_layer_fct = lambda layer_ix, total_layers, layer:nn.BatchNorm1d(layer.module.out_features)
+        elif nn_layer_dropout:
+            post_layer_fct = lambda layer_ix, total_layers, layer: nn.Dropout(self.nn_dropout)
+        else:
+            post_layer_fct = lambda layer_ix, total_layers, layer: None
+
+        if na_layer_norm and na_batch_norm and na_layer_dropout:
+            post_act_fct = lambda layer_ix, total_layers, layer: nn.Sequential(nn.Dropout(self.nn_dropout),nn.BatchNorm1d(layer.module.out_features), nn.LayerNorm(layer.module.out_features))
+        elif na_layer_norm and na_layer_dropout:
+            post_act_fct = lambda layer_ix, total_layers, layer: nn.Sequential(nn.Dropout(self.nn_dropout), nn.LayerNorm(layer.module.out_features))
+        elif na_batch_norm and na_layer_dropout:
+            post_act_fct = lambda layer_ix, total_layers, layer: nn.Sequential(nn.Dropout(self.nn_dropout), nn.BatchNorm1d(layer.module.out_features))
+        elif na_layer_norm and na_batch_norm:
+            post_act_fct = lambda layer_ix, total_layers, layer: nn.Sequential(nn.BatchNorm1d(layer.module.out_features), nn.LayerNorm(layer.module.out_features))
+        elif na_layer_norm:
+            post_act_fct = lambda layer_ix, total_layers, layer: nn.LayerNorm(layer.module.out_features)
+        elif na_batch_norm:
+            post_act_fct = lambda layer_ix, total_layers, layer:nn.BatchNorm1d(layer.module.out_features)
+        elif na_layer_dropout:
+            post_act_fct = lambda layer_ix, total_layers, layer: nn.Dropout(self.nn_dropout)
+        else:
+            post_act_fct = lambda layer_ix, total_layers, layer: None
+
+        if self.hidden_layer_activation == 'relu':
+            activate_fct = nn.ReLU
+        elif self.hidden_layer_activation == 'softplus':
+            activate_fct = nn.Softplus
+        elif self.hidden_layer_activation == 'leakyrelu':
+            activate_fct = nn.LeakyReLU
+        elif self.hidden_layer_activation == 'linear':
+            activate_fct = nn.Identity
+
+        if self.supervised_mode:
+            self.encoder_n = MLP(
+                [self.input_size] + hidden_sizes + [self.code_size],
+                activation=activate_fct,
+                output_activation=None,
+                post_layer_fct=post_layer_fct,
+                post_act_fct=post_act_fct,
+                allow_broadcast=self.allow_broadcast,
+                use_cuda=self.use_cuda,
+            )
+        else:
+            self.encoder_n = MLP(
+                [self.latent_dim] + hidden_sizes + [self.code_size],
+                activation=activate_fct,
+                output_activation=None,
+                post_layer_fct=post_layer_fct,
+                post_act_fct=post_act_fct,
+                allow_broadcast=self.allow_broadcast,
+                use_cuda=self.use_cuda,
+            )
+
+        self.encoder_zn = MLP(
+            [self.input_size] + hidden_sizes + [[latent_dim, latent_dim]],
+            activation=activate_fct,
+            output_activation=[None, Exp],
+            post_layer_fct=post_layer_fct,
+            post_act_fct=post_act_fct,
+            allow_broadcast=self.allow_broadcast,
+            use_cuda=self.use_cuda,
+        )
+        
+        if self.loss_func == 'negbinomial':
+            self.encoder_inverse_dispersion = MLP(
+                [self.latent_dim] + hidden_sizes + [[self.input_size, self.input_size]],
+                activation=activate_fct,
+                output_activation=[Exp, Exp],
+                post_layer_fct=post_layer_fct,
+                post_act_fct=post_act_fct,
+                allow_broadcast=self.allow_broadcast,
+                use_cuda=self.use_cuda,
+            )
+        
+        if self.cell_factor_size>0:
+            self.cell_factor_effect = nn.ModuleList()
+            for i in np.arange(self.cell_factor_size):
+                if self.use_bias[i]:
+                    if self.turn_off_cell_specific:
+                        self.cell_factor_effect.append(MLP(
+                            [1] + self.decoder_hidden_layers + [self.latent_dim],
+                            activation=activate_fct,
+                            output_activation=None,
+                            post_layer_fct=post_layer_fct,
+                            post_act_fct=post_act_fct,
+                            allow_broadcast=self.allow_broadcast,
+                            use_cuda=self.use_cuda,
+                            )
+                        )
+                    else:
+                        self.cell_factor_effect.append(MLP(
+                            [self.latent_dim+1] + self.decoder_hidden_layers + [self.latent_dim],
+                            activation=activate_fct,
+                            output_activation=None,
+                            post_layer_fct=post_layer_fct,
+                            post_act_fct=post_act_fct,
+                            allow_broadcast=self.allow_broadcast,
+                            use_cuda=self.use_cuda,
+                            )
+                        )
+                else:
+                    if self.turn_off_cell_specific:
+                        self.cell_factor_effect.append(ZeroBiasMLP(
+                            [1] + self.decoder_hidden_layers + [self.latent_dim],
+                            activation=activate_fct,
+                            output_activation=None,
+                            post_layer_fct=post_layer_fct,
+                            post_act_fct=post_act_fct,
+                            allow_broadcast=self.allow_broadcast,
+                            use_cuda=self.use_cuda,
+                            )
+                        )
+                    else:
+                        self.cell_factor_effect.append(ZeroBiasMLP(
+                            [self.latent_dim+1] + self.decoder_hidden_layers + [self.latent_dim],
+                            activation=activate_fct,
+                            output_activation=None,
+                            post_layer_fct=post_layer_fct,
+                            post_act_fct=post_act_fct,
+                            allow_broadcast=self.allow_broadcast,
+                            use_cuda=self.use_cuda,
+                            )
+                        )
+            
+        self.decoder_concentrate = MLP(
+                    [self.latent_dim] + self.decoder_hidden_layers + [self.input_size],
+                    activation=activate_fct,
+                    output_activation=None,
+                    post_layer_fct=post_layer_fct,
+                    post_act_fct=post_act_fct,
+                    allow_broadcast=self.allow_broadcast,
+                    use_cuda=self.use_cuda,
+                )
+
+        if self.latent_dist == 'studentt':
+            self.codebook = MLP(
+                [self.code_size] + hidden_sizes + [[latent_dim,latent_dim]],
+                activation=activate_fct,
+                output_activation=[Exp,None],
+                post_layer_fct=post_layer_fct,
+                post_act_fct=post_act_fct,
+                allow_broadcast=self.allow_broadcast,
+                use_cuda=self.use_cuda,
+            )
+        else:
+            self.codebook = MLP(
+                [self.code_size] + hidden_sizes + [latent_dim],
+                activation=activate_fct,
+                output_activation=None,
+                post_layer_fct=post_layer_fct,
+                post_act_fct=post_act_fct,
+                allow_broadcast=self.allow_broadcast,
+                use_cuda=self.use_cuda,
+            )
+
+        if self.use_cuda:
+            self.cuda()
+
+    def get_device(self):
+        return next(self.parameters()).device
+
+    def cutoff(self, xs, thresh=None):
+        eps = torch.finfo(xs.dtype).eps
+        
+        if not thresh is None:
+            if eps < thresh:
+                eps = thresh
+
+        xs = xs.clamp(min=eps)
+
+        if torch.any(torch.isnan(xs)):
+            xs[torch.isnan(xs)] = eps
+
+        return xs
+
+    def softmax(self, xs):
+        #xs = SoftmaxTransform()(xs)
+        xs = dist.Multinomial(total_count=1, logits=xs).mean
+        return xs
+    
+    def sigmoid(self, xs):
+        #sigm_enc = nn.Sigmoid()
+        #xs = sigm_enc(xs)
+        #xs = clamp_probs(xs)
+        xs = dist.Bernoulli(logits=xs).mean
+        return xs
+
+    def softmax_logit(self, xs):
+        eps = torch.finfo(xs.dtype).eps
+        xs = self.softmax(xs)
+        xs = torch.logit(xs, eps=eps)
+        return xs
+
+    def logit(self, xs):
+        eps = torch.finfo(xs.dtype).eps
+        xs = torch.logit(xs, eps=eps)
+        return xs
+
+    def dirimulti_param(self, xs):
+        xs = self.dirimulti_mass * self.sigmoid(xs)
+        return xs
+
+    def multi_param(self, xs):
+        xs = self.softmax(xs)
+        return xs
+
+    def model1(self, xs):
+        pyro.module('DensityFlow2', self)
+
+        eps = torch.finfo(xs.dtype).eps
+        batch_size = xs.size(0)
+        self.options = dict(dtype=xs.dtype, device=xs.device)
+        
+        if self.loss_func=='negbinomial':
+            total_count = pyro.param("inverse_dispersion", self.inverse_dispersion *
+                                     xs.new_ones(self.input_size), constraint=constraints.positive)
+            
+        if self.use_zeroinflate:
+            gate_logits = pyro.param("dropout_rate", xs.new_zeros(self.input_size))
+            
+        acs_scale = pyro.param("codebook_scale", xs.new_ones(self.latent_dim), constraint=constraints.positive)
+        
+        I = torch.eye(self.code_size)
+        if self.latent_dist=='studentt':
+            acs_dof,acs_loc = self.codebook(I)
+        else:
+            acs_loc = self.codebook(I)
+
+        with pyro.plate('data'):
+            prior = torch.zeros(batch_size, self.code_size, **self.options)
+            ns = pyro.sample('n', dist.OneHotCategorical(logits=prior))
+
+            zn_loc = torch.matmul(ns,acs_loc)
+            #zn_scale = torch.matmul(ns,acs_scale)
+            zn_scale = acs_scale
+
+            if self.latent_dist == 'studentt':
+                prior_dof = torch.matmul(ns,acs_dof)
+                zns = pyro.sample('zn', dist.StudentT(df=prior_dof, loc=zn_loc, scale=zn_scale).to_event(1))
+            elif self.latent_dist == 'laplacian':
+                zns = pyro.sample('zn', dist.Laplace(zn_loc, zn_scale).to_event(1))
+            elif self.latent_dist == 'cauchy':
+                zns = pyro.sample('zn', dist.Cauchy(zn_loc, zn_scale).to_event(1))
+            elif self.latent_dist == 'normal':
+                zns = pyro.sample('zn', dist.Normal(zn_loc, zn_scale).to_event(1))
+            elif self.latent_dist == 'gumbel':
+                zns = pyro.sample('zn', dist.Gumbel(zn_loc, zn_scale).to_event(1))
+
+            zs = zns
+            concentrate = self.decoder_concentrate(zs)
+            if self.loss_func in ['bernoulli']:
+                log_theta = concentrate
+            else:
+                rate = concentrate.exp()
+                theta = dist.DirichletMultinomial(total_count=1, concentration=rate).mean
+                if self.loss_func == 'poisson':
+                    rate = theta * torch.sum(xs, dim=1, keepdim=True)
+
+            if self.loss_func == 'negbinomial':
+                if self.use_zeroinflate:
+                    pyro.sample('x', dist.ZeroInflatedDistribution(dist.NegativeBinomial(total_count=total_count, probs=theta),gate_logits=gate_logits).to_event(1), obs=xs)
+                else:
+                    pyro.sample('x', dist.NegativeBinomial(total_count=total_count, probs=theta).to_event(1), obs=xs)
+            elif self.loss_func == 'poisson':
+                if self.use_zeroinflate:
+                    pyro.sample('x', dist.ZeroInflatedDistribution(dist.Poisson(rate=rate),gate_logits=gate_logits).to_event(1), obs=xs.round())
+                else:
+                    pyro.sample('x', dist.Poisson(rate=rate).to_event(1), obs=xs.round())
+            elif self.loss_func == 'multinomial':
+                #pyro.sample('x', dist.Multinomial(total_count=int(1e8), probs=theta), obs=xs)
+                pyro.sample('x', dist.Multinomial(total_count=int(1e8), logits=concentrate), obs=xs)
+            elif self.loss_func == 'bernoulli':
+                if self.use_zeroinflate:
+                    pyro.sample('x', dist.ZeroInflatedDistribution(dist.Bernoulli(logits=log_theta),gate_logits=gate_logits).to_event(1), obs=xs)
+                else:
+                    pyro.sample('x', dist.Bernoulli(logits=log_theta).to_event(1), obs=xs)
+
+    def guide1(self, xs):
+        with pyro.plate('data'):
+            #zn_loc, zn_scale = self.encoder_zn(xs)
+            zn_loc, zn_scale = self._get_basal_embedding(xs)
+            zns = pyro.sample('zn', dist.Normal(zn_loc, zn_scale).to_event(1))
+
+            alpha = self.encoder_n(zns)
+            ns = pyro.sample('n', dist.OneHotCategorical(logits=alpha))
+
+    def model2(self, xs, us=None):
+        pyro.module('DensityFlow2', self)
+
+        eps = torch.finfo(xs.dtype).eps
+        batch_size = xs.size(0)
+        self.options = dict(dtype=xs.dtype, device=xs.device)
+        
+        if self.loss_func=='negbinomial':
+            #total_count = pyro.param("inverse_dispersion", self.inverse_dispersion *
+            #                         xs.new_ones(self.input_size), constraint=constraints.positive)
+            with pyro.plate("genes", self.input_size):
+                inverse_dispersion = pyro.sample("inverse_dispersion", dist.LogNormal(self.inverse_dispersion, 0.5).to_event(1))
+            
+        if self.use_zeroinflate:
+            gate_logits = pyro.param("dropout_rate", xs.new_zeros(self.input_size))
+            
+        acs_scale = pyro.param("codebook_scale", xs.new_ones(self.latent_dim), constraint=constraints.positive)
+
+        I = torch.eye(self.code_size)
+        if self.latent_dist=='studentt':
+            acs_dof,acs_loc = self.codebook(I)
+        else:
+            acs_loc = self.codebook(I)
+
+        with pyro.plate('data'):
+            prior = torch.zeros(batch_size, self.code_size, **self.options)
+            ns = pyro.sample('n', dist.OneHotCategorical(logits=prior))
+
+            zn_loc = torch.matmul(ns,acs_loc)
+            #zn_scale = torch.matmul(ns,acs_scale)
+            zn_scale = acs_scale
+
+            if self.latent_dist == 'studentt':
+                prior_dof = torch.matmul(ns,acs_dof)
+                zns = pyro.sample('zn', dist.StudentT(df=prior_dof, loc=zn_loc, scale=zn_scale).to_event(1))
+            elif self.latent_dist == 'laplacian':
+                zns = pyro.sample('zn', dist.Laplace(zn_loc, zn_scale).to_event(1))
+            elif self.latent_dist == 'cauchy':
+                zns = pyro.sample('zn', dist.Cauchy(zn_loc, zn_scale).to_event(1))
+            elif self.latent_dist == 'normal':
+                zns = pyro.sample('zn', dist.Normal(zn_loc, zn_scale).to_event(1))
+            elif self.latent_dist == 'gumbel':
+                zns = pyro.sample('zn', dist.Gumbel(zn_loc, zn_scale).to_event(1))
+
+            '''if self.cell_factor_size>0:
+                zus = self._total_shifts(zns, us)
+                zs = zns+zus
+            else:
+                zs = zns'''
+
+            zs = zns
+            concentrate = self.decoder_concentrate(zs)
+            for i in np.arange(self.cell_factor_size):
+                zus = self._cell_shift(zs, i, us[:,i].reshape(-1,1))
+                concentrate += self.decoder_concentrate(zus)
+                
+            if self.loss_func in ['bernoulli']:
+                log_theta = concentrate
+            elif self.loss_func in ['negbinomial']:
+                mu = concentrate.exp()
+            else:
+                rate = concentrate.exp()
+                theta = dist.DirichletMultinomial(total_count=1, concentration=rate).mean
+                if self.loss_func == 'poisson':
+                    rate = theta * torch.sum(xs, dim=1, keepdim=True)
+
+            if self.loss_func == 'negbinomial':
+                logits = (mu.log()-inverse_dispersion.log()).clamp(min=-10, max=10)
+                if self.use_zeroinflate:
+                    pyro.sample('x', dist.ZeroInflatedDistribution(dist.NegativeBinomial(total_count=inverse_dispersion, 
+                                                                                         logits=logits),gate_logits=gate_logits).to_event(1), obs=xs)
+                else:
+                    pyro.sample('x', dist.NegativeBinomial(total_count=inverse_dispersion, 
+                                                           logits=logits).to_event(1), obs=xs)
+            elif self.loss_func == 'poisson':
+                if self.use_zeroinflate:
+                    pyro.sample('x', dist.ZeroInflatedDistribution(dist.Poisson(rate=rate),gate_logits=gate_logits).to_event(1), obs=xs.round())
+                else:
+                    pyro.sample('x', dist.Poisson(rate=rate).to_event(1), obs=xs.round())
+            elif self.loss_func == 'multinomial':
+                #pyro.sample('x', dist.Multinomial(total_count=int(1e8), probs=theta), obs=xs)
+                pyro.sample('x', dist.Multinomial(total_count=int(1e8), logits=concentrate), obs=xs)
+            elif self.loss_func == 'bernoulli':
+                if self.use_zeroinflate:
+                    pyro.sample('x', dist.ZeroInflatedDistribution(dist.Bernoulli(logits=log_theta),gate_logits=gate_logits).to_event(1), obs=xs)
+                else:
+                    pyro.sample('x', dist.Bernoulli(logits=log_theta).to_event(1), obs=xs)
+
+    def guide2(self, xs, us=None):
+        with pyro.plate('data'):
+            #zn_loc, zn_scale = self.encoder_zn(xs)
+            zn_loc, zn_scale = self._get_basal_embedding(xs)
+            zns = pyro.sample('zn', dist.Normal(zn_loc, zn_scale).to_event(1))
+
+            alpha = self.encoder_n(zns)
+            ns = pyro.sample('n', dist.OneHotCategorical(logits=alpha))
+            
+            if self.loss_func == 'negbinomial':
+                id_loc,id_scale = self.encoder_inverse_dispersion(zns)
+                with pyro.plate("genes", self.input_size):
+                    pyro.sample("inverse_dispersion", dist.LogNormal(id_loc, id_scale).to_event(1))
+
+    def model3(self, xs, ys, embeds=None):
+        pyro.module('DensityFlow2', self)
+
+        eps = torch.finfo(xs.dtype).eps
+        batch_size = xs.size(0)
+        self.options = dict(dtype=xs.dtype, device=xs.device)
+        
+        if self.loss_func=='negbinomial':
+            total_count = pyro.param("inverse_dispersion", self.inverse_dispersion *
+                                     xs.new_ones(self.input_size), constraint=constraints.positive)
+            
+        if self.use_zeroinflate:
+            gate_logits = pyro.param("dropout_rate", xs.new_zeros(self.input_size))
+            
+        acs_scale = pyro.param("codebook_scale", xs.new_ones(self.latent_dim), constraint=constraints.positive)
+
+        I = torch.eye(self.code_size)
+        if self.latent_dist=='studentt':
+            acs_dof,acs_loc = self.codebook(I)
+        else:
+            acs_loc = self.codebook(I)
+
+        with pyro.plate('data'):
+            #prior = torch.zeros(batch_size, self.code_size, **self.options)
+            prior = self.encoder_n(xs)
+            ns = pyro.sample('n', dist.OneHotCategorical(logits=prior), obs=ys)
+
+            zn_loc = torch.matmul(ns,acs_loc)
+            #prior_scale = torch.matmul(ns,acs_scale)
+            zn_scale = acs_scale
+
+            if self.latent_dist=='studentt':
+                prior_dof = torch.matmul(ns,acs_dof)
+                if embeds is None:
+                    zns = pyro.sample('zn', dist.StudentT(df=prior_dof, loc=zn_loc, scale=zn_scale).to_event(1))
+                else:
+                    zns = pyro.sample('zn', dist.StudentT(df=prior_dof, loc=zn_loc, scale=zn_scale).to_event(1), obs=embeds)
+            elif self.latent_dist=='laplacian':
+                if embeds is None:
+                    zns = pyro.sample('zn', dist.Laplace(zn_loc, zn_scale).to_event(1))
+                else:
+                    zns = pyro.sample('zn', dist.Laplace(zn_loc, zn_scale).to_event(1), obs=embeds)
+            elif self.latent_dist=='cauchy':
+                if embeds is None:
+                    zns = pyro.sample('zn', dist.Cauchy(zn_loc, zn_scale).to_event(1))
+                else:
+                    zns = pyro.sample('zn', dist.Cauchy(zn_loc, zn_scale).to_event(1), obs=embeds)
+            elif self.latent_dist=='normal':
+                if embeds is None:
+                    zns = pyro.sample('zn', dist.Normal(zn_loc, zn_scale).to_event(1))
+                else:
+                    zns = pyro.sample('zn', dist.Normal(zn_loc, zn_scale).to_event(1), obs=embeds)
+            elif self.z_dist == 'gumbel':
+                if embeds is None:
+                    zns = pyro.sample('zn', dist.Gumbel(zn_loc, zn_scale).to_event(1))
+                else:
+                    zns = pyro.sample('zn', dist.Gumbel(zn_loc, zn_scale).to_event(1), obs=embeds)
+
+            zs = zns
+
+            concentrate = self.decoder_concentrate(zs)
+            if self.loss_func in ['bernoulli']:
+                log_theta = concentrate
+            else:
+                rate = concentrate.exp()
+                theta = dist.DirichletMultinomial(total_count=1, concentration=rate).mean
+                if self.loss_func == 'poisson':
+                    rate = theta * torch.sum(xs, dim=1, keepdim=True)
+
+            if self.loss_func == 'negbinomial':
+                if self.use_zeroinflate:
+                    pyro.sample('x', dist.ZeroInflatedDistribution(dist.NegativeBinomial(total_count=total_count, probs=theta),gate_logits=gate_logits).to_event(1), obs=xs)
+                else:
+                    pyro.sample('x', dist.NegativeBinomial(total_count=total_count, probs=theta).to_event(1), obs=xs)
+            elif self.loss_func == 'poisson':
+                if self.use_zeroinflate:
+                    pyro.sample('x', dist.ZeroInflatedDistribution(dist.Poisson(rate=rate),gate_logits=gate_logits).to_event(1), obs=xs.round())
+                else:
+                    pyro.sample('x', dist.Poisson(rate=rate).to_event(1), obs=xs.round())
+            elif self.loss_func == 'multinomial':
+                #pyro.sample('x', dist.Multinomial(total_count=int(1e8), probs=theta), obs=xs)
+                pyro.sample('x', dist.Multinomial(total_count=int(1e8), logits=concentrate), obs=xs)
+            elif self.loss_func == 'bernoulli':
+                if self.use_zeroinflate:
+                    pyro.sample('x', dist.ZeroInflatedDistribution(dist.Bernoulli(logits=log_theta),gate_logits=gate_logits).to_event(1), obs=xs)
+                else:
+                    pyro.sample('x', dist.Bernoulli(logits=log_theta).to_event(1), obs=xs)
+
+    def guide3(self, xs, ys, embeds=None):
+        with pyro.plate('data'):
+            if embeds is None:
+                #zn_loc, zn_scale = self.encoder_zn(xs)
+                zn_loc, zn_scale = self._get_basal_embedding(xs)
+                zns = pyro.sample('zn', dist.Normal(zn_loc, zn_scale).to_event(1))
+            else:
+                zns = embeds
+
+    def model4(self, xs, us, ys, embeds=None):
+        pyro.module('DensityFlow2', self)
+
+        eps = torch.finfo(xs.dtype).eps
+        batch_size = xs.size(0)
+        self.options = dict(dtype=xs.dtype, device=xs.device)
+        
+        if self.loss_func=='negbinomial':
+            total_count = pyro.param("inverse_dispersion", self.inverse_dispersion *
+                                     xs.new_ones(self.input_size), constraint=constraints.positive)
+            
+        if self.use_zeroinflate:
+            gate_logits = pyro.param("dropout_rate", xs.new_zeros(self.input_size))
+            
+        acs_scale = pyro.param("codebook_scale", xs.new_ones(self.latent_dim), constraint=constraints.positive)
+
+        I = torch.eye(self.code_size)
+        if self.latent_dist=='studentt':
+            acs_dof,acs_loc = self.codebook(I)
+        else:
+            acs_loc = self.codebook(I)
+
+        with pyro.plate('data'):
+            #prior = torch.zeros(batch_size, self.code_size, **self.options)
+            prior = self.encoder_n(xs)
+            ns = pyro.sample('n', dist.OneHotCategorical(logits=prior), obs=ys)
+
+            zn_loc = torch.matmul(ns,acs_loc)
+            #prior_scale = torch.matmul(ns,acs_scale)
+            zn_scale = acs_scale
+
+            if self.latent_dist=='studentt':
+                prior_dof = torch.matmul(ns,acs_dof)
+                if embeds is None:
+                    zns = pyro.sample('zn', dist.StudentT(df=prior_dof, loc=zn_loc, scale=zn_scale).to_event(1))
+                else:
+                    zns = pyro.sample('zn', dist.StudentT(df=prior_dof, loc=zn_loc, scale=zn_scale).to_event(1), obs=embeds)
+            elif self.latent_dist=='laplacian':
+                if embeds is None:
+                    zns = pyro.sample('zn', dist.Laplace(zn_loc, zn_scale).to_event(1))
+                else:
+                    zns = pyro.sample('zn', dist.Laplace(zn_loc, zn_scale).to_event(1), obs=embeds)
+            elif self.latent_dist=='cauchy':
+                if embeds is None:
+                    zns = pyro.sample('zn', dist.Cauchy(zn_loc, zn_scale).to_event(1))
+                else:
+                    zns = pyro.sample('zn', dist.Cauchy(zn_loc, zn_scale).to_event(1), obs=embeds)
+            elif self.latent_dist=='normal':
+                if embeds is None:
+                    zns = pyro.sample('zn', dist.Normal(zn_loc, zn_scale).to_event(1))
+                else:
+                    zns = pyro.sample('zn', dist.Normal(zn_loc, zn_scale).to_event(1), obs=embeds)
+            elif self.z_dist == 'gumbel':
+                if embeds is None:
+                    zns = pyro.sample('zn', dist.Gumbel(zn_loc, zn_scale).to_event(1))
+                else:
+                    zns = pyro.sample('zn', dist.Gumbel(zn_loc, zn_scale).to_event(1), obs=embeds)
+
+            '''if self.cell_factor_size>0:
+                zus = self._total_shifts(zns, us)
+                zs = zns+zus
+            else:
+                zs = zns'''
+
+            zs = zns
+            concentrate = self.decoder_concentrate(zs)
+            for i in np.arange(self.cell_factor_size):
+                zus = self._cell_shift(zs, i, us[:,i].reshape(-1,1))
+                concentrate += self.decoder_concentrate(zus)
+            
+            if self.loss_func in ['bernoulli']:
+                log_theta = concentrate
+            else:
+                rate = concentrate.exp()
+                theta = dist.DirichletMultinomial(total_count=1, concentration=rate).mean
+                if self.loss_func == 'poisson':
+                    rate = theta * torch.sum(xs, dim=1, keepdim=True)
+
+            if self.loss_func == 'negbinomial':
+                if self.use_zeroinflate:
+                    pyro.sample('x', dist.ZeroInflatedDistribution(dist.NegativeBinomial(total_count=total_count, probs=theta),gate_logits=gate_logits).to_event(1), obs=xs)
+                else:
+                    pyro.sample('x', dist.NegativeBinomial(total_count=total_count, probs=theta).to_event(1), obs=xs)
+            elif self.loss_func == 'poisson':
+                if self.use_zeroinflate:
+                    pyro.sample('x', dist.ZeroInflatedDistribution(dist.Poisson(rate=rate),gate_logits=gate_logits).to_event(1), obs=xs.round())
+                else:
+                    pyro.sample('x', dist.Poisson(rate=rate).to_event(1), obs=xs.round())
+            elif self.loss_func == 'multinomial':
+                #pyro.sample('x', dist.Multinomial(total_count=int(1e8), probs=theta), obs=xs)
+                pyro.sample('x', dist.Multinomial(total_count=int(1e8), logits=concentrate), obs=xs)
+            elif self.loss_func == 'bernoulli':
+                if self.use_zeroinflate:
+                    pyro.sample('x', dist.ZeroInflatedDistribution(dist.Bernoulli(logits=log_theta),gate_logits=gate_logits).to_event(1), obs=xs)
+                else:
+                    pyro.sample('x', dist.Bernoulli(logits=log_theta).to_event(1), obs=xs)
+
+    def guide4(self, xs, us, ys, embeds=None):
+        with pyro.plate('data'):
+            if embeds is None:
+                #zn_loc, zn_scale = self.encoder_zn(xs)
+                zn_loc, zn_scale = self._get_basal_embedding(xs)
+                zns = pyro.sample('zn', dist.Normal(zn_loc, zn_scale).to_event(1))
+            else:
+                zns = embeds
+
+    def _total_shifts(self, zns, us):
+        zus = None
+        for i in np.arange(self.cell_factor_size):
+            if i==0:
+                #if self.turn_off_cell_specific:
+                #    zus = self.cell_factor_effect[i](us[:,i].reshape(-1,1))
+                #else:
+                #    zus = self.cell_factor_effect[i]([zns,us[:,i].reshape(-1,1)])
+                zus = self._cell_shift(zns, i, us[:,i].reshape(-1,1))
+            else:
+                #if self.turn_off_cell_specific:
+                #    zus = zus + self.cell_factor_effect[i](us[:,i].reshape(-1,1))
+                #else:
+                #    zus = zus + self.cell_factor_effect[i]([zns,us[:,i].reshape(-1,1)])
+                zus = zus + self._cell_shift(zns, i, us[:,i].reshape(-1,1))
+        return zus
+                        
+    def _get_codebook_identity(self):
+        return torch.eye(self.code_size, **self.options)
+    
+    def _get_codebook(self):
+        I = torch.eye(self.code_size, **self.options)
+        if self.latent_dist=='studentt':
+            _,cb = self.codebook(I)
+        else:
+            cb = self.codebook(I)
+        return cb
+    
+    def get_codebook(self):
+        """
+        Return the mean part of metacell codebook
+        """
+        cb = self._get_codebook()
+        cb = tensor_to_numpy(cb)
+        return cb
+
+    def _get_basal_embedding(self, xs):           
+        loc, scale = self.encoder_zn(xs)
+        return loc, scale 
+    
+    def get_basal_embedding(self, 
+                             xs, 
+                             batch_size: int = 1024):
+        """
+        Return cells' basal latent representations
+
+        Parameters
+        ----------
+        xs
+            Single-cell expression matrix. It should be a Numpy array or a Pytorch Tensor.
+        batch_size
+            Size of batch processing.
+        use_decoder
+            If toggled on, the latent representations will be reconstructed from the metacell codebook
+        soft_assign
+            If toggled on, the assignments of cells will use probabilistic values.
+        """
+        xs = self.preprocess(xs)
+        xs = convert_to_tensor(xs, device=self.get_device())
+        dataset = CustomDataset(xs)
+        dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=False)
+
+        Z = []
+        with tqdm(total=len(dataloader), desc='', unit='batch') as pbar:
+            for X_batch, _ in dataloader:
+                zns,_ = self._get_basal_embedding(X_batch)
+                Z.append(tensor_to_numpy(zns))
+                pbar.update(1)
+
+        Z = np.concatenate(Z)
+        return Z
+    
+    def _code(self, xs):
+        if self.supervised_mode:
+            alpha = self.encoder_n(xs)
+        else:
+            #zns,_ = self.encoder_zn(xs)
+            zns,_ = self._get_basal_embedding(xs)
+            alpha = self.encoder_n(zns)
+        return alpha
+    
+    def code(self, xs, batch_size=1024):
+        xs = self.preprocess(xs)
+        xs = convert_to_tensor(xs, device=self.get_device())
+        dataset = CustomDataset(xs)
+        dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=False)
+
+        A = []
+        with tqdm(total=len(dataloader), desc='', unit='batch') as pbar:
+            for X_batch, _ in dataloader:
+                a = self._code(X_batch)
+                A.append(tensor_to_numpy(a))
+                pbar.update(1)
+
+        A = np.concatenate(A)
+        return A
+    
+    def _soft_assignments(self, xs):
+        alpha = self._code(xs)
+        alpha = self.softmax(alpha)
+        return alpha
+    
+    def soft_assignments(self, xs, batch_size=1024):
+        """
+        Map cells to metacells and return the probabilistic values of metacell assignments
+        """
+        xs = self.preprocess(xs)
+        xs = convert_to_tensor(xs, device=self.get_device())
+        dataset = CustomDataset(xs)
+        dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=False)
+
+        A = []
+        with tqdm(total=len(dataloader), desc='', unit='batch') as pbar:
+            for X_batch, _ in dataloader:
+                a = self._soft_assignments(X_batch)
+                A.append(tensor_to_numpy(a))
+                pbar.update(1)
+
+        A = np.concatenate(A)
+        return A
+    
+    def _hard_assignments(self, xs):
+        alpha = self._code(xs)
+        res, ind = torch.topk(alpha, 1)
+        ns = torch.zeros_like(alpha).scatter_(1, ind, 1.0)
+        return ns
+    
+    def hard_assignments(self, xs, batch_size=1024):
+        """
+        Map cells to metacells and return the assigned metacell identities.
+        """
+        xs = self.preprocess(xs)
+        xs = convert_to_tensor(xs, device=self.get_device())
+        dataset = CustomDataset(xs)
+        dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=False)
+
+        A = []
+        with tqdm(total=len(dataloader), desc='', unit='batch') as pbar:
+            for X_batch, _ in dataloader:
+                a = self._hard_assignments(X_batch)
+                A.append(tensor_to_numpy(a))
+                pbar.update(1)
+
+        A = np.concatenate(A)
+        return A
+    
+    def predict(self, xs, us, perturbs_predict:list, perturbs_reference:list, library_sizes=None):
+        perturbs_reference = np.array(perturbs_reference)
+        
+        # basal embedding
+        zs = self.get_basal_embedding(xs)
+        for pert in perturbs_predict:
+            pert_idx = int(np.where(perturbs_reference==pert)[0])
+            us_i = us[:,pert_idx].reshape(-1,1)
+            
+            # factor effect of xs 
+            dzs0 = self.get_cell_shift(zs, perturb_idx=pert_idx, perturb_us=us_i)
+            
+            # perturbation effect 
+            ps = np.ones_like(us_i)
+            if np.sum(np.abs(ps-us_i))>=1:
+                dzs = self.get_cell_shift(zs, perturb_idx=pert_idx, perturb_us=ps)          
+                zs = zs + dzs0 + dzs
+            else:
+                zs = zs + dzs0
+        
+        if library_sizes is None:
+            library_sizes = np.sum(xs, axis=1, keepdims=True)
+        elif type(library_sizes) == list:
+            library_sizes = np.array(library_sizes)
+            library_sizes = library_sizes.reshape(-1,1)
+        elif len(library_sizes.shape)==1:
+            library_sizes = library_sizes.reshape(-1,1)
+            
+        counts = self.get_counts(zs, library_sizes=library_sizes)
+        
+        return counts, zs
+    
+    def _cell_shift(self, zs, perturb_idx, perturb):
+        #zns,_ = self.encoder_zn(xs)    
+        #zns,_ = self._get_basal_embedding(xs)   
+        zns = zs
+        if perturb.ndim==2:
+            if self.turn_off_cell_specific:
+                ms = self.cell_factor_effect[perturb_idx](perturb)
+            else:
+                ms = self.cell_factor_effect[perturb_idx]([zns, perturb])
+        else:
+            if self.turn_off_cell_specific:
+                ms = self.cell_factor_effect[perturb_idx](perturb.reshape(-1,1))
+            else:
+                ms = self.cell_factor_effect[perturb_idx]([zns, perturb.reshape(-1,1)])
+            
+        return ms 
+
+    def get_cell_shift(self, 
+                             zs, 
+                             perturb_idx,
+                             perturb_us,
+                             batch_size: int = 1024):
+        """
+        Return cells' changes in the latent space induced by specific perturbation of a factor
+
+        """
+        #xs = self.preprocess(xs)
+        zs = convert_to_tensor(zs, device=self.get_device())
+        ps = convert_to_tensor(perturb_us, device=self.get_device())
+        dataset = CustomDataset2(zs,ps)
+        dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=False)
+
+        Z = []
+        with tqdm(total=len(dataloader), desc='', unit='batch') as pbar:
+            for Z_batch, P_batch, _ in dataloader:
+                zns = self._cell_shift(Z_batch, perturb_idx, P_batch)
+                Z.append(tensor_to_numpy(zns))
+                pbar.update(1)
+
+        Z = np.concatenate(Z)
+        return Z
+    
+    def _get_theta(self, delta_zs):
+        return self.decoder_concentrate(delta_zs)
+    
+    def get_theta(self, 
+                             delta_zs, 
+                             batch_size: int = 1024):
+        """
+        Return cells' changes in the feature space induced by specific perturbation of a factor
+
+        """
+        delta_zs = convert_to_tensor(delta_zs, device=self.get_device())
+        dataset = CustomDataset(delta_zs)
+        dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=False)
+
+        R = []
+        with tqdm(total=len(dataloader), desc='', unit='batch') as pbar:
+            for delta_Z_batch, _ in dataloader:
+                r = self._get_theta(delta_Z_batch)
+                R.append(tensor_to_numpy(r))
+                pbar.update(1)
+
+        R = np.concatenate(R)
+        return R
+    
+    def _count(self, concentrate, library_size=None):
+        if self.loss_func == 'bernoulli':
+            #counts = self.sigmoid(concentrate)
+            counts = dist.Bernoulli(logits=concentrate).to_event(1).mean
+        elif self.loss_func == 'multinomial':
+            theta = dist.Multinomial(total_count=int(1e8), logits=concentrate).mean
+            counts = theta * library_size
+        else:
+            rate = concentrate.exp()
+            theta = dist.DirichletMultinomial(total_count=1, concentration=rate).mean
+            counts = theta * library_size
+        return counts
+    
+    def get_counts(self, concentrate, 
+                   library_sizes,
+                   batch_size: int = 1024):
+
+        concentrate = convert_to_tensor(concentrate, device=self.get_device())
+        
+        if type(library_sizes) == list:
+            library_sizes = np.array(library_sizes).reshape(-1,1)
+        elif len(library_sizes.shape)==1:
+            library_sizes = library_sizes.reshape(-1,1)
+        ls = convert_to_tensor(library_sizes, device=self.get_device())
+            
+        dataset = CustomDataset2(concentrate,ls)
+        dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=False)
+        
+        E = []
+        with tqdm(total=len(dataloader), desc='', unit='batch') as pbar:
+            for C_batch, L_batch, _ in dataloader:
+                counts = self._count(C_batch, L_batch)
+                E.append(tensor_to_numpy(counts))
+                pbar.update(1)
+        
+        E = np.concatenate(E)
+        return E
+    
+    def preprocess(self, xs, threshold=0):
+        if self.loss_func == 'bernoulli':
+            ad = sc.AnnData(xs)
+            binarize(ad, threshold=threshold)
+            xs = ad.X.copy()
+        else:
+            xs = np.round(xs)
+            
+        if sparse.issparse(xs):
+            xs = xs.toarray()
+        return xs 
+    
+    def fit(self, xs, 
+            us = None, 
+            ys = None,
+            zs = None,
+            num_epochs: int = 500, 
+            learning_rate: float = 0.0001, 
+            batch_size: int = 256, 
+            algo: Literal['adam','rmsprop','adamw'] = 'adam', 
+            beta_1: float = 0.9, 
+            weight_decay: float = 0.005, 
+            decay_rate: float = 0.9,
+            config_enum: str = 'parallel',
+            threshold: int = 0,
+            use_jax: bool = True):
+        """
+        Train the DensityFlow2 model.
+
+        Parameters
+        ----------
+        xs
+            Single-cell experssion matrix. It should be a Numpy array or a Pytorch Tensor. Rows are cells and columns are features.
+        us
+            cell-level factor matrix. 
+        ys
+            Desired factor matrix. It should be a Numpy array or a Pytorch Tensor. Rows are cells and columns are desired factors.
+        num_epochs
+            Number of training epochs.
+        learning_rate
+            Parameter for training.
+        batch_size
+            Size of batch processing.
+        algo
+            Optimization algorithm.
+        beta_1
+            Parameter for optimization.
+        weight_decay
+            Parameter for optimization.
+        decay_rate 
+            Parameter for optimization.
+        use_jax
+            If toggled on, Jax will be used for speeding up. CAUTION: This will raise errors because of unknown reasons when it is called in
+            the Python script or Jupyter notebook. It is OK if it is used when runing DensityFlow2 in the shell command.
+        """
+        xs = self.preprocess(xs, threshold=threshold)
+        xs = convert_to_tensor(xs, dtype=self.dtype, device=self.get_device())
+        if us is not None:
+            us = convert_to_tensor(us, dtype=self.dtype, device=self.get_device())
+        if ys is not None:
+            ys = convert_to_tensor(ys, dtype=self.dtype, device=self.get_device())
+        if zs is not None:
+            zs = convert_to_tensor(zs, dtype=self.dtype, device=self.get_device())
+
+        dataset = CustomDataset4(xs, us, ys, zs)
+        dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=True)
+
+        # setup the optimizer
+        optim_params = {'lr': learning_rate, 'betas': (beta_1, 0.999), 'weight_decay': weight_decay}
+
+        if algo.lower()=='rmsprop':
+            optimizer = torch.optim.RMSprop
+        elif algo.lower()=='adam':
+            optimizer = torch.optim.Adam
+        elif algo.lower() == 'adamw':
+            optimizer = torch.optim.AdamW
+        else:
+            raise ValueError("An optimization algorithm must be specified.")
+        scheduler = ExponentialLR({'optimizer': optimizer, 'optim_args': optim_params, 'gamma': decay_rate})
+
+        pyro.clear_param_store()
+
+        # set up the loss(es) for inference, wrapping the guide in config_enumerate builds the loss as a sum
+        # by enumerating each class label form the sampled discrete categorical distribution in the model
+        Elbo = JitTraceEnum_ELBO if use_jax else TraceEnum_ELBO
+        elbo = Elbo(max_plate_nesting=1, strict_enumeration_warning=False)
+        if us is None:
+            if ys is None:
+                guide = config_enumerate(self.guide1, config_enum, expand=True)
+                loss_basic = SVI(self.model1, guide, scheduler, loss=elbo)
+            else:
+                guide = config_enumerate(self.guide3, config_enum, expand=True)
+                loss_basic = SVI(self.model3, guide, scheduler, loss=elbo)
+        else:
+            if ys is None:
+                guide = config_enumerate(self.guide2, config_enum, expand=True)
+                loss_basic = SVI(self.model2, guide, scheduler, loss=elbo)
+            else:
+                guide = config_enumerate(self.guide4, config_enum, expand=True)
+                loss_basic = SVI(self.model4, guide, scheduler, loss=elbo)
+
+        # build a list of all losses considered
+        losses = [loss_basic]
+        num_losses = len(losses)
+
+        with tqdm(total=num_epochs, desc='Training', unit='epoch') as pbar:
+            for epoch in range(num_epochs):
+                epoch_losses = [0.0] * num_losses
+                for batch_x, batch_u, batch_y, batch_z, _ in dataloader:
+                    if us is None:
+                        batch_u = None
+                    if ys is None:
+                        batch_y = None
+                    if zs is None:
+                        batch_z = None
+
+                    for loss_id in range(num_losses):
+                        if batch_u is None:
+                            if batch_y is None:
+                                new_loss = losses[loss_id].step(batch_x)
+                            else:
+                                new_loss = losses[loss_id].step(batch_x, batch_y, batch_z)
+                        else:
+                            if batch_y is None:
+                                new_loss = losses[loss_id].step(batch_x, batch_u)
+                            else:
+                                new_loss = losses[loss_id].step(batch_x, batch_u, batch_y, batch_z)
+                        epoch_losses[loss_id] += new_loss
+
+                avg_epoch_losses_ = map(lambda v: v / len(dataloader), epoch_losses)
+                avg_epoch_losses = map(lambda v: "{:.4f}".format(v), avg_epoch_losses_)
+
+                # store the loss
+                str_loss = " ".join(map(str, avg_epoch_losses))
+
+                # Update progress bar
+                pbar.set_postfix({'loss': str_loss})
+                pbar.update(1)
+                
+    @classmethod
+    def save_model(cls, model, file_path, compression=False):
+        """Save the model to the specified file path."""
+        file_path = os.path.abspath(file_path)
+
+        model.eval()
+        if compression:
+            with gzip.open(file_path, 'wb') as pickle_file:
+                pickle.dump(model, pickle_file)
+        else:
+            with open(file_path, 'wb') as pickle_file:
+                pickle.dump(model, pickle_file)
+
+        print(f'Model saved to {file_path}')
+
+    @classmethod
+    def load_model(cls, file_path):
+        """Load the model from the specified file path and return an instance."""
+        print(f'Model loaded from {file_path}')
+
+        file_path = os.path.abspath(file_path)
+        if file_path.endswith('gz'):
+            with gzip.open(file_path, 'rb') as pickle_file:
+                model = pickle.load(pickle_file)
+        else:
+            with open(file_path, 'rb') as pickle_file:
+                model = pickle.load(pickle_file)
+        
+        return model
+
+        
+EXAMPLE_RUN = (
+    "example run: DensityFlow2 --help"
+)
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description="DensityFlow2\n{}".format(EXAMPLE_RUN))
+
+    parser.add_argument(
+        "--cuda", action="store_true", help="use GPU(s) to speed up training"
+    )
+    parser.add_argument(
+        "--jit", action="store_true", help="use PyTorch jit to speed up training"
+    )
+    parser.add_argument(
+        "-n", "--num-epochs", default=200, type=int, help="number of epochs to run"
+    )
+    parser.add_argument(
+        "-enum",
+        "--enum-discrete",
+        default="parallel",
+        help="parallel, sequential or none. uses parallel enumeration by default",
+    )
+    parser.add_argument(
+        "-data",
+        "--data-file",
+        default=None,
+        type=str,
+        help="the data file",
+    )
+    parser.add_argument(
+        "-cf",
+        "--cell-factor-file",
+        default=None,
+        type=str,
+        help="the file for the record of cell-level factors",
+    )
+    parser.add_argument(
+        "-bs",
+        "--batch-size",
+        default=1000,
+        type=int,
+        help="number of cells to be considered in a batch",
+    )
+    parser.add_argument(
+        "-lr",
+        "--learning-rate",
+        default=0.0001,
+        type=float,
+        help="learning rate for Adam optimizer",
+    )
+    parser.add_argument(
+        "-cs",
+        "--codebook-size",
+        default=100,
+        type=int,
+        help="size of vector quantization codebook",
+    )
+    parser.add_argument(
+        "--z-dist",
+        default='gumbel',
+        type=str,
+        choices=['normal','laplacian','studentt','gumbel','cauchy'],
+        help="distribution model for latent representation",
+    )
+    parser.add_argument(
+        "-zd",
+        "--z-dim",
+        default=10,
+        type=int,
+        help="size of the tensor representing the latent variable z variable",
+    )
+    parser.add_argument(
+        "-likeli",
+        "--likelihood",
+        default='negbinomial',
+        type=str,
+        choices=['negbinomial', 'multinomial', 'poisson', 'bernoulli'],
+        help="specify the distribution likelihood function",
+    )
+    parser.add_argument(
+        "-zi",
+        "--zeroinflate",
+        action="store_true",
+        help="use zero-inflated estimation",
+    )
+    parser.add_argument(
+        "-id",
+        "--inverse-dispersion",
+        default=10.0,
+        type=float,
+        help="inverse dispersion prior for negative binomial",
+    )
+    parser.add_argument(
+        "-hl",
+        "--hidden-layers",
+        nargs="+",
+        default=[500],
+        type=int,
+        help="a tuple (or list) of MLP layers to be used in the neural networks "
+        "representing the parameters of the distributions in our model",
+    )
+    parser.add_argument(
+        "-hla",
+        "--hidden-layer-activation",
+        default='relu',
+        type=str,
+        choices=['relu','softplus','leakyrelu','linear'],
+        help="activation function for hidden layers",
+    )
+    parser.add_argument(
+        "-plf",
+        "--post-layer-function",
+        nargs="+",
+        default=['layernorm'],
+        type=str,
+        help="post functions for hidden layers, could be none, dropout, layernorm, batchnorm, or combination, default is 'dropout layernorm'",
+    )
+    parser.add_argument(
+        "-paf",
+        "--post-activation-function",
+        nargs="+",
+        default=['none'],
+        type=str,
+        help="post functions for activation layers, could be none or dropout, default is 'none'",
+    )
+    parser.add_argument(
+        "-64",
+        "--float64",
+        action="store_true",
+        help="use double float precision",
+    )
+    parser.add_argument(
+        "-dr",
+        "--decay-rate",
+        default=0.9,
+        type=float,
+        help="decay rate for Adam optimizer",
+    )
+    parser.add_argument(
+        "--layer-dropout-rate",
+        default=0.1,
+        type=float,
+        help="droput rate for neural networks",
+    )
+    parser.add_argument(
+        "-b1",
+        "--beta-1",
+        default=0.95,
+        type=float,
+        help="beta-1 parameter for Adam optimizer",
+    )  
+    parser.add_argument(
+        "--seed",
+        default=None,
+        type=int,
+        help="seed for controlling randomness in this example",
+    )
+    parser.add_argument(
+        "--save-model",
+        default=None,
+        type=str,
+        help="path to save model for prediction",
+    )
+    args = parser.parse_args()
+    return args
+
+def main():
+    args = parse_args()
+    assert (
+        (args.data_file is not None) and (
+            os.path.exists(args.data_file))
+    ), "data file must be provided"
+
+    if args.seed is not None:
+        set_random_seed(args.seed)
+
+    if args.float64:
+        dtype = torch.float64
+        torch.set_default_dtype(torch.float64)
+    else:
+        dtype = torch.float32
+        torch.set_default_dtype(torch.float32)
+
+    xs = dt.fread(file=args.data_file, header=True).to_numpy()
+    us = None 
+    if args.cell_factor_file is not None:
+        us = dt.fread(file=args.cell_factor_file, header=True).to_numpy()
+
+    input_size = xs.shape[1]
+    cell_factor_size = 0 if us is None else us.shape[1]
+
+    ###########################################
+    df = DensityFlow2(
+        input_size=input_size,
+        cell_factor_size=cell_factor_size,
+        inverse_dispersion=args.inverse_dispersion,
+        z_dim=args.z_dim,
+        hidden_layers=args.hidden_layers,
+        hidden_layer_activation=args.hidden_layer_activation,
+        use_cuda=args.cuda,
+        config_enum=args.enum_discrete,
+        use_zeroinflate=args.zeroinflate,
+        loss_func=args.likelihood,
+        nn_dropout=args.layer_dropout_rate,
+        post_layer_fct=args.post_layer_function,
+        post_act_fct=args.post_activation_function,
+        codebook_size=args.codebook_size,
+        z_dist = args.z_dist,
+        dtype=dtype,
+    )
+
+    df.fit(xs, us=us, 
+             num_epochs=args.num_epochs,
+             learning_rate=args.learning_rate,
+             batch_size=args.batch_size,
+             beta_1=args.beta_1,
+             decay_rate=args.decay_rate,
+             use_jax=args.jit,
+             config_enum=args.enum_discrete,
+             )
+
+    if args.save_model is not None:
+        if args.save_model.endswith('gz'):
+            DensityFlow2.save_model(df, args.save_model, compression=True)
+        else:
+            DensityFlow2.save_model(df, args.save_model)
+    
+
+
+if __name__ == "__main__":
+    main()