parallelFPGrowth.py

# Parallel FPGrowth is one of the fundamental algorithm to discover frequent patterns in a transactional database. It stores the database in compressed fp-tree decreasing the memory usage and extracts the patterns from tree.It employs downward closure property to  reduce the search space effectively.
#
#  **Importing this algorithm into a python program**
# ----------------------------------------------------
#
#             import PAMI.frequentPattern.pyspark.parallelFPGrowth as alg
#
#             obj = alg.parallelFPGrowth(iFile, minSup, numWorkers)
#
#             obj.mine()
#
#             frequentPatterns = obj.getPatterns()
#
#             print("Total number of Frequent Patterns:", len(frequentPatterns))
#
#             obj.save(oFile)
#
#             Df = obj.getPatternInDataFrame()
#
#             memUSS = obj.getMemoryUSS()
#
#             print("Total Memory in USS:", memUSS)
#
#             memRSS = obj.getMemoryRSS()
#
#             print("Total Memory in RSS", memRSS)
#
#             run = obj.getRuntime()
#
#             print("Total ExecutionTime in seconds:", run)
#






__copyright__ = """
Copyright (C)  2021 Rage Uday Kiran

     This program is free software: you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
     the Free Software Foundation, either version 3 of the License, or
     (at your option) any later version.

     This program is distributed in the hope that it will be useful,
     but WITHOUT ANY WARRANTY; without even the implied warranty of
     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
     GNU General Public License for more details.

     You should have received a copy of the GNU General Public License
     along with this program.  If not, see <https://www.gnu.org/licenses/>.
"""



# from pyspark import SparkConf, SparkContext
from collections import defaultdict
from PAMI.frequentPattern.pyspark import abstract as _ab
from operator import add
from pyspark import SparkConf as _SparkConf, SparkContext as _SparkContext
from deprecated import deprecated


class Node:
    """
    :Attribute:
        item : int
            Storing item of a node
        count : int
            To maintain the support count of node
        children : dict
            To maintain the children of node
        prefix : list
            To maintain the prefix of node
    """
    def __init__(self, item, prefix):
        self.item = item
        self.count = 0
        self.children = {}
        self.prefix = prefix


class Tree:
    """
    :Attribute:
        root : Node
            The first node of the tree set to Null
        nodeLink : dict
            Store nodes that have the same item
    :Methods:

        addTransaction(transaction, count)
            Create tree from transaction and count
        addNodeToNodeLink(node)
            Add nodes that have the same item to self.nodeLink
        generateConditionalTree(item)
            Create conditional pattern base of item
    """
    def __init__(self):
        self.root = Node(None, [])
        self.nodeLink = {}
        self.itemCount = defaultdict(int)


    def addTransaction(self, transaction, count):
        """
        Add transaction to tree
        :param transaction: Transaction to add
        :type transaction: list
        :param count: Number of nodes
        :type count: int
        """
        current = self.root
        for item in transaction:
            if item not in current.children:
                current.children[item] = Node(item, transaction[0:transaction.index(item)])
                current.children[item].count += count
                self.addNodeToNodeLink(current.children[item])
            else:
                current.children[item].count += count
            self.itemCount[item] += count
            current = current.children[item]


    def addNodeToNodeLink(self, node):
        """
        Add node to self.nodeLink
        :param node: Node to add
        :type node: Node
        """
        if node.item not in self.nodeLink:
            self.nodeLink[node.item] = [node]
        else:
            self.nodeLink[node.item].append(node)


    def generateConditionalTree(self, item):
        """
        Generate conditional tree based on item
        :param item: Item to be considered as a condition
        :type item: str or int
        :return: Tree
        """
        tree = Tree()
        for node in self.nodeLink[item]:
            tree.addTransaction(node.prefix, node.count)
        return tree




class parallelFPGrowth(_ab._frequentPatterns):
    """

    :Description: Parallel FPGrowth is one of the fundamental algorithm to discover frequent patterns in a transactional database. It stores the database in compressed fp-tree decreasing the memory usage and extracts the patterns from tree.It employs downward closure property to  reduce the search space effectively.

    :Reference: Li, Haoyuan et al. “Pfp: parallel fp-growth for query recommendation.” ACM Conference on Recommender Systems (2008).

    :param  iFile: str :
                   Name of the Input file to mine complete set of frequent patterns
    :param  oFile: str :
                   Name of the output file to store complete set of frequent patterns
    :param  minSup: int :
                   The user can specify minSup either in count or proportion of database size. If the program detects the data type of minSup is integer, then it treats minSup is expressed in count. Otherwise, it will be treated as float.
    :param  sep: str :
                   This variable is used to distinguish items from one another in a transaction. The default seperator is tab space. However, the users can override their default separator.
    :param  numPartitions: int :
                   The number of partitions. On each worker node, an executor process is started and this process performs processing.The processing unit of worker node is partition


    :Attributes:

        startTime : float
          To record the start time of the mining process

        endTime : float
          To record the completion time of the mining process

        finalPatterns : dict
          Storing the complete set of patterns in a dictionary variable

        memoryUSS : float
          To store the total amount of USS memory consumed by the program

        memoryRSS : float
          To store the total amount of RSS memory consumed by the program

        lno : int
                the number of transactions
    
    **Methods to execute code on terminal**
    ----------------------------------------------------

    .. code-block:: console

      Format:

      (.venv) $ python3 parallelFPGrowth.py <inputFile> <outputFile> <minSup> <numWorkers>

      Example Usage:

      (.venv) $ python3 parallelFPGrowth.py sampleDB.txt patterns.txt 10.0 3

    .. note:: minSup will be considered in percentage of database transactions


    **Importing this algorithm into a python program**
    ----------------------------------------------------
    .. code-block:: python
    
                    import PAMI.frequentPattern.pyspark.parallelFPGrowth as alg
    
                    obj = alg.parallelFPGrowth(iFile, minSup, numWorkers)
    
                    obj.mine()
    
                    frequentPatterns = obj.getPatterns()
    
                    print("Total number of Frequent Patterns:", len(frequentPatterns))
    
                    obj.save(oFile)
    
                    Df = obj.getPatternInDataFrame()
    
                    memUSS = obj.getMemoryUSS()
    
                    print("Total Memory in USS:", memUSS)
    
                    memRSS = obj.getMemoryRSS()
    
                    print("Total Memory in RSS", memRSS)
    
                    run = obj.getRuntime()
    
                    print("Total ExecutionTime in seconds:", run)
    
    
    **Credits:**
    ----------------------------------------------------
    
             The complete program was written by Yudai Masu  under the supervision of Professor Rage Uday Kiran.

    """

    _minSup = float()
    _numPartitions = int()
    _startTime = float()
    _endTime = float()
    _finalPatterns = dict()
    _FPList = list()
    _iFile = " "
    _oFile = " "
    _sep = " "
    _memoryUSS = float()
    _memoryRSS = float()
    _lno = int()


    def __init__(self, iFile, minSup, numWorkers, sep='\t'):
        super().__init__(iFile, minSup, int(numWorkers), sep)

    @deprecated("It is recommended to use 'mine()' instead of 'startMine()' for mining process. Starting from January 2025, 'startMine()' will be completely terminated.")
    def startMine(self):
        """
        Frequent pattern mining process will start from here
        """

        self._startTime = _ab._time.time()

        conf = _SparkConf().setAppName("Parallel FPGrowth").setMaster("local[*]")
        sc = _SparkContext(conf=conf)

        rdd = sc.textFile(self._iFile, self._numPartitions)\
            .map(lambda x: x.rstrip().split('\t'))\
            .persist()

        self._lno = rdd.count()
        self._minSup = self._convert(self._minSup)

        freqItems = rdd.flatMap(lambda trans: [(item, 1) for item in trans])\
            .reduceByKey(add)\
            .filter(lambda x: x[1] >= self._minSup)\
            .sortBy(lambda x: x[1], ascending=False)\
            .collect()
        self._finalPatterns = dict(freqItems)
        self._FPList = [x[0] for x in freqItems]
        rank = dict([(item, index) for (index, item) in enumerate(self._FPList)])

        workByPartition = rdd.flatMap(lambda x: self.genCondTransaction(x, rank)).groupByKey()

        trees = workByPartition.foldByKey(Tree(), lambda tree, data: self.buildTree(tree, data))
        freqPatterns = trees.flatMap(lambda tree_tuple: self.genAllFrequentPatterns(tree_tuple))
        result = freqPatterns.map(lambda ranks_count: (tuple([self._FPList[z] for z in ranks_count[0]]), ranks_count[1]))\
            .collect()

        self._finalPatterns.update(dict(result))

        self._endTime = _ab._time.time()
        process = _ab._psutil.Process(_ab._os.getpid())
        self._memoryUSS = process.memory_full_info().uss
        self._memoryRSS = process.memory_info().rss
        sc.stop()

        print("Frequent patterns were generated successfully using Parallel FPGrowth algorithm")

    def mine(self):
        """
        Frequent pattern mining process will start from here
        """

        self._startTime = _ab._time.time()

        conf = _SparkConf().setAppName("Parallel FPGrowth").setMaster("local[*]")
        sc = _SparkContext(conf=conf)

        rdd = sc.textFile(self._iFile, self._numPartitions)\
            .map(lambda x: x.rstrip().split('\t'))\
            .persist()

        self._lno = rdd.count()
        self._minSup = self._convert(self._minSup)

        freqItems = rdd.flatMap(lambda trans: [(item, 1) for item in trans])\
            .reduceByKey(add)\
            .filter(lambda x: x[1] >= self._minSup)\
            .sortBy(lambda x: x[1], ascending=False)\
            .collect()
        self._finalPatterns = dict(freqItems)
        self._FPList = [x[0] for x in freqItems]
        rank = dict([(item, index) for (index, item) in enumerate(self._FPList)])

        workByPartition = rdd.flatMap(lambda x: self.genCondTransaction(x, rank)).groupByKey()

        trees = workByPartition.foldByKey(Tree(), lambda tree, data: self.buildTree(tree, data))
        freqPatterns = trees.flatMap(lambda tree_tuple: self.genAllFrequentPatterns(tree_tuple))
        result = freqPatterns.map(lambda ranks_count: (tuple([self._FPList[z] for z in ranks_count[0]]), ranks_count[1]))\
            .collect()

        self._finalPatterns.update(dict(result))

        self._endTime = _ab._time.time()
        process = _ab._psutil.Process(_ab._os.getpid())
        self._memoryUSS = process.memory_full_info().uss
        self._memoryRSS = process.memory_info().rss
        sc.stop()

        print("Frequent patterns were generated successfully using Parallel FPGrowth algorithm")


    def getPartitionId(self, value):
        """
        Get partition id of item
        :param value: value to get partition id
        :type value: int
        :return: integer
        """
        return value % self._numPartitions

    def genCondTransaction(self, trans, rank):
        """
        Generate conditional transactions from transaction
        :param trans : transactions to generate conditional transactions
        :type trans: list
        :param rank: rank of conditional transactions to generate conditional transactions
        :type rank: dict
        :return: list
        """
        newTrans = [rank[item] for item in trans if item in rank.keys()]
        newTrans = sorted(newTrans)
        condTrans = {}
        for i in reversed(newTrans):
            partition = self.getPartitionId(i)
            if partition not in condTrans:
                condTrans[partition] = newTrans[:newTrans.index(i)+1]
        return [x for x in condTrans.items()]

    @staticmethod
    def buildTree(tree, data):
        """
        Build tree from data
        :param tree: tree to build
        :type tree: Tree
        :param data: data to build
        :type data: list
        :return: tree
        """
        for trans in data:
            tree.addTransaction(trans, 1)
        return tree

    def genAllFrequentPatterns(self, tree_tuple):
        """
        Generate all frequent patterns
        :param tree_tuple: (partition id, tree)
        :type tree_tuple: tuple
        :return: dict
        """
        itemList = sorted(tree_tuple[1].itemCount.items(), key=lambda x: x[1])
        itemList = [x[0] for x in itemList]
        freqPatterns = {}
        for item in itemList:
            if self.getPartitionId(item) == tree_tuple[0]:
                freqPatterns.update(self.genFreqPatterns(item, [item], tree_tuple[1]))
        return freqPatterns.items()

    def genFreqPatterns(self, item, prefix, tree):
        """
        Generate new frequent patterns based on item.
        :param item: item
        :type item: int
        :param prefix: prefix frequent pattern
        :type prefix: str
        :param tree: tree to generate patterns
        :type tree: Tree
        :return:
        """
        condTree = tree.generateConditionalTree(item)
        freqPatterns = {}
        freqItems = {}
        for i in condTree.nodeLink.keys():
            freqItems[i] = 0
            for node in condTree.nodeLink[i]:
                freqItems[i] += node.count
        freqItems = {key: value for key, value in freqItems.items() if value >= self._minSup}

        for i in freqItems:
            pattern = prefix + [i]
            freqPatterns[tuple(pattern)] = freqItems[i]
            freqPatterns.update(self.genFreqPatterns(i, pattern, condTree))
        return freqPatterns

    def getMemoryUSS(self):
        """
        Total amount of USS memory consumed by the mining process will be retrieved from this function
        :return: returning USS memory consumed by the mining process
        :rtype: float
        """

        return self._memoryUSS

    def getMemoryRSS(self):
        """
        Total amount of RSS memory consumed by the mining process will be retrieved from this function
        :return: returning RSS memory consumed by the mining process
        :rtype: float
        """

        return self._memoryRSS

    def getRuntime(self):
        """
        Calculating the total amount of runtime taken by the mining process
        :return: returning total amount of runtime taken by the mining process
        :rtype: float
        """

        return self._endTime - self._startTime

    def getPatternsAsDataFrame(self):
        """
        Storing final frequent patterns in a dataframe
        :return: returning frequent patterns in a dataframe
        :rtype: pd.DataFrame
        """

        dataFrame = {}
        data = []
        for a, b in self._finalPatterns.items():
            data.append([a, b])
            dataFrame = _ab._pd.DataFrame(data, columns=['Patterns', 'Support'])
        return dataFrame

    def save(self, outFile):
        """
        Complete set of frequent patterns will be loaded in to an output file
        :param outFile: name of the output file
        :type outFile: csvfile
        """
        self._oFile = outFile
        writer = open(self._oFile, 'w+')
        for x, y in self._finalPatterns.items():
            if type(x) == tuple:
                pattern = ""
                for item in x:
                    pattern = pattern + str(item) + " "
                s1 = pattern + ":" + str(y)
            else:
                s1 = str(x) + ":" + str(y)
            writer.write("%s \n" % s1)

    def getPatterns(self):
        """
        Function to send the set of frequent patterns after completion of the mining process
        :return: returning frequent patterns
        :rtype: dict
        """
        return self._finalPatterns

    def printResults(self):
        """
        This function is used to print the results
        """
        print("Total number of Frequent Patterns:", len(self.getPatterns()))
        print("Total Memory in USS:", self.getMemoryUSS())
        print("Total Memory in RSS", self.getMemoryRSS())
        print("Total ExecutionTime in ms:", self.getRuntime())


    def _convert(self, value):
        """
        To convert the user specified minSup value
        :param value: user specified minSup value
        :type value: int or float or str
        :return: converted type
        """
        if type(value) is int:
            value = int(value)
        elif type(value) is float:
            value = (self._lno * value)
        elif type(value) is str:
            if '.' in value:
                value = float(value)
                value = (self._lno * value)
            else:
                value = int(value)
        else:
            print("minSup is not correct")
        return value


if __name__ == "__main__":
    _ap = str()
    if len(_ab._sys.argv) == 5 or len(_ab._sys.argv) == 6:
        if len(_ab._sys.argv) == 6:
            _ap = parallelFPGrowth(_ab._sys.argv[1], _ab._sys.argv[3], _ab._sys.argv[4], _ab._sys.argv[5])
        if len(_ab._sys.argv) == 5:
            _ap = parallelFPGrowth(_ab._sys.argv[1], _ab._sys.argv[3], _ab._sys.argv[4])
        _ap.startMine()
        _ap.mine()
        _finalPatterns = _ap.getPatterns()
        print("Total number of Frequent Patterns:", len(_finalPatterns))
        # _ap.save(_ab._sys.argv[2])
        _memUSS = _ap.getMemoryUSS()
        print("Total Memory in USS:", _memUSS)
        _memRSS = _ap.getMemoryRSS()
        print("Total Memory in RSS", _memRSS)
        _run = _ap.getRuntime()
        print("Total ExecutionTime in ms:", _run)
    else:
        print("Error! The number of input parameters do not match the total number of parameters provided")