Cleanup dataset for visualization

Question

This code cleans up a big dataset into a very clean and a flat file that can be further used downstream like visualization.

I am expecting to improve the code to essentially run it faster and cleanup the code to avoid any inefficient functions that should not take further resources than it should.

The problem is that I have written the code as and when I have experienced any functions as needed to be added on the fly and now, I am not very happy with it.

I am seeking help from this community on the best code that can be written which performs the same work that this code does. I would also be learning on the best practices while working with pandas.

There is so much on the internet on Pandas that every function and code seems to be a clean code but when i look at it collectively, its so bad.

My dataframe:

df.head()

   Unnamed: 0                       game score home_odds draw_odds away_odds country                 league             datetime
0           0  Sport Recife - Imperatriz   2:2      1.36      4.31      7.66  Brazil  Copa do Nordeste 2020  2020-02-07 00:00:00
1           1           ABC - America RN   2:1      2.62      3.30      2.48  Brazil  Copa do Nordeste 2020  2020-02-02 22:00:00
2           2  Frei Paulistano - Nautico   0:2      5.19      3.58      1.62  Brazil  Copa do Nordeste 2020  2020-02-02 00:00:00
3           3    Botafogo PB - Confianca   1:1      2.06      3.16       3.5  Brazil  Copa do Nordeste 2020  2020-02-02 22:00:00
4           4          Fortaleza - Ceara   1:1      2.19      2.98      3.38  Brazil  Copa do Nordeste 2020  2020-02-02 22:00:00


df.describe()
         Unnamed: 0
count  1.115767e+06
mean   5.574871e+05
std    3.220941e+05
min    0.000000e+00
25%    2.785455e+05
50%    5.574870e+05
75%    8.364285e+05
max    1.115370e+06

I use this code to cleanup the dataframe:

from datetime import datetime
import pandas as pd
import numpy as np
from tabulate import tabulate

start = datetime.now()

df = pd.read_csv()

#This part essentially splits columns and harmonises the entie dataframe

# This code harmonises the game column e.g. "Talleres (R.E) - Defensores Unidos" should be as "Talleres - "Defensores Unidos" removing any brackets and its values and removes any date values in the column
df['game'] = df['game'].astype(str).str.replace('(\(\w+\))', '', regex=True)
df['game'] = df['game'].astype(str).str.replace('(\s\d+\S\d+)$', '', regex=True)
# This code removes any numerical values in the league column. Many times the league column has years concatenated which is what we don't want e.g "Brazil  Copa do Nordeste 2020" should be "Brazil  Copa do Nordeste"
df['league'] = df['league'].astype(str).str.replace('(\s\d+\S\d+)$', '', regex=True)
# This part splits the game column into two competing teams i.e. home team and away team by the delimiter "-"
df[['home_team', 'away_team']] = df['game'].str.split(' - ', expand=True, n=1)
# This part splits the score column into two competing teams i.e. home score and away score by the delimiter ":"
df[['home_score', 'away_score']] = df['score'].str.split(':', expand=True)
# This code removes any non numerical values in the home score and away score columns. e.ge scores can't have "aet", "canc", ".", etc. We dont want anything that cannot be identified as filetype:int in pandas
df['away_score'] = df['away_score'].astype(str).str.replace('[a-zA-Z\s\D]', '', regex=True)
df['home_score'] = df['home_score'].astype(str).str.replace('[a-zA-Z\s\D]', '', regex=True)
df = df[df.home_score != "."]
df = df[df.home_score != ".."]
df = df[df.home_score != "."]
df = df[df.home_odds != "-"]
df = df[df.draw_odds != "-"]
df = df[df.away_odds != "-"]
m = df[['home_odds', 'draw_odds', 'away_odds']].astype(str).agg(lambda x: x.str.count('/'), 1).ne(0).all(1)
n = df[['home_score']].agg(lambda x: x.str.count('-'), 1).ne(0).all(1)
o = df[['away_score']].agg(lambda x: x.str.count('-'), 1).ne(0).all(1)

# I get UserWarning: Boolean Series key will be reindexed to match DataFrame index. at these parts
df = df[~n]
df = df[~m]
df = df[~n]
df = df[~o]
df = df[df.home_score != '']
df = df[df.away_score != '']
df = df.dropna()

# We now would be keeping only the columns we want
df = df.loc[:, df.columns.intersection(
    ['datetime', 'country', 'league', 'home_team', 'away_team', 'home_odds', 'draw_odds', 'away_odds', 'home_score',
     'away_score'])]

#We are making sure that the columns are as per data types that we would want pandas to identify. Pandas does not seem to do a very good job identifying data types correctly.
colt = {
    'country': str,
    'league': str,
    'home_team': str,
    'away_team': str,
    'home_odds': float,
    'draw_odds': float,
    'away_odds': float,
    'home_score': int,
    'away_score': int
}
df = df.astype(colt)

# This part removes any leading and trailing whitespaces in the string columns
df = df.applymap(lambda x: x.strip() if isinstance(x, str) else x)

# Cleaning data where odds are greater than 100 and less than -1 and dropping duplicates
df = df[df['home_odds'] <= 100]
df = df[df['draw_odds'] <= 100]
df = df[df['away_odds'] <= 100]
df = df.drop_duplicates(['datetime', 'home_score', 'away_score', 'country', 'league', 'home_team', 'away_team'],
                        keep='last')
df = df.applymap(lambda x: x.strip() if isinstance(x, str) else x)
df.to_csv()
time_taken = end - start
print('Time taken to complete: ', time_taken)

df.head()
      home_odds    draw_odds    away_odds  country    league            datetime             home_team        away_team      home_score    away_score
--  -----------  -----------  -----------  ---------  ----------------  -------------------  ---------------  -----------  ------------  ------------
 0         1.36         4.31         7.66  Brazil     Copa do Nordeste  2020-02-07 00:00:00  Sport Recife     Imperatriz              2             2
 1         2.62         3.3          2.48  Brazil     Copa do Nordeste  2020-02-02 22:00:00  ABC              America RN              2             1
 2         5.19         3.58         1.62  Brazil     Copa do Nordeste  2020-02-02 00:00:00  Frei Paulistano  Nautico                 0             2
 3         2.06         3.16         3.5   Brazil     Copa do Nordeste  2020-02-02 22:00:00  Botafogo PB      Confianca               1             1
 4         2.19         2.98         3.38  Brazil     Copa do Nordeste  2020-02-02 22:00:00  Fortaleza        Ceara                   1             1

It takes me 9 minutes to run this code with warnings:

sys:1: DtypeWarning: Columns (3,4,5) have mixed types.Specify dtype option on import or set low_memory=False.
G:/My Drive/Odds/Code/5. Creating updated training data.py:33: UserWarning: Boolean Series key will be reindexed to match DataFrame index.
  df = df[~n]
G:/My Drive/Odds/Code/5. Creating updated training data.py:34: UserWarning: Boolean Series key will be reindexed to match DataFrame index.
  df = df[~o]

How can I cleanup this code and run it faster using pandas?

Also, I have a GPU so I can exploit cudf however I am using the python 3.7 environment and cudf does not seem to support it and am not familiar with conda yet.

Answer 1

Any time that you have a section labelled This part, that's a good indication that you should have a function. Code like this should not exist in a flat file with no structure. Among many other reasons,

• once one of your subroutines finishes, any intermediate variables that go out of scope like m, n etc. will be eligible for garbage collection
• performing profiling will be much more possible - please do this; otherwise you're effectively flying blind
• it will be more legible and self-documenting
• it will be possible for your IDE to perform folding, etc.

Until you have profiling data it's effectively impossible to say what part of your code is the slowest. Some random guesses:

• where you do astype(str), that's likely not necessary
• when you run successive predicate selection like

df = df[df.home_score != "."]
df = df[df.home_score != ".."]

you should instead do only one [] indexing operation, and combine your predicates logically using &

• See if there is a performance difference if you prefer in-place operations, including dropping columns and NA, which are both possible in-place
• Does your colt actually do anything? Is there any evidence that your column types are non-uniform? Could you, instead of doing a blanket type coerce, target specific columns that need conversion?
• Avoid applymap. There is a vectorised strip available. You do this twice - why?

Non-performance topics:

• Don't name variables m, n, o - these are opaque and non-descriptive
• Rather than writing "this part harmonises ...", you can just write "harmonise ..." - i.e. use the imperative mood which is common in function documentation.

A first stab, addressing some of the above, looks like

import pandas as pd
from datetime import datetime


def harmonize_game(df: pd.DataFrame) -> pd.DataFrame:
    """
    harmonise the game column e.g. "Talleres (R.E) - Defensores Unidos" should
    be split as "Talleres - "Defensores Unidos" and removes any date values in
    the column
    """
    df['game'] = df['game'].astype(str).str.replace(r'(\(\w+\))', '', regex=True)
    df['game'] = df['game'].astype(str).str.replace(r'(\s\d+\S\d+)$', '', regex=True)

    # Remove any numerical values in the league column. Many times the league
    # column has years concatenated which is what we don't want e.g "Brazil
    # Copa do Nordeste 2020" should be "Brazil  Copa do Nordeste"
    df['league'] = df['league'].astype(str).str.replace(r'(\s\d+\S\d+)$', '', regex=True)

    # Split the game column into tow competing teams i.e. home team and away team by the delimiter "-"
    df[['home_team', 'away_team']] = df['game'].str.split(' - ', expand=True, n=1)

    # Split the game column into tow competing teams i.e. home team and away team by the delimiter ":"
    df[['home_score', 'away_score']] = df['score'].str.split(':', expand=True)
    return df


def numerical_scores(df: pd.DataFrame) -> pd.DataFrame:
    """
    Remove any non numerical values in the home score and away score columns.
    e.ge scores can have "aet", "canc", ".", etc. We don't want anything that
    cannot be identified as filetype:int in pandas
    """
    df['away_score'] = df['away_score'].astype(str).str.replace(r'[a-zA-Z\s\D]', '', regex=True)
    df['home_score'] = df['home_score'].astype(str).str.replace(r'[a-zA-Z\s\D]', '', regex=True)
    df = df[df.home_score != "."]
    df = df[df.home_score != ".."]
    df = df[df.home_score != "."]
    df = df[df.home_odds != "-"]
    df = df[df.draw_odds != "-"]
    df = df[df.away_odds != "-"]
    m = df[['home_odds', 'draw_odds', 'away_odds']].astype(str).agg(lambda x: x.str.count('/'), 1).ne(0).all(1)
    n = df[['home_score']].agg(lambda x: x.str.count('-'), 1).ne(0).all(1)
    o = df[['away_score']].agg(lambda x: x.str.count('-'), 1).ne(0).all(1)
    df = df[~m]
    df = df[~n]
    df = df[~o]
    df = df[df.home_score != '']
    df = df[df.away_score != '']
    df = df.dropna()
    return df


def coerce_columns(df: pd.DataFrame) -> pd.DataFrame:
    """Keep only the columns we want"""
    df = df.loc[
        :, df.columns.intersection([
            'datetime', 'country', 'league', 'home_team', 'away_team',
            'home_odds', 'draw_odds', 'away_odds', 'home_score',
            'away_score',
        ])
    ]

    # Make sure that the columns are as per data types that we would want pandas
    # to identify. Pandas does not seem to do a very good job identifying data
    # types correctly.
    colt = {
        'country': str,
        'league': str,
        'home_team': str,
        'away_team': str,
        'home_odds': float,
        'draw_odds': float,
        'away_odds': float,
        'home_score': int,
        'away_score': int
    }
    df = df.astype(colt)
    return df


def strip_strings(df: pd.DataFrame) -> pd.DataFrame:
    """remove any leading and trailing whitespaces in the string columns"""
    return df.applymap(lambda x: x.strip() if isinstance(x, str) else x)


def clean_odds(df: pd.DataFrame) -> pd.DataFrame:
    """Cleaning data where odds are greater than 100 and less than -1 and dropping duplicates"""
    df = df[df['home_odds'] <= 100]
    df = df[df['draw_odds'] <= 100]
    df = df[df['away_odds'] <= 100]
    df = df.drop_duplicates(
        ['datetime', 'home_score', 'away_score', 'country', 'league', 'home_team', 'away_team'],
        keep='last',
    )
    df = df.applymap(lambda x: x.strip() if isinstance(x, str) else x)
    return df


def clean(df: pd.DataFrame) -> pd.DataFrame:
    df = harmonize_game(df)
    df = numerical_scores(df)
    df = coerce_columns(df)
    df = strip_strings(df)
    df = clean_odds(df)
    return df


def test() -> None:
    df = pd.DataFrame(
        (
            (0, 'Sport Recife - Imperatriz', '2:2', 1.36, 4.31, 7.66, 'Brazil', 'Copa do Nordeste 2020', datetime.strptime('2020-02-07 00:00:00', '%Y-%m-%d %H:%M:%S')),
            (1,          'ABC - America RN', '2:1', 2.62, 3.30, 2.48, 'Brazil', 'Copa do Nordeste 2020', datetime.strptime('2020-02-02 22:00:00', '%Y-%m-%d %H:%M:%S')),
            (2, 'Frei Paulistano - Nautico', '0:2', 5.19, 3.58, 1.62, 'Brazil', 'Copa do Nordeste 2020', datetime.strptime('2020-02-02 00:00:00', '%Y-%m-%d %H:%M:%S')),
            (3,   'Botafogo PB - Confianca', '1:1', 2.06, 3.16, 3.50, 'Brazil', 'Copa do Nordeste 2020', datetime.strptime('2020-02-02 22:00:00', '%Y-%m-%d %H:%M:%S')),
            (4,         'Fortaleza - Ceara', '1:1', 2.19, 2.98, 3.38, 'Brazil', 'Copa do Nordeste 2020', datetime.strptime('2020-02-02 22:00:00', '%Y-%m-%d %H:%M:%S')),
        ),
        columns=(
             'Unnamed: 0',           'game',
                                           'score',
                                             'home_odds',
                                                   'draw_odds',
                                                         'away_odds', 'country',               'league', 'datetime',
        ),
    )
    clean(df)


if __name__ == '__main__':
    test()