Implementation:Recommenders team Recommenders SARSingleNode Init

Knowledge Sources	Recommenders
Domains	Recommender Systems, Collaborative Filtering, Item Similarity
Last Updated	2026-02-10 00:00 GMT

Overview

Concrete tool for configuring and initializing a SAR (Smart Adaptive Recommendations) single-node collaborative filtering model provided by the recommenders library.

Description

The SARSingleNode.__init__ method initializes all model parameters and internal state for the SAR algorithm. It validates the chosen similarity type against the set of supported metrics (cooccurrence, cosine, inclusion index, jaccard, lexicographers mutual information, lift, mutual information), converts the time-decay coefficient from days to seconds, and sets up internal data structures (index mappings, affinity matrices, similarity matrices) as None, to be populated during the fit step. It also validates that the threshold is positive and configures whether score normalization is enabled.

Usage

Import and instantiate this class at the model configuration stage of your pipeline, before calling fit to train on data. Choose the similarity type and time-decay settings based on your dataset characteristics and evaluation goals.

Code Reference

Source Location

• Repository: recommenders
• File: recommenders/models/sar/sar_singlenode.py
• Lines: L43-L135

Signature

class SARSingleNode:
    def __init__(
        self,
        col_user=DEFAULT_USER_COL,
        col_item=DEFAULT_ITEM_COL,
        col_rating=DEFAULT_RATING_COL,
        col_timestamp=DEFAULT_TIMESTAMP_COL,
        col_prediction=DEFAULT_PREDICTION_COL,
        similarity_type="jaccard",
        time_decay_coefficient=30,
        time_now=None,
        timedecay_formula=False,
        threshold=1,
        normalize=False,
    )

Import

from recommenders.models.sar.sar_singlenode import SARSingleNode

I/O Contract

Inputs

Name	Type	Required	Description
col_user	str	No (default: DEFAULT_USER_COL)	Column name for user IDs in the input DataFrame.
col_item	str	No (default: DEFAULT_ITEM_COL)	Column name for item IDs in the input DataFrame.
col_rating	str	No (default: DEFAULT_RATING_COL)	Column name for rating values in the input DataFrame.
col_timestamp	str	No (default: DEFAULT_TIMESTAMP_COL)	Column name for timestamps. Required only when timedecay_formula is True.
col_prediction	str	No (default: DEFAULT_PREDICTION_COL)	Column name used for prediction output.
similarity_type	str	No (default: "jaccard")	Item-item similarity metric. One of: "cooccurrence", "cosine", "inclusion index", "jaccard", "lexicographers mutual information", "lift", "mutual information".
time_decay_coefficient	float	No (default: 30)	Half-life in days for time-decay weighting. Internally converted to seconds.
time_now	int or None	No (default: None)	Reference timestamp for time-decay calculation. If None, the maximum timestamp in the data is used during fit.
timedecay_formula	bool	No (default: False)	Whether to apply exponential time-decay weighting to ratings.
threshold	int	No (default: 1)	Minimum item-item co-occurrence count. Pairs below this threshold are set to zero. Must be >= 1.
normalize	bool	No (default: False)	Whether to normalize prediction scores to the scale of the original ratings.

Outputs

Name	Type	Description
return	SARSingleNode	Initialized SAR model instance with internal state set to None (user_affinity, item_similarity, item_frequencies, user2index, item2index, etc.), ready for training via fit().

Usage Examples

Basic Usage

from recommenders.models.sar.sar_singlenode import SARSingleNode

# Initialize with default Jaccard similarity
model = SARSingleNode(
    col_user="userID",
    col_item="itemID",
    col_rating="rating",
    col_timestamp="timestamp",
    similarity_type="jaccard",
)

# Initialize with cosine similarity and time decay
model = SARSingleNode(
    similarity_type="cosine",
    timedecay_formula=True,
    time_decay_coefficient=15,
)

# Initialize with lift similarity and score normalization
model = SARSingleNode(
    similarity_type="lift",
    normalize=True,
    threshold=3,
)

Dependencies

• numpy - Numerical operations and array manipulation
• pandas - DataFrame handling for input data
• scipy.sparse - Sparse matrix representations for affinity and similarity matrices

Related Pages

Implements Principle

• Principle:Recommenders_team_Recommenders_SAR_Algorithm

Requires Environment

• Environment:Recommenders_team_Recommenders_Python_Core_Dependencies