def _calc_embeddings(
self, model_traces: List[List[str]], real_traces: List[List[str]]
) -> Tuple[np.ndarray, np.ndarray, Any]:
"""Calculates the embeddings of the traces.
:param model_traces: The traces coming from the model.
:param real_traces: The traces coming from the real log.
:return: Dicts for the model and real log contains
index of activities and its frequencies
and a distance matrix for Euclidean distances of
every two actives in all traces.
"""
emb_gen = ActivityEmbeddingGenerator(
model_traces + real_traces,
act2vec_windows_size=self.window_size,
num_ns=self.num_negative,
auto_train=False,
num_epochs=self.num_epochs,
batch_size=self.batch_size,
embedding_size=self.embedding_size,
)
# start to train the models
emb_gen.start_training()
model_embedding, real_embedding, context = emb_gen.get_activity_embedding(
model_traces, real_traces, norm=True)
dist_matrix = calc_euclidean(context)
model_embedding = _calc_d(model_embedding, len(dist_matrix))
real_embedding = _calc_d(real_embedding, len(dist_matrix))
# return frequency tables for the model log and the real log
# and an embedding lookup table
return model_embedding, real_embedding, dist_matrix
def test_wmd():
"""Is the wmd calculated correctly?"""
model_embedding = [{0: 1, 1: 1}]
real_embedding = [{1: 1}]
context = np.array([[1, 4], [5, 1]])
# calculate Euclidean distance matrix
distance_matrix = calc_euclidean(context)
# calc d for embeddings
vocab_len = len(context)
d_model = _calc_d(model_embedding, vocab_len)
d_real = _calc_d(real_embedding, vocab_len)
assert calc_wmd(d_model[0], d_real[0],
distance_matrix) == pytest.approx(2.5)
def main():
# create some embeddings as example
# (int, int, ...): int =
# embedding of a activity: count of this activity within a trace
model_embedding = [{0: 3, 1: 1, 2: 2}]
real_embedding = [{0: 2}]
context = np.array([[0.4, 0.3], [0.2, 0.6], [0.5, 0.9]])
# calculate Euclidean distance matrix
distance_matrix = calc_euclidean(context)
# calc d for embeddings
vocab_len = len(context)
d_model = _calc_d(model_embedding, vocab_len)
d_real = _calc_d(real_embedding, vocab_len)
# calculate WMD between these two traces
print("WMD: ", calc_wmd(d_model[0], d_real[0], distance_matrix))
print("ICT: ", calc_ict(d_model[0], d_real[0], distance_matrix))