from environments.random_search import RandomSearch
DATA_DIR = "/home/suching/reproducibility/data/sst/"
BOW_LINEAR = {
"CUDA_DEVICE": 0,
"USE_SPACY_TOKENIZER": 0,
"SEED": RandomSearch.random_integer(0, 100),
"DATA_DIR": DATA_DIR,
"THROTTLE": None,
"EMBEDDING": "BOW_COUNTS",
"ENCODER": None,
"LEARNING_RATE": RandomSearch.random_loguniform(1e-6, 1e-1),
"DROPOUT": RandomSearch.random_uniform(0, 0.5),
"BATCH_SIZE": 32,
}
# CLASSIFIER_SEARCH = {
# "CUDA_DEVICE": 0,
# "USE_SPACY_TOKENIZER": 0,
# "SEED": RandomSearch.random_integer(0, 100),
# "DATA_DIR": DATA_DIR,
# "THROTTLE": None,
# "EMBEDDING": "ELMO_TRANSFORMER",
# "ENCODER": "LSTM",
# "HIDDEN_SIZE": RandomSearch.random_integer(64, 512),
# "NUM_ENCODER_LAYERS": RandomSearch.random_integer(1, 3),
# "MAX_FILTER_SIZE": RandomSearch.random_integer(5, 10),
# "NUM_FILTERS": RandomSearch.random_integer(64, 512),
# "AGGREGATIONS": RandomSearch.random_choice("final_state", "maxpool", "meanpool", "attention"),
# "LEARNING_RATE": RandomSearch.random_loguniform(1e-6, 1e-1),
"LEARNING_RATE":
0.004,
"DROPOUT":
0.5,
"VAMPIRE_DIRECTORY":
os.environ.get("VAMPIRE_DIR", None),
"VAMPIRE_DIM":
os.environ.get("VAMPIRE_DIM", None),
"BATCH_SIZE":
32,
"NUM_ENCODER_LAYERS":
1,
"NUM_OUTPUT_LAYERS":
2,
"MAX_FILTER_SIZE":
RandomSearch.random_integer(3, 6),
"NUM_FILTERS":
RandomSearch.random_integer(64, 512),
"HIDDEN_SIZE":
RandomSearch.random_integer(64, 512),
"AGGREGATIONS":
RandomSearch.random_subset("maxpool", "meanpool", "attention",
"final_state"),
"MAX_CHARACTER_FILTER_SIZE":
RandomSearch.random_integer(3, 6),
"NUM_CHARACTER_FILTERS":
RandomSearch.random_integer(16, 64),
"CHARACTER_HIDDEN_SIZE":
RandomSearch.random_integer(16, 128),
"CHARACTER_EMBEDDING_DIM":
RandomSearch.random_integer(16, 64),
"DEV_PATH": os.environ["DATA_DIR"] + "/dev.jsonl",
"TEST_PATH": os.environ["DATA_DIR"] + "/test.jsonl",
"THROTTLE": os.environ.get("THROTTLE", None),
"USE_SPACY_TOKENIZER": 1,
"FREEZE_EMBEDDINGS": ["VAMPIRE"],
"EMBEDDINGS": ["VAMPIRE", "RANDOM"],
"ENCODER": "AVERAGE",
"EMBEDDING_DROPOUT": 0.5,
"LEARNING_RATE": 0.004,
"DROPOUT": 0.5,
"VAMPIRE_DIRECTORY": os.environ.get("VAMPIRE_DIR", None),
"VAMPIRE_DIM": os.environ.get("VAMPIRE_DIM", None),
"BATCH_SIZE": 32,
"NUM_ENCODER_LAYERS": 1,
"NUM_OUTPUT_LAYERS": 2,
"MAX_FILTER_SIZE": RandomSearch.random_integer(3, 6),
"NUM_FILTERS": RandomSearch.random_integer(64, 512),
"HIDDEN_SIZE": RandomSearch.random_integer(64, 512),
"AGGREGATIONS": RandomSearch.random_subset("maxpool", "meanpool", "attention", "final_state"),
"MAX_CHARACTER_FILTER_SIZE": RandomSearch.random_integer(3, 6),
"NUM_CHARACTER_FILTERS": RandomSearch.random_integer(16, 64),
"CHARACTER_HIDDEN_SIZE": RandomSearch.random_integer(16, 128),
"CHARACTER_EMBEDDING_DIM": RandomSearch.random_integer(16, 64),
"CHARACTER_ENCODER": RandomSearch.random_choice("LSTM", "CNN", "AVERAGE"),
"NUM_CHARACTER_ENCODER_LAYERS": RandomSearch.random_choice(1, 2),
}
VAMPIRE = {
"LAZY_DATASET_READER": os.environ.get("LAZY", 0),
"KL_ANNEALING": "linear",
"SIGMOID_WEIGHT_1": 0.25,