scDREAMER API
- class scDREAMER.scDREAMER(sess, batch, cell_type, name, epoch=300, lr_ae=0.0002, lr_dis=0.0007, lr_bc=0.0007, beta1=0.9, batch_size=128, X_dim=2000, z_dim=10, dataset_name='Pancreas', num_layers=1, g_h_dim=[512, 256, 0, 0], d_h_dim=[512, 256, 0, 0], gen_activation='sig', leak=0.2, keep_param=0.9, trans='sparse', is_bn=False, g_iter=2, lam=1.0, sampler='normal', shuffle_type=1)
scDREAMER class parameter setting
- param batch:
obs key containing the batch information
- type batch:
string
- param cell_type:
obs key containing the cell type information
- type cell_type:
string
- param name:
path to the AnnData(adata) object. adata.X contains counts info. adata.obs contains batch and celltype info
- type name:
AnnData
- param epoch:
number of epoches to train the model, defaults to 300
- type epoch:
int, optional
- param lr_ae:
learning rate, defaults to 0.0007
- type lr_ae:
float, optional
- param lr_bc:
learning rate, defaults to 0.0007
- type lr_bc:
float, optional
- param lr_dis:
learning rate, defaults to 0.0007
- type lr_dis:
float, optional
- param beta1:
beta1, defaults to 0.9
- type beta1:
float, optional
- param batch_size:
batch size, defaults to 128
- type batch_size:
int, optional
- param X_dim:
Top heighly variable genes, defaults to 2000
- type X_dim:
int, optional
- param z_dim:
Embeddings dimention, defaults to 10
- type z_dim:
int, optional
- param dataset_name:
dataset name, defaults to “Pancreas”
- type dataset_name:
str, optional
- param num_layers:
number of hidden layers, defaults to 1
- type num_layers:
int, optional
- param g_h_dim:
neurons in encoder hidden layers (uses only num_layers), defaults to [512, 256, 0, 0]
- type g_h_dim:
list, optional
- param d_h_dim:
neurons in decoder hidden layers (uses only num_layers), defaults to [512, 256, 0, 0]
- type d_h_dim:
list, optional
- param gen_activation:
activation function, defaults to “sig”
- type gen_activation:
str, optional
- param leak:
leak, defaults to 0.2
- type leak:
float, optional
- param keep_param:
keep, defaults to 0.9
- type keep_param:
float, optional
- param trans:
translation, defaults to “sparse”
- type trans:
str, optional
- param sampler:
z sampler, defaults to “normal”
- type sampler:
str, optional
- param shuffle_type:
type of shuffling, defaults to 1
- type shuffle_type:
int, optional
- batchClassifier(z, z_dim, reuse=False)
batchClassifier takes the latent space representation and try to differentiate between different batches :param z: tensor of shape [batch_size, z_dim] :param batch: tensor of shape [batch_size] -> batchinfo of the train data :param reuse: True -> Reuse the discriminator variables,False -> Create or search of variables before creating :return: tensor of shape [batch_size, 1]
- build_model()
Build the complete tensorflow network of the model
- decoder(z, reuse=False)
Decoder part of the autoencoder.
- Parameters:
z (tensor) – input to the decoder
reuse (bool, optional) – True -> Reuse the decoder variables, False -> Create or search of variables before creating, defaults to False
- Returns:
tensor which should ideally be the input given to the encoder.
- Return type:
tensor
- discriminator2(x, x_dim, reuse=False)
Discriminator that is used to match the input x with reconstructed x.
- Parameters:
x – tensor of shape [batch_size, x_dim]
reuse – True -> Reuse the discriminator variables, False -> Create or search of variables before creating
- Returns:
tensor of shape [batch_size, 1]
- encoder(x, reuse=False)
Encoder network of VAE
- Parameters:
x (tensor) – tensor to pass encoder shape
reuse (bool, optional) – reuse the weight of network(used at the time of building the network), defaults to False
- train_cluster()
Train the model and saves the integrated lowerdimentional embeedings for every 50 epoch