def test_embedding_layer():
random_state = np.random.RandomState(1999)
graph = OrderedDict()
max_index = 100
proj_dim = 12
fake_str_int = [[1, 5, 7, 1, 6, 2], [2, 3, 6, 2], [3, 3, 3, 3, 3, 3, 3]]
minibatch, mask = make_embedding_minibatch(
fake_str_int, slice(0, 3))
(emb_slices,), (emb_mask,) = add_embedding_datasets_to_graph(
[minibatch], [mask], "emb", graph)
emb = embedding_layer(emb_slices, max_index, proj_dim, graph,
'emb', random_state)
followup_dim = 17
proj = linear_layer([emb], graph, 'proj', followup_dim,
random_state=random_state)
f = theano.function(emb_slices, [proj], mode="FAST_COMPILE")
out, = f(*minibatch)
assert(out.shape[-1] == 17)
assert(out.shape[-2] == len(fake_str_int))
def test_embedding_layer():
random_state = np.random.RandomState(1999)
graph = OrderedDict()
max_index = 100
proj_dim = 12
fake_str_int = [[1, 5, 7, 1, 6, 2], [2, 3, 6, 2], [3, 3, 3, 3, 3, 3, 3]]
minibatch, mask = make_embedding_minibatch(fake_str_int, slice(0, 3))
(emb_slices, ), (emb_mask, ) = add_embedding_datasets_to_graph([minibatch],
[mask],
"emb",
graph)
emb = embedding_layer(emb_slices, max_index, proj_dim, graph, 'emb',
random_state)
followup_dim = 17
proj = linear_layer([emb], graph, 'proj', followup_dim, random_state)
f = theano.function(emb_slices, [proj], mode="FAST_COMPILE")
out, = f(*minibatch)
assert (out.shape[-1] == 17)
assert (out.shape[-2] == len(fake_str_int))
def test_make_embedding_minibatch():
fake_str_int = [[1, 5, 7, 1, 6, 2], [2, 3, 6, 2], [3, 3, 3, 3, 3, 3, 3]]
uniform_minibatch, mask = make_embedding_minibatch(
fake_str_int, slice(0, 2))
babi = fetch_babi(task_number=2)
X_story = babi["stories"]
X_query = babi["queries"]
y_answer = babi["target"]
train_indices = babi["train_indices"]
valid_indices = babi["valid_indices"]
vocab_size = babi["vocabulary_size"]
random_state = np.random.RandomState(1999)
graph = OrderedDict()
minibatch_size = 32
n_emb = 50
n_hid = 100
X_story_mb, X_story_mask = make_embedding_minibatch(
X_story, slice(0, minibatch_size))
X_query_mb, X_query_mask = make_embedding_minibatch(
X_query, slice(0, minibatch_size))
embedding_datasets = [X_story_mb, X_query_mb]
masks = [X_story_mask, X_query_mask]
r = add_embedding_datasets_to_graph(embedding_datasets, masks, "babi_data",
graph)
(X_story_syms, X_query_syms), (X_story_mask_sym, X_query_mask_sym) = r
y_sym = add_datasets_to_graph([y_answer], ["y"], graph)
l1_story = embedding_layer(X_story_syms, vocab_size, n_emb, graph, 'l1_story',
random_state)
masked_story = X_story_mask_sym.dimshuffle(0, 1, 'x') * l1_story
babi = fetch_babi(task_number=2)
X_story = babi["stories"]
X_query = babi["queries"]
y_answer = babi["target"]
train_indices = babi["train_indices"]
valid_indices = babi["valid_indices"]
vocab_size = babi["vocabulary_size"]
random_state = np.random.RandomState(1999)
graph = OrderedDict()
minibatch_size = 32
n_emb = 50
n_hid = 100
X_story_mb, X_story_mask = make_embedding_minibatch(
X_story, slice(0, minibatch_size))
X_query_mb, X_query_mask = make_embedding_minibatch(
X_query, slice(0, minibatch_size))
embedding_datasets = [X_story_mb, X_query_mb]
masks = [X_story_mask, X_query_mask]
r = add_embedding_datasets_to_graph(embedding_datasets, masks, "babi_data",
graph)
(X_story_syms, X_query_syms), (X_story_mask_sym, X_query_mask_sym) = r
y_sym = add_datasets_to_graph([y_answer], ["y"], graph)
l1_story = embedding_layer(X_story_syms, vocab_size, n_emb, graph, 'l1_story',
random_state=random_state)
masked_story = X_story_mask_sym.dimshuffle(0, 1, 'x') * l1_story
def test_make_embedding_minibatch():
fake_str_int = [[1, 5, 7, 1, 6, 2], [2, 3, 6, 2], [3, 3, 3, 3, 3, 3, 3]]
uniform_minibatch, mask = make_embedding_minibatch(fake_str_int, slice(0, 2))
