Skip to content

Practicus AI Pattern Mining Documentation

This example documents a set of built-in pattern mining snippets that can be used in the platform:

  • apriori_association_rules
  • sequence_markov_transitions
  • carma_windowed_association_rules
  • fpgrowth_association_rules
  • sequence_mining_contiguous

1. apriori_association_rules

Purpose:

Finds association rules (e.g. Milk -> Bread) from transactional data using the Apriori algorithm. This is a classic market basket analysis / birliktelik analizi method.

Expected input DataFrame columns:

  • transaction_id_col: column containing the transaction or basket ID (e.g., OrderID, BasketId).
  • item_col: column containing the item / product identifier (e.g., ProductName, ItemCode).

Each row represents one (transaction, item) pair.

Parameters:

  • df: the input DataFrame.
  • transaction_id_col (str): name of the transaction ID column.
  • item_col (str): name of the item column.
  • min_support (float, default 0.05): minimum fraction of transactions that must contain an itemset.
  • min_confidence (float, default 0.6): minimum confidence for a rule A -> B.
  • min_lift (float or None, default None): if set, rules with lift below this threshold are removed.
  • max_rules (int or None, default 100): maximum number of rules to return.

Output:

A new DataFrame with one row per rule, typically with columns like:

  • antecedent: left-hand side itemset as a string (e.g., "Milk", "Milk, Bread").
  • consequent: right-hand side itemset as a string.
  • support: support of the full rule (fraction of all transactions).
  • confidence: confidence of the rule.
  • lift: lift of the rule.

The platform can store this output as a new worksheet, e.g. Apriori_Rules_<original_table>.

2. carma_windowed_association_rules

Purpose:

Implements a CARMA-style incremental association mining by running Apriori on a sliding window of the most recent transactions. This approximates streaming / real-time association rule mining.

Expected input DataFrame columns:

  • transaction_id_col: transaction or basket ID.
  • item_col: item / product identifier.
  • timestamp_col (optional): timestamp or order column used to determine which transactions are the most recent.

Parameters:

  • df: the input DataFrame.
  • transaction_id_col (str): name of the transaction ID column.
  • item_col (str): name of the item column.
  • timestamp_col (str or None, default None): name of the timestamp column. If None, the existing row order is used.
  • window_size_transactions (int, default 10000): number of most recent distinct transactions to include in the analysis.
  • min_support (float, default 0.05): minimum support within the window.
  • min_confidence (float, default 0.6): minimum confidence.
  • min_lift (float or None, default None): optional minimum lift.
  • max_rules (int or None, default 100): limit on number of rules returned.

Output:

Same shape as apriori_association_rules: a table of rules with antecedent, consequent, support, confidence and lift. The difference is that all metrics are computed only on the most recent window of transactions.

3. sequence_mining_contiguous

Purpose:

Performs Sequence mining by finding frequent contiguous subsequences of events within ordered sequences.

Examples: - ViewProduct -> AddToCart -> Purchase - Login -> Search -> Logout

This satisfies the "Sequence" requirement for ardışıklık tespiti.

Expected input DataFrame columns:

  • sequence_id_col: identifier of a sequence (e.g., SessionID, CustomerID, CaseId).
  • event_col: event or action name at each step (e.g., "View", "AddToCart", "Purchase").
  • order_by_col (optional): column used to sort events inside each sequence (e.g., a timestamp or step index).

Each row represents one event in a sequence.

Parameters:

  • df: the input DataFrame.
  • sequence_id_col (str): name of the sequence ID column.
  • event_col (str): name of the event/action column.
  • order_by_col (str or None, default None): name of the column used to sort events.
  • min_support (float, default 0.02): minimum fraction of sequences that must contain a pattern.
  • max_pattern_length (int, default 5): maximum length of a pattern (e.g., 3 for A -> B -> C).
  • top_k (int or None, default 100): maximum number of patterns to return.

Output:

A DataFrame where each row is a discovered pattern, with columns such as:

  • sequence: the pattern as a string (e.g., "ViewProduct -> AddToCart -> Purchase").
  • support: number of sequences in which this pattern appears.
  • support_fraction: support divided by total number of sequences.
  • length: number of events in the pattern.

4. fpgrowth_association_rules

Purpose:

Performs association rule mining using the FP-Growth algorithm instead of Apriori. This demonstrates support for multiple birliktelik analizi algorithms and can be more efficient on large datasets.

Expected input DataFrame columns:

  • transaction_id_col: transaction or basket ID.
  • item_col: item / product identifier.

Parameters:

  • df: the input DataFrame.
  • transaction_id_col (str): name of the transaction ID column.
  • item_col (str): name of the item column.
  • min_support (float, default 0.05): minimum support threshold for frequent itemsets.
  • min_confidence (float, default 0.6): minimum confidence for rules.
  • min_lift (float or None, default None): optional minimum lift.
  • max_rules (int or None, default 100): maximum number of rules to return.

Output:

Same structure as the Apriori-based snippet: a rule table with antecedent, consequent, support, confidence and lift.

5. sequence_markov_transitions

Purpose:

Computes first-order Markov transition probabilities between events in sequences. This focuses on pairwise transitions rather than longer patterns.

Example output: - ViewProduct -> AddToCart with probability 0.4 - AddToCart -> Purchase with probability 0.7

This is another way to perform ardışıklık analizi.

Expected input DataFrame columns:

  • sequence_id_col: identifier of a sequence (e.g., SessionID, CustomerID).
  • event_col: event or action at each step.
  • order_by_col (optional): column used to sort events inside each sequence.

Parameters:

  • df: the input DataFrame.
  • sequence_id_col (str): name of the sequence ID column.
  • event_col (str): name of the event column.
  • order_by_col (str or None, default None): name of the ordering column.
  • min_transition_count (int, default 1): minimum number of occurrences of a transition to be included.
  • min_transition_probability (float, default 0.0): minimum P(target | source) to be included.

Output:

A DataFrame with one row per transition, typically with columns:

  • source_event: previous event (e.g., "ViewProduct").
  • target_event: next event (e.g., "AddToCart").
  • count: number of times this transition appears.
  • probability: estimated probability P(target_event | source_event).

6. Example Usage With a Simple CSV

All of these snippets can be demonstrated on a small CSV file with columns:

  • transaction_id: transaction / basket identifier.
  • item: product name.
  • sequence_id: session or journey identifier.
  • event: action taken (view, add to cart, purchase, etc.).
  • timestamp: event time, used to order events.

Example:

transaction_id,item,sequence_id,event,timestamp
T1,Milk,S1,ViewProduct,2025-01-01T09:00:00
T1,Bread,S1,AddToCart,2025-01-01T09:01:00
T1,Bread,S1,Purchase,2025-01-01T09:02:00
T2,Bread,S2,ViewProduct,2025-01-01T09:03:00
T2,Cheese,S2,AddToCart,2025-01-01T09:04:00
T2,Cheese,S2,Purchase,2025-01-01T09:05:00
T3,Cheese,S3,ViewProduct,2025-01-01T09:06:00
T3,Apples,S3,AddToCart,2025-01-01T09:07:00
T3,Apples,S3,Purchase,2025-01-01T09:08:00
T4,Apples,S4,ViewProduct,2025-01-01T09:09:00
T4,Oranges,S4,AddToCart,2025-01-01T09:10:00

With this CSV loaded into a DataFrame df, typical parameter mappings are:

apriori_association_rules

  • transaction_id_col = "transaction_id"
  • item_col = "item"

carma_windowed_association_rules

  • transaction_id_col = "transaction_id"
  • item_col = "item"
  • timestamp_col = "timestamp"

fpgrowth_association_rules

  • transaction_id_col = "transaction_id"
  • item_col = "item"

sequence_mining_contiguous

  • sequence_id_col = "sequence_id"
  • event_col = "event"
  • order_by_col = "timestamp"

sequence_markov_transitions

  • sequence_id_col = "sequence_id"
  • event_col = "event"
  • order_by_col = "timestamp"

The platform can expose these parameters in the UI and run the corresponding snippet on the selected table.

Previous: Spark Tutorial | Next: Model Tracking > Experiment Tracking > Experiment Tracking Logging