Highway Networks implement in pytorch
Highway Networks
in detail: Highway Networks
the file of hyperparams.py contains all hyperparams that need to modify, based on yours nedds, select neural networks what you want and config the hyperparams.
the file of main-hyperparams.py is the main function,run the command ("python main_hyperparams.py") to execute the demo.
the folder of models contains all neural networks models.
the file of train_ALL_CNN.py is the train function about CNN
the file of train_ALL_LSTM.py is the train function about LSTM
the folder of loaddata contains some file of load dataset
the folder of word2vec is the file of word embedding that you want to use
the folder of data contains the dataset file,contains train data,dev data,test data.
the file of Parameters.txt is being used to save all parameters values.
the file of Test_Result.txt is being used to save the result of test,in the demo,save a model and test a model immediately,and int the end of training, will calculate the best result value.
model-BiLSTM-1.py is a simple bidirection LSTM neural networks model.
model-BiLSTM-List.py is a simple bidirection LSTM neural networks model.
model-BiLSTM-Cat.py is a simple bidirection LSTM variant neural networks model.
model-CNN.py is a simple CNN neural networks model.
model-HBiLSTM.py is a simple HIghway BiLstm neural networks model.
model-HBiLSTM-CAT.py is a simple HIghway BiLstm variant neural networks model.
model-HCNN.py is a simple HIghway CNN neural networks model.
model-HighWay.py is a simple Highway networks model.
model-HighWayBiLSTM.py is a simple Highway BiLstm variant networks model.
model-HighWayCNN.py is a simple Highway CNN variant networks model.
model-HighWay-BiLSTM.py is a HighWay NetWorks variant model with use in the BiLSTM model.
model-HighWay-CNN.py is a HighWay NetWorks model variant with use in the CNN model.
learning_rate: initial learning rate.
epochs:number of epochs for train
batch_size:batch size for training
log_interval:how many steps to wait before logging training status
test_interval:how many steps to wait before testing
save_interval:how many steps to wait before saving
save_dir:where to save the snapshot
datafile_path:datafile path
name_trainfile:name of the train file
name_devfile:name of the dev file
name_testfile: name of the test file
char_data: whether to use the strategy of char-level data
shuffle:whether to shuffle the dataset when load dataset
epochs_shuffle:whether to shuffle the dataset when train in every epoch
TWO-CLASS-TASK:execute two-classification-task
dropout:the probability for dropout
max_norm:l2 constraint of parameters
clip-max-norm:the values of prevent the explosion and Vanishing in Gradient
kernel_sizes:comma-separated kernel size to use for convolution
kernel_num:number of each kind of kernel
static:whether to update the gradient during train
Adam:select the optimizer of adam
SGD:select the optimizer of SGD
Adadelta:select the optimizer of Adadelta
optim-momentum-value:the parameter in the optimizer
wide_conv:whether to use wide convcolution True : wide False : narrow
min_freq:min freq to include during built the vocab when use torchtext, default is 1
word_Embedding: use word embedding
embed_dim:number of embedding dimension
word-Embedding-Path:the path of word embedding file
lstm-hidden-dim:the hidden dim with lstm model
lstm-num-layers:the num of hidden layers with lstm
no_cuda: use cuda
num_threads:set the value of threads when run the demo
init_weight:whether to init weight
init-weight-value:the value of init weight
weight-decay:L2 weight_decay,default value is zero in optimizer
seed_num:set the num of random seed
rm_model:whether to delete the model after test acc so that to save space