def get_tok2vec_kwargs():
# This actually creates models, so seems best to put it in a function.
return {
"embed":
MultiHashEmbed(
width=32,
rows=[500, 500, 500],
attrs=["NORM", "PREFIX", "SHAPE"],
include_static_vectors=False,
),
"encode":
MaxoutWindowEncoder(width=32, depth=2, maxout_pieces=2, window_size=1),
}
def my_parser():
tok2vec = build_Tok2Vec_model(
MultiHashEmbed(
width=321,
attrs=["LOWER", "SHAPE"],
rows=[5432, 5432],
include_static_vectors=False,
),
MaxoutWindowEncoder(width=321, window_size=3, maxout_pieces=4, depth=2),
)
parser = build_tb_parser_model(
tok2vec=tok2vec,
state_type="parser",
extra_state_tokens=True,
hidden_width=65,
maxout_pieces=5,
use_upper=True,
)
return parser
def test_multi_hash_embed():
embed = MultiHashEmbed(
width=32,
rows=[500, 500, 500],
attrs=["NORM", "PREFIX", "SHAPE"],
include_static_vectors=False,
)
hash_embeds = [node for node in embed.walk() if node.name == "hashembed"]
assert len(hash_embeds) == 3
# Check they look at different columns.
assert list(sorted(he.attrs["column"] for he in hash_embeds)) == [0, 1, 2]
# Check they use different seeds
assert len(set(he.attrs["seed"] for he in hash_embeds)) == 3
# Check they all have the same number of rows
assert [he.get_dim("nV") for he in hash_embeds] == [500, 500, 500]
# Now try with different row factors
embed = MultiHashEmbed(
width=32,
rows=[1000, 50, 250],
attrs=["NORM", "PREFIX", "SHAPE"],
include_static_vectors=False,
)
hash_embeds = [node for node in embed.walk() if node.name == "hashembed"]
assert [he.get_dim("nV") for he in hash_embeds] == [1000, 50, 250]