def quora_dataset_config7(is_debug, output_log=None):
from model.self_attention_model import PreprocessWrapper
from read_data.quora_question_pair.load_data import load_parsed_quora_data
train, valid, test, embedding_matrix, character_size, word_pad_id, character_pad_id = \
load_parsed_quora_data(debug=is_debug,
word_vector_name="glove_300d",
n_gram=1)
from model.feed_forward_network import FFNWithCrossCompare
return {
"model_fn": FFNWithCrossCompare,
"model_dict": {
"word_embedding": embedding_matrix,
},
"pre_process_module_fn": PreprocessWrapper,
"pre_process_module_dict": {
"pad_idx": word_pad_id,
"character_pad_idx": character_pad_id,
},
"data": [train, valid, test],
"batch_size": 128,
"train_loss": BCELoss,
"clip_norm": None,
"name": "FFN_try",
"optimizer": optim.Adam,
"need_pad": True,
"optimizer_dict": {"betas": (0.9, 0.999), "weight_decay": 3e-7, },
"epcohes": 80,
"lr": 1e-3,
"evaluate_object_list": [SequenceBinaryClassExactMatch(gpu_index=get_gpu_index())],
"epoch_ratio": 0.25,
}
def quora_dataset_config5(is_debug, output_log=None):
from model.self_attention_model import PreprocessWrapper
from read_data.quora_question_pair.load_data import load_parsed_quora_data
train, valid, test, embedding_matrix, character_size, word_pad_id, character_pad_id = \
load_parsed_quora_data(debug=is_debug,
word_vector_name="glove_300d",
n_gram=1)
from qanet.qanet import QANet
return {
"model_fn": QANet,
"model_dict": {
"word_embedding_matrix": embedding_matrix, "char_embedding_matrix": None,
"params": {
"word_embed_dim": 300,
"highway_n_layers": 2,
"hidden_size": 128,
"embed_encoder_resize_kernel_size": 7,
"embed_encoder_resize_pad": 3,
"embed_encoder_n_blocks": 1,
"embed_encoder_n_conv": 4,
"embed_encoder_kernel_size": 7,
"embed_encoder_pad": 3,
"embed_encoder_conv_type": "depthwise_separable",
"embed_encoder_with_self_attn": False,
"embed_encoder_n_heads": 8,
"model_encoder_n_blocks": 7,
"model_encoder_n_conv": 2,
"model_encoder_kernel_size": 7,
"model_encoder_pad": 3,
"model_encoder_conv_type": "depthwise_separable",
"model_encoder_with_self_attn": False,
"model_encoder_n_heads": 8,
"batch_size": 32,
}
},
"pre_process_module_fn": PreprocessWrapper,
"pre_process_module_dict": {
"pad_idx": word_pad_id,
"character_pad_idx": character_pad_id,
},
"data": [train, valid, test],
"batch_size": 64,
"train_loss": BCELoss,
"clip_norm": None,
"name": "qa_net",
"optimizer": optim.Adam,
"need_pad": True,
"optimizer_dict": {"betas": (0.8, 0.999), "weight_decay": 3e-7, },
"epcohes": 80,
"lr": 0.001,
"evaluate_object_list": [SequenceBinaryClassExactMatch(gpu_index=get_gpu_index())],
"epoch_ratio": 0.25,
}
def quora_dataset_config3(is_debug, output_log=None):
from model.self_attention_model import SelfAttentionPairModel, PreprocessWrapper
from read_data.quora_question_pair.load_data import load_parsed_quora_data
train, valid, test, embedding_matrix, character_size, word_pad_id, character_pad_id = \
load_parsed_quora_data(debug=is_debug,
word_vector_name="fasttext",
n_gram=1)
return {
"model_fn": SelfAttentionPairModel,
"model_dict": {
"word_embedding": embedding_matrix,
"character_number": character_size,
"character_embedding_dim": 600,
"character_n_filters": 200,
"character_kernel_size": 5,
"character_padding": 2,
"self_attention_layer": 5,
},
"pre_process_module_fn": PreprocessWrapper,
"pre_process_module_dict": {
"pad_idx": word_pad_id,
"character_pad_idx": character_pad_id,
},
"data": [train, valid, test],
"batch_size": 8,
"train_loss": BCELoss,
"clip_norm": 10,
"name": "sequence_graph_encoder_decoder_for_method_name",
"optimizer": optim.Adam,
"need_pad": True,
"optimizer_dict": {"betas": (0.8, 0.999), "weight_decay": 3e-7, },
"epcohes": 80,
"lr": 1e-4,
"evaluate_object_list": [SequenceBinaryClassExactMatch(gpu_index=get_gpu_index())],
}
def quora_dataset_config1(is_debug, output_log=None):
from model.sentence_pair_graph import SequenceGraphModel, PreprocessWrapper
from read_data.quora_question_pair.load_data import load_parsed_quora_data
train, valid, test, embedding_matrix, character_size, word_pad_id, character_pad_id = \
load_parsed_quora_data(debug=is_debug,
word_vector_name="glove_300d",
n_gram=1)
return {
"model_fn": SequenceGraphModel,
"model_dict": {
"word_embedding": embedding_matrix,
"character_number": character_size,
"character_embedding_dim": 16,
"character_n_filters": 32,
"character_kernel_size": 5,
"character_padding": 2,
"n_link_type": 3,
"hidden_state_size": 200,
"n_dynamic_link_layer": 2,
"n_fix_graph": 1,
"graph_itr": 5,
"n_classes": 2,
"summary_node": False,
"tie_weight": False,
},
"pre_process_module_fn": PreprocessWrapper,
"pre_process_module_dict": {
"pad_idx": word_pad_id,
"character_pad_idx": character_pad_id,
"summary_node": True,
},
"data": [train, valid, test],
"batch_size": 8,
"train_loss": BCELoss,
"clip_norm": 10,
"name": "sequence_graph_encoder_decoder_for_method_name",
"optimizer": optim.Adam,
"need_pad": True,
"optimizer_dict": {"betas": (0.8, 0.999), "weight_decay": 3e-7, },
"epcohes": 80,
"lr": 1e-4,
"evaluate_object_list": [SequenceBinaryClassExactMatch(gpu_index=get_gpu_index())],
}
def quora_dataset_config4(is_debug, output_log=None):
from model.self_attention_model import PreprocessWrapper
from model.graph_cluster_model import GraphClusterModel
from read_data.quora_question_pair.load_data import load_parsed_quora_data
train, valid, test, embedding_matrix, character_size, word_pad_id, character_pad_id = \
load_parsed_quora_data(debug=is_debug,
word_vector_name="glove_300d",
n_gram=1)
return {
"model_fn": GraphClusterModel,
"model_dict": {
"word_embedding": embedding_matrix,
"character_number": character_size,
"character_embedding_dim": 16,
"character_n_filters": 32,
"character_kernel_size": 5,
"character_padding": 2,
"hidden_size": 128, "conv_type": "depthwise_separable",
"resize_kernel_size": 7, "resize_pad_size": 3,
"n_encoder_conv_layer": 2, "encoder_kernel_size": 7, "encoder_padding": 3,
"n_self_attention_heads": 4, "route_number": 3, "n_capsules": 32,
"capsules_dim": 128, "n_compare_layer": 2, "n_layer_output_conv": 2,
"n_layer_output_feedforward": 3, "hidden_size_output_feedforward": 128, "n_classes": 2, "dropout": 0.2
},
"pre_process_module_fn": PreprocessWrapper,
"pre_process_module_dict": {
"pad_idx": word_pad_id,
"character_pad_idx": character_pad_id,
},
"data": [train, valid, test],
"batch_size": 16,
"train_loss": BCELoss,
"clip_norm": None,
"name": "sequence_graph_encoder_decoder_for_method_name",
"optimizer": optim.Adam,
"need_pad": True,
"optimizer_dict": {"betas": (0.8, 0.999), "weight_decay": 3e-7, },
"epcohes": 80,
"lr": 1e-2,
"evaluate_object_list": [SequenceBinaryClassExactMatch(gpu_index=get_gpu_index())],
}