Skip to content

Commit

Permalink
Set the input_dropout rate in the main_branch to always be zero, even…
Browse files Browse the repository at this point in the history 
… if the user defines a different input_dropout value (so that input_dropout is only applied after the actual input layers)

Former-commit-id: 762e68d0722f6b0c4aef381df8ea07eefb8369ac
  • Loading branch information
@dlrsb
dlrsb committed May 10, 2022
1 parent ab1c896 commit 2839fa9
Show file tree
Hide file tree
Showing 2 changed files with 14 additions and 14 deletions.
18 changes: 9 additions & 9 deletions src/models/drug_pairs_build_functions.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,7 @@ def expr_drug_dense_model(expr_dim=None, drug_dim=None, expr_hlayers_sizes='[10]
# Additional dense layers after concatenating:
main_branch = dense_submodel(concat, hlayers_sizes=predictor_hlayers_sizes,
l1_regularization=l1, l2_regularization=l2,
hidden_activation=hidden_activation, input_dropout=input_dropout,
hidden_activation=hidden_activation, input_dropout=0,
hidden_dropout=hidden_dropout)
# Add output layer
output = Dense(1, activation='linear', kernel_initializer=initializer, name='output')(main_branch)
Expand Down Expand Up @@ -74,7 +74,7 @@ def expr_drug_textcnn_model(expr_dim=None, drug_dim=None, expr_hlayers_sizes='[1
drug2_input = Input(shape=(drug_seq_length,), dtype=tf.int32, name='drugB')
expr = dense_submodel(expr_input, hlayers_sizes=expr_hlayers_sizes, l1_regularization=l1, l2_regularization=l2,
hidden_activation=hidden_activation, input_dropout=input_dropout,
hidden_dropout=hidden_dropout, )
hidden_dropout=hidden_dropout)
drug_submodel = textcnn_submodel(seq_length=drug_seq_length, n_embedding=drug_n_embedding,
char_dict=drug_char_dict, kernel_sizes=drug_kernel_sizes,
num_filters=drug_num_filters, dropout=drug_dropout, l1=drug_l1, l2=drug_l2)
Expand Down Expand Up @@ -206,10 +206,10 @@ def expr1dconv_drug_textcnn_model(expr_dim=None, drug_dim=None, expr_num_filters


def expr2dconv_drug_dense_model(expr_dim=None, drug_dim=None, expr_num_filters='[32, 32]',
expr_kernel_size=(3, 3), expr_kernel_size_rest=(3, 3),
expr_pool_size=(2, 2), expr_batchnorm=True, drug_hlayers_sizes='[10]',
predictor_hlayers_sizes='[10]', initializer='he_normal', hidden_activation='relu', l1=0,
l2=0, input_dropout=0, hidden_dropout=0, optimizer='Adam', learn_rate=0.001):
expr_kernel_size=(3, 3), expr_pool_size=(2, 2), expr_batchnorm=True,
drug_hlayers_sizes='[10]', predictor_hlayers_sizes='[10]', initializer='he_normal',
hidden_activation='relu', l1=0, l2=0, input_dropout=0, hidden_dropout=0,
optimizer='Adam', learn_rate=0.001):
"""Build a multi-input deep learning model with separate feature-encoding subnetworks for expression data, drugA and
drugB. The expression subnetwork is a 2D CNN and the drug subnetworks use fully-connected layers."""
expr_input = Input(shape=expr_dim, name='expr')
Expand Down Expand Up @@ -644,7 +644,7 @@ def two_drug_gcn_model(n_atom_features=30, gcn_layers='[64, 64]', residual=True,

main_branch = dense_submodel(concat, hlayers_sizes=predictor_hlayers_sizes,
l1_regularization=l1, l2_regularization=l2,
hidden_activation=hidden_activation, input_dropout=input_dropout,
hidden_activation=hidden_activation, input_dropout=0,
hidden_dropout=hidden_dropout)
# Add output layer
output = Dense(1, activation='linear', kernel_initializer=initializer, name='output')(main_branch)
Expand Down Expand Up @@ -691,7 +691,7 @@ def expr_drug_gcn_model(expr_dim=None, drug_dim=None, expr_hlayers_sizes='[10]',

main_branch = dense_submodel(concat, hlayers_sizes=predictor_hlayers_sizes,
l1_regularization=l1, l2_regularization=l2,
hidden_activation=hidden_activation, input_dropout=input_dropout,
hidden_activation=hidden_activation, input_dropout=0,
hidden_dropout=hidden_dropout)
# Add output layer
output = Dense(1, activation='linear', kernel_initializer=initializer, name='output')(main_branch)
Expand Down Expand Up @@ -741,7 +741,7 @@ def expr_drug_gat_model(expr_dim=None, drug_dim=None, expr_hlayers_sizes='[10]',

main_branch = dense_submodel(concat, hlayers_sizes=predictor_hlayers_sizes,
l1_regularization=l1, l2_regularization=l2,
hidden_activation=hidden_activation, input_dropout=input_dropout,
hidden_activation=hidden_activation, input_dropout=0,
hidden_dropout=hidden_dropout)
# Add output layer
output = Dense(1, activation='linear', kernel_initializer=initializer, name='output')(main_branch)
Expand Down
10 changes: 5 additions & 5 deletions src/models/single_drug_build_functions.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,7 @@ def expr_drug_dense_model(expr_dim=None, drug_dim=None, expr_hlayers_sizes='[10]
# Additional dense layers after concatenating:
main_branch = dense_submodel(concat, hlayers_sizes=predictor_hlayers_sizes,
l1_regularization=l1, l2_regularization=l2,
hidden_activation=hidden_activation, input_dropout=input_dropout,
hidden_activation=hidden_activation, input_dropout=0,
hidden_dropout=hidden_dropout)
# Add output layer
output = Dense(1, activation='linear', kernel_initializer=initializer, name='output')(main_branch)
Expand Down Expand Up @@ -67,7 +67,7 @@ def expr_drug_textcnn_model(expr_dim=None, drug_dim=None, expr_hlayers_sizes='[1
drug_input = Input(shape=(drug_seq_length,), dtype=tf.int32, name='drugA')
expr = dense_submodel(expr_input, hlayers_sizes=expr_hlayers_sizes, l1_regularization=l1, l2_regularization=l2,
hidden_activation=hidden_activation, input_dropout=input_dropout,
hidden_dropout=hidden_dropout, )
hidden_dropout=hidden_dropout)
drug_submodel = textcnn_submodel(seq_length=drug_seq_length, n_embedding=drug_n_embedding,
char_dict=drug_char_dict, kernel_sizes=drug_kernel_sizes,
num_filters=drug_num_filters, dropout=drug_dropout, l1=drug_l1, l2=drug_l2)
Expand Down Expand Up @@ -121,7 +121,7 @@ def expr_drug_gcn_model(expr_dim=None, drug_dim=None, expr_hlayers_sizes='[10]',

main_branch = dense_submodel(concat, hlayers_sizes=predictor_hlayers_sizes,
l1_regularization=l1, l2_regularization=l2,
hidden_activation=hidden_activation, input_dropout=input_dropout,
hidden_activation=hidden_activation, input_dropout=0,
hidden_dropout=hidden_dropout)
# Add output layer
output = Dense(1, activation='linear', kernel_initializer=initializer, name='output')(main_branch)
Expand Down Expand Up @@ -168,7 +168,7 @@ def expr_drug_gat_model(expr_dim=None, drug_dim=None, expr_hlayers_sizes='[10]',

main_branch = dense_submodel(concat, hlayers_sizes=predictor_hlayers_sizes,
l1_regularization=l1, l2_regularization=l2,
hidden_activation=hidden_activation, input_dropout=input_dropout,
hidden_activation=hidden_activation, input_dropout=0,
hidden_dropout=hidden_dropout)
# Add output layer
output = Dense(1, activation='linear', kernel_initializer=initializer, name='output')(main_branch)
Expand Down Expand Up @@ -365,7 +365,7 @@ def expr_mut_cnv_drug_dense_model(expr_dim=None, mut_dim=None, cnv_dim=None, dru
output = Dense(1, activation='linear', kernel_initializer=initializer, name='output')(main_branch)

# create Model object
model = Model(inputs=[mut_input, cnv_input, expr_input, drug_input], outputs=[output])
model = Model(inputs=[expr_input, mut_input, cnv_input, drug_input], outputs=[output])

# Define optimizer
opt_class = dict(getmembers(optimizers))[optimizer]
Expand Down

0 comments on commit 2839fa9

Please sign in to comment.