I have a Python script that given some configuration, generates a random CSV file that can be used for testing purposes.
I want to know if it adheres to the best Python and coding practices. It works for cases where I need upwards of 10K+ rows fast enough for my requirements, so I am not too worried about performance although inputs on performance are also appreciated.
Input:
- Schema: as a dict, information about each column name, data type and some other constraints (like fixed length/in a range/ from a given list)
- Number of rows
- Name of the output CSV file
Script:
import random as rnd
import csv
from abc import ABC, abstractmethod
# csv creator, creates a csv files with a given config
roundPrecision = 3
class BoundType(ABC):
def __init__(self, dtype, params):
self.dType = dtype
self.params = params
@abstractmethod
def generate(self):
pass
class FixedLength(BoundType):
# params is length
def generate(self):
length = self.params.get("len", 1)
if self.dType == "int":
return rnd.randint(10 ** (length - 1), 10 ** length - 1)
elif self.dType == "float":
return FixedLength("int", self.params).generate() + round(rnd.random(), roundPrecision)
elif self.dType == "string":
alphabet = list("ABCDEFGHIJKLMNOPQRSTUVWXYZ")
word = [rnd.choice(alphabet) for _ in range(length)]
return ''.join(word)
else:
return None
class FixedRange(BoundType):
# params is range
def generate(self):
lo, hi = (self.params.get("lohi"))
if self.dType == "int":
return rnd.randint(lo, hi)
elif self.dType == "float":
return round(rnd.uniform(lo, hi), roundPrecision)
else:
return None
class FromPossibleValues(BoundType):
# params is a list
def generate(self):
possibleval = self.params.get("set", set())
return rnd.choice(possibleval)
def createcsv(rows, filename, schema):
with open(f'./output/{filename}.csv', 'w', encoding='UTF8', newline='') as f:
writer = csv.writer(f)
writer.writerow(schema.keys())
for _ in range(rows):
writer.writerow([x.generate() for x in schema.values()])
Test:
from csvGen.csvGenerator import FixedLength, FixedRange, FromPossibleValues, createcsv
schema = {
"col1": FixedLength("int", {"len": 5}),
"col2": FixedLength("float", {"len": 5}),
"col3": FixedLength("string", {"len": 5}),
"col4": FixedRange("int", {"lohi": (10, 15)}),
"col5": FixedRange("float", {"lohi": (5.5, 6.7)}),
"col6": FromPossibleValues("int", {"set": [1, 2, 3, 4, 5]}),
"col7": FromPossibleValues("int", {"set": [1.1, 2.2, 3.3]}),
"col8": FromPossibleValues("int", {"set": ["A", "AB"]})
}
rows = 10
fileName = "eightVals"
createcsv(rows, fileName, schema)
This is what the output looks like for the given test :
col1 | col2 | col3 | col4 | col5 | col6 | col7 | col8 |
---|---|---|---|---|---|---|---|
51685 | 71830.471 | PAXBK | 12 | 6.192 | 1 | 2.2 | AB |
60384 | 42341.991 | RHNUK | 11 | 6.037 | 1 | 1.1 | AB |
73505 | 30997.171 | DVOGT | 10 | 6.69 | 5 | 2.2 | A |
60528 | 85072.731 | FWWXW | 10 | 5.761 | 1 | 2.2 | A |
23048 | 65401.245 | EVPUX | 13 | 6.474 | 4 | 1.1 | AB |
74748 | 66969.774 | PEULP | 15 | 6.546 | 3 | 2.2 | AB |
88763 | 34749.184 | VOAUO | 10 | 6.402 | 4 | 2.2 | AB |
77351 | 44566.163 | JOBQF | 13 | 5.683 | 1 | 2.2 | AB |
50820 | 73002.154 | EACZT | 15 | 5.711 | 1 | 1.1 | AB |
53037 | 89225.572 | YTLBI | 13 | 6.328 | 1 | 2.2 | AB |
-
1\$\begingroup\$ @Peter, your edit was mostly good, but please don't change the code comments for grammar/spelling, as that's an aspect of review. \$\endgroup\$Toby Speight– Toby Speight2022年10月18日 15:24:47 +00:00Commented Oct 18, 2022 at 15:24
2 Answers 2
This seems overbuilt, and would also be more tersely expressed with Numpy/Pandas. Additionally, your manual loops - whereas you're satisfied with current performance - will be faster in vectorised form.
This program can be as simple as
import numpy as np
import pandas as pd
from string import ascii_uppercase
from numpy.random import default_rng
round_precision = 3
nrows = 10
rand = default_rng()
array_3 = rand.choice(tuple(ascii_uppercase), size=(nrows, 5))
df = pd.DataFrame({
'col1': rand.integers(low=1e4, high=1e5, size=nrows),
'col2': rand.uniform(low=1e4, high=1e5, size=nrows).round(decimals=round_precision),
'col3': pd.DataFrame(array_3).agg(''.join, axis=1),
'col4': rand.integers(low=10, high=16, size=nrows),
'col5': rand.uniform(low=5.5, high=6.7, size=nrows).round(decimals=round_precision),
'col6': rand.integers(low=1, high=6, size=nrows),
'col7': rand.choice(np.linspace(start=1.1, stop=3.3, num=3), size=nrows),
'col8': rand.choice(('A', 'AB'), size=nrows),
})
df.to_csv('eightVals.csv', index=None)
with output
col1,col2,col3,col4,col5,col6,col7,col8
90510,54337.357,TITJJ,12,6.254,2,2.2,A
82827,10498.364,MLZFY,11,5.881,5,2.2,A
81292,95873.98,UOHVL,14,5.931,2,1.1,AB
44278,69859.781,RGQVO,11,6.492,2,2.2,AB
26424,18106.356,XPCYV,15,6.176,5,2.2,AB
12324,52143.779,VUEWA,15,5.959,2,3.3,A
38774,34881.201,URXXL,15,6.545,2,2.2,AB
25801,31699.204,XEWGS,14,6.427,3,3.3,A
59555,98153.322,WYGBV,11,6.062,1,2.2,A
44449,86569.519,MDNXN,15,6.229,1,3.3,A
If this needs to be reusable (and I'm not entirely convinced that it is, given what you've shown), you can adapt the above to utility functions. I don't know that classes are necessary.
-
1\$\begingroup\$ Storing the config separately makes a lot of sense in data pipelines IMHO. Classes don't seem necessary just for that indeed though. \$\endgroup\$Lodinn– Lodinn2022年10月18日 12:53:17 +00:00Commented Oct 18, 2022 at 12:53
I like how you started designing with generate
method as interface for your generators. I like how you define your schema. I like also how the final loop and code works.
Really can't agree with the way you implemented your generate
methods. First it was incredibly confusing and it took me unnecessarily too long to figure out, what are these classes about. To be more specific:
- Why passing params as dict? Why not pass named parameters and make them instance variables? It would make code much more readable and easier to use. Only reason I can think about is that you could instantiate all those classes with same type of parameters (but you don't need to do that).
- That big
if
in yourgenerate
method is code smell. Each of those if branches should be it's own class with it's owngenerate
implementation. - Names
FixedLength
andFromPossibleValues
can be better.
Some possibilities examples how to implement your classes better (imho):
class FixedLengthIntGenerator(BoundType):
def __init__(self, length):
self.length = length
def generate(self):
length = self.length
return rnd.randint(10 ** (length - 1), 10 ** length - 1)
# TODO: other classes for string, float, range ...
class FromPresetChoicesGenerator(BoundType):
def __init__(self, choices):
self.choices = choices
def generate(self):
return rnd.choice(self.choices)
Yes it is more verbose, but it is more readable, extensible, different implementations are not so bound. Also it's not necessary to pass this "dType" anymore. Instead, you choose what implementation to use.
-
3\$\begingroup\$ If you use a class having two methods, one of which is
__init__
, you should be using a function instead. Additionally using inheritance without a single call tosuper()
or any of the parent class' features, is a big no no for me. \$\endgroup\$Richard Neumann– Richard Neumann2022年10月18日 05:52:45 +00:00Commented Oct 18, 2022 at 5:52