|
| 1 | +from __future__ import print_function |
| 2 | +import sys |
| 3 | +from pyspark.sql import SparkSession |
| 4 | + |
| 5 | +#----------------------------------------------------- |
| 6 | +# Find Minimum and Maximum of all input by |
| 7 | +# using the mapPartitions() transformations. |
| 8 | +# |
| 9 | +# The idea is that each partition will find |
| 10 | +# (local_min, local_max, local_count) |
| 11 | +# and then we find (final_min, final_max, final_count) |
| 12 | +# for all partitions. |
| 13 | +# |
| 14 | +# input ---- N partitioned ----> partition-1, partition-2, ... partition_N |
| 15 | +# |
| 16 | +# partition-1 => local_1 = (local_min_1, local_max_1, local_count_1) |
| 17 | +# partition-2 => local_2 = (local_min_2, local_max_2, local_count_2) |
| 18 | +# ... |
| 19 | +# partition-N => local_N = (local_min_N, local_max_N, local_count_N) |
| 20 | +# |
| 21 | +# final_min_max = minmax(local_1, local_2, ..., local_N) |
| 22 | +# |
| 23 | +#------------------------------------------------------ |
| 24 | +# Input Parameters: |
| 25 | +# INPUT_PATH as a file of numbers |
| 26 | +# |
| 27 | +# Example: sample_numbers.txt |
| 28 | +# |
| 29 | +# $ cat sample_numbers.txt |
| 30 | +#23,24,22,44,66,77,44,44,555,666 |
| 31 | +#12,4,555,66,67,68,57,55,56,45,45,45,66,77 |
| 32 | +#34,35,36,97300,78,79 |
| 33 | +#120,44,444,445,345,345,555 |
| 34 | +#11,33,34,35,36,37,47,7777,8888,6666,44,55 |
| 35 | +#10,11,44,66,77,78,79,80,90,98,99,100,102,103,104,105 |
| 36 | +#6,7,8,9,10 |
| 37 | +#8,9,10,12,12 |
| 38 | +#7777 |
| 39 | +#222,333,444,555,666,111,112,5,113,114 |
| 40 | +#5555,4444,24 |
| 41 | +# |
| 42 | +# |
| 43 | +#------------------------------------------------------- |
| 44 | +# @author Mahmoud Parsian |
| 45 | +#------------------------------------------------------- |
| 46 | + |
| 47 | +# |
| 48 | +# |
| 49 | +#========================================== |
| 50 | +# Find (min, max, count) for a given single partition. |
| 51 | +# |
| 52 | +# partition_iterator is an iterator over |
| 53 | +# elements of a single partition. |
| 54 | +# partition_iterator : iterator over |
| 55 | +# set of input records and each input record |
| 56 | +# has the format as: |
| 57 | +# <number><,><number><,>...<number> |
| 58 | +# |
| 59 | +def minmax(partition_iterator): |
| 60 | +# |
| 61 | + print("type(partition_iterator)=", type(partition_iterator)) |
| 62 | + #('type(partition_iterator)=', <type 'itertools.chain'>) |
| 63 | + # type(partition_iterator)= <type 'generator'> |
| 64 | + # |
| 65 | + try: |
| 66 | + first_record = next(partition_iterator) |
| 67 | + print("first_record=", first_record) |
| 68 | + except StopIteration: |
| 69 | + # for empty partitions |
| 70 | + return [None] |
| 71 | +# |
| 72 | + numbers = [int(n) for n in first_record.split(",")] |
| 73 | + local_min = min(numbers) |
| 74 | + local_max = max(numbers) |
| 75 | + local_count = len(numbers) |
| 76 | +# |
| 77 | + # handle remaining records in a partition |
| 78 | + for record in partition_iterator: |
| 79 | + #print("record=", record) |
| 80 | + numbers = [int(n) for n in record.split(",")] |
| 81 | + min2 = min(numbers) |
| 82 | + max2 = max(numbers) |
| 83 | + # update min, max, and count |
| 84 | + local_count += len(numbers) |
| 85 | + local_max = max(local_max, max2) |
| 86 | + local_min = min(local_min, min2) |
| 87 | +# end-for |
| 88 | + return [(local_min, local_max, local_count)] |
| 89 | +#end-def |
| 90 | +# |
| 91 | +#========================================== |
| 92 | +# |
| 93 | +# find final (min, max, count) from all partitions |
| 94 | +# min_max_count_list = [ |
| 95 | +# (min_1, max_1, count_1), |
| 96 | +# (min_2, max_2, count_2), |
| 97 | +# ... |
| 98 | +# (min_N, max_N, count_N) |
| 99 | +# ] |
| 100 | +# |
| 101 | +def find_min_max_count(min_max_count_list): |
| 102 | + first_time = True |
| 103 | + # iterate tuple3 in min_max_count_list: |
| 104 | + for local_min, local_max, local_count in min_max_count_list: |
| 105 | + if (first_time): |
| 106 | + final_min = local_min |
| 107 | + final_max = local_max |
| 108 | + final_count = local_count |
| 109 | + first_time = False |
| 110 | + else: |
| 111 | + final_min = min(final_min, local_min) |
| 112 | + final_max = max(final_max, local_max) |
| 113 | + final_count += local_count |
| 114 | + #end-for |
| 115 | + return (final_min, final_max, final_count) |
| 116 | +#end-def |
| 117 | +#========================================== |
| 118 | +# |
| 119 | +def debug_partition(iterator): |
| 120 | + print("===begin-partition===") |
| 121 | + for x in iterator: |
| 122 | + print(x) |
| 123 | + print("===end-partition===") |
| 124 | +#end-def |
| 125 | +# |
| 126 | +#========================================== |
| 127 | +# main(): |
| 128 | + |
| 129 | +if len(sys.argv) != 2: |
| 130 | + print("Usage: ", __file__, "<input-path>", file=sys.stderr) |
| 131 | + exit(-1) |
| 132 | + |
| 133 | +# create an instance of SparkSession |
| 134 | +spark = SparkSession.builder.appName("minmax").getOrCreate() |
| 135 | +# |
| 136 | + |
| 137 | +# handle input parameter |
| 138 | +input_path = sys.argv[1] |
| 139 | +print("input_path=", input_path) |
| 140 | + |
| 141 | +#===================================== |
| 142 | +# read input and apply mapPartitions() |
| 143 | +#===================================== |
| 144 | +# rdd: RDD[String] |
| 145 | +rdd = spark.sparkContext.textFile(input_path) |
| 146 | +print("rdd=", rdd) |
| 147 | +print("rdd.count=", rdd.count()) |
| 148 | +print("rdd.collect()=", rdd.collect()) |
| 149 | +print("rdd.getNumPartitions()=", rdd.getNumPartitions()) |
| 150 | +# |
| 151 | +#===================================== |
| 152 | +# find (min, max, count) per partition |
| 153 | +# custom function is minmax |
| 154 | +#===================================== |
| 155 | +# min_max_count: RDD[(min, max, count)] |
| 156 | +# min_max_count: [(min_1, max_1, count_1), |
| 157 | +# (min_2, max_2, count_2), |
| 158 | +# ..., |
| 159 | +# (min_N, max_N, count_N)] |
| 160 | +# |
| 161 | +# Apply mapPartitions() and then drop None elements |
| 162 | +min_max_count = rdd.mapPartitions(minmax).filter(lambda x: x is not None) |
| 163 | +# |
| 164 | +print("min_max_count=", min_max_count) |
| 165 | +print("min_max_count.count=", min_max_count.count()) |
| 166 | +min_max_count_list = min_max_count.collect() |
| 167 | +print("min_max_count.collect()=", min_max_count_list) |
| 168 | + |
| 169 | +#===================================== |
| 170 | +# find final (min, max, count) from all partitions |
| 171 | +#===================================== |
| 172 | +final_min, final_max, final_count = find_min_max_count(min_max_count_list) |
| 173 | +print("final: (min, max, count)= (", final_min, ", ", final_max, ", ", final_count, ")") |
| 174 | + |
| 175 | +# done! |
| 176 | +spark.stop() |
0 commit comments