temp-fix-copy-intx

rankfm/evaluation.py

@@ -20,7 +20,7 @@ def hit_rate(model, test_interactions, k=10, filter_previous=False):
  assert model.is_fit, "you must fit the model prior to evaluating hold-out metrics"
 
  # transform interactions into a user -> items dictionary
- test_user_items = pd.DataFrame(test_interactions, columns=['user_id', 'item_id'])
+ test_user_items = pd.DataFrame(test_interactions.values, columns=['user_id', 'item_id'])
  test_user_items = test_user_items.groupby('user_id')['item_id'].apply(set).to_dict()
  test_users = list(test_user_items.keys())
 
@@ -47,7 +47,7 @@ def reciprocal_rank(model, test_interactions, k=10, filter_previous=False):
  assert model.is_fit, "you must fit the model prior to evaluating hold-out metrics"
 
  # transform interactions into a user -> items dictionary
- test_user_items = pd.DataFrame(test_interactions, columns=['user_id', 'item_id'])
+ test_user_items = pd.DataFrame(test_interactions.values, columns=['user_id', 'item_id'])
  test_user_items = test_user_items.groupby('user_id')['item_id'].apply(set).to_dict()
  test_users = list(test_user_items.keys())
 
@@ -75,7 +75,7 @@ def discounted_cumulative_gain(model, test_interactions, k=10, filter_previous=F
  assert model.is_fit, "you must fit the model prior to evaluating hold-out metrics"
 
  # transform interactions into a user -> items dictionary
- test_user_items = pd.DataFrame(test_interactions, columns=['user_id', 'item_id'])
+ test_user_items = pd.DataFrame(test_interactions.values, columns=['user_id', 'item_id'])
  test_user_items = test_user_items.groupby('user_id')['item_id'].apply(set).to_dict()
  test_users = list(test_user_items.keys())
 
@@ -103,7 +103,7 @@ def precision(model, test_interactions, k=10, filter_previous=False):
  assert model.is_fit, "you must fit the model prior to evaluating hold-out metrics"
 
  # transform interactions into a user -> items dictionary
- test_user_items = pd.DataFrame(test_interactions, columns=['user_id', 'item_id'])
+ test_user_items = pd.DataFrame(test_interactions.values, columns=['user_id', 'item_id'])
  test_user_items = test_user_items.groupby('user_id')['item_id'].apply(set).to_dict()
  test_users = list(test_user_items.keys())
 
@@ -130,7 +130,7 @@ def recall(model, test_interactions, k=10, filter_previous=False):
  assert model.is_fit, "you must fit the model prior to evaluating hold-out metrics"
 
  # transform interactions into a user -> items dictionary
- test_user_items = pd.DataFrame(test_interactions, columns=['user_id', 'item_id'])
+ test_user_items = pd.DataFrame(test_interactions.values, columns=['user_id', 'item_id'])
  test_user_items = test_user_items.groupby('user_id')['item_id'].apply(set).to_dict()
  test_users = list(test_user_items.keys())
 
@@ -157,7 +157,7 @@ def diversity(model, test_interactions, k=10, filter_previous=False):
  assert model.is_fit, "you must fit the model prior to evaluating hold-out metrics"
 
  # get the unique set of test users
- test_user_items = pd.DataFrame(test_interactions, columns=['user_id', 'item_id'])
+ test_user_items = pd.DataFrame(test_interactions.values, columns=['user_id', 'item_id'])
  test_users = test_user_items['user_id'].unique()
 
  # generate topK recommendations for all test users also present in the training data

rankfm/rankfm.py

@@ -100,7 +100,7 @@ def _init_all(self, interactions, user_features=None, item_features=None):
  """
 
  # save the unique lists of users/items in terms of original identifiers
- interactions_df = pd.DataFrame(interactions, columns=['user_id', 'item_id'])
+ interactions_df = pd.DataFrame(interactions.values, columns=['user_id', 'item_id'])
  self.user_id = pd.Series(np.sort(np.unique(interactions_df['user_id'])))
  self.item_id = pd.Series(np.sort(np.unique(interactions_df['item_id'])))
 
@@ -139,7 +139,7 @@ def _init_interactions(self, interactions):
  # map the raw user/item identifiers to internal zero-based index positions
  # NOTE: any user/item pairs not found in the existing indexes will be dropped
 
- self.interactions = pd.DataFrame(interactions.copy(), columns=['user_id', 'item_id'])
+ self.interactions = pd.DataFrame(interactions.values.copy(), columns=['user_id', 'item_id'])
  self.interactions['user_id'] = self.interactions['user_id'].map(self.user_to_index).astype(np.int32)
  self.interactions['item_id'] = self.interactions['item_id'].map(self.item_to_index).astype(np.int32)
  self.interactions = self.interactions.rename({'user_id': 'user_idx', 'item_id': 'item_idx'}, axis=1).dropna().astype(np.int32)
@@ -289,7 +289,7 @@ def predict(self, pairs, cold_start='nan'):
  assert pairs.shape[1] == 2, "[pairs] should be: [user_id, item_id]"
  assert self.is_fit, "you must fit the model prior to generating predictions"
 
- pred_pairs = pd.DataFrame(pairs.copy(), columns=['user_id', 'item_id'])
+ pred_pairs = pd.DataFrame(pairs.values.copy(), columns=['user_id', 'item_id'])
  pred_pairs['user_id'] = pred_pairs['user_id'].map(self.user_to_index)
  pred_pairs['item_id'] = pred_pairs['item_id'].map(self.item_to_index)