def eval_fit_recsys(cls, recsys: RecSys, split_list: List[Split],
test_items_list: List[pd.DataFrame]):
"""
Method which is usually called by the 'PredictionCalculator' module that generates recommendation lists. For
every user, items that will be ranked are specified by the 'test_items_list' parameter.
Rankings generated will be stored into a class attribute (rank_truth_list), which is a list that contains
Split objects: every object has two DataFrames, the first one has recommendation lists for every user,
the second one has the 'ground truth' for every user.
If the class attribute is non-empty, then the 'AlreadyFittedRecSys' exception is raised, so remember to
clean the class attribute by calling the private method '_clean_pred_truth_list(...)' upon every new evaluation
Args:
recsys (RecSys): Recommender system which will generate predictions that will later be evaluated
split_list (List[Split]): List of Split objects where every Split contains two DataFrames, the first has
the 'train set' for every user, the second has the 'test set' for every user
test_items_list (List[pd.DataFrame]): List of DataFrames, one for every Split object inside the split_list
parameter, where every DataFrame contains for every user the list of items that must be ranked
Raises:
AlreadyFittedRecSys exception when the class attribute 'rank_truth_list' is non-empty, meaning that
recommendation lists are already been calculated
"""
if len(cls.rank_truth_list) != 0:
raise AlreadyFittedRecSys
for counter, (split, test_items_frame) in enumerate(zip(
split_list, test_items_list),
start=1):
train = split.train
test = split.test
rank_truth = Split()
rank_truth.truth = test
frame_to_concat = []
user_list_to_fit = set(train.from_id)
for user in progbar(
user_list_to_fit,
prefix='Calculating rank for user {} - split {}'.format(
'{}', counter),
substitute_with_current=True):
user_ratings_train = train.loc[train['from_id'] == user]
test_items = list(
test_items_frame.query('from_id == @user').to_id)
result = recsys._eval_fit_rank(user_ratings_train, test_items)
frame_to_concat.append(result)
rank_truth.pred = pd.concat(frame_to_concat)
cls.rank_truth_list.append(rank_truth)
def eval_fit_recsys(cls, recsys: RecSys, split_list: List[Split],
test_items_list: List[pd.DataFrame]):
if len(cls.rank_truth_list) != 0:
raise AlreadyFittedRecSys
for counter, (split, test_items_frame) in enumerate(zip(
split_list, test_items_list),
start=1):
train = split.train
test = split.test
rank_truth = Split()
rank_truth.truth = test
frame_to_concat = []
user_list_to_fit = set(train.from_id)
for user in progbar(
user_list_to_fit,
prefix='Calculating rank for user {} - split {}'.format(
'{}', counter),
substitute_with_current=True):
user_ratings_train = train.loc[train['from_id'] == user]
test_items = list(
test_items_frame.query('from_id == @user').to_id)
result = recsys._eval_fit_rank(user_ratings_train, test_items)
frame_to_concat.append(result)
rank_truth.pred = pd.concat(frame_to_concat)
cls.rank_truth_list.append(rank_truth)