def test_recommendation_with_testfm(self):
"""
[recommendation.api.GetRecommendation] Test recommendation with testfm
"""
data = np.array(zip(*map(lambda x: (x["user_id"]-1, x["item_id"]-1, 1.),
Inventory.objects.all().values("user_id", "item_id"))), dtype=np.float32)
users, items = zip(*Inventory.objects.all().values_list("user_id", "item_id"))
df = pd.DataFrame({"user": pd.Series(users), "item": pd.Series(items)}, dtype=np.float32)
evaluator = Evaluator(use_multi_threading=False)
tensor = TensorCoFi.get_model_from_cache()
tfm_tensor = PyTensorCoFi()
tfm_tensor.data_map = tensor.data_map
tfm_tensor.users_size = lambda: tensor.users_size()
tfm_tensor.items_size = lambda: tensor.items_size()
tfm_tensor.get_score = lambda user, item: \
np.dot(tfm_tensor.factors[0][tfm_tensor.data_map[tfm_tensor.get_user_column()][user]],
tfm_tensor.factors[1][tfm_tensor.data_map[tfm_tensor.get_item_column()][item]].transpose())
tfm_tensor.train(data.transpose())
items = df.item.unique()
t = evaluator.evaluate_model(tensor, df, all_items=items, non_relevant_count=100)
tfm = evaluator.evaluate_model(tfm_tensor, df, all_items=items, non_relevant_count=100)
assert abs(t[0] - tfm[0]) < 0.15, \
"Difference between testfm implementation and frappe is to high (%f, %f)" % (t[0], tfm[0])
