def a_ndcg(true,
pred,
aspects,
k=10,
alpha=0.5,
user_col='user_id',
item_col='item_id'):
"""Measures redundancy-aware quality and diversity."""
if type(true) is pd.DataFrame:
true = pandas_to_dict(true, user_col, item_col)
if type(pred) is pd.DataFrame:
pred = pandas_to_dict(pred, user_col, item_col)
return user_mean_sub(_a_ndcg, true, pred, aspects, k, alpha)
def coverage(items, recs, k=None, user_col='user_id', item_col='item_id'):
"""What percentage of items appears in recommendations?
Args:
items: list of unique item ids
recs: dict of recommendations
k: topk items to use from recs
Returns: float
"""
if type(recs) is pd.DataFrame:
recs = pandas_to_dict(recs, user_col, item_col)
topk = list(set(flatten_list(top_k(recs, k).values())))
return np.isin(items, topk).mean()
def popularity(log, pred, k=10, user_col='user_id', item_col='item_id'):
"""
Mean popularity of recommendations.
Args:
log: pandas DataFrame with interactions
pred: pandas DataFrame with recommendations
k: top k items to use from recs
user_col: column name for user ids
item_col: column name for item ids
"""
if type(pred) is pd.DataFrame:
pred = pandas_to_dict(pred, user_col, item_col)
scores = item_pop(log, user_col, item_col)
return user_apply(_popularity, scores, pred, k, 0)
def surprisal(log, pred, k=10, user_col='user_id', item_col='item_id'):
if type(pred) is pd.DataFrame:
pred = pandas_to_dict(pred, user_col, item_col)
scores = -np.log2(item_pop(log, user_col, item_col))
fill = np.log2(log[user_col].nunique())
return user_apply(_popularity, scores, pred, k, fill)