Basic Data Wrangling and Plotting Tutorial

Hongtao Hao / 2022-10-05 

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import altair as alt

Download data here: aiddata.csv

Data import and basic manipulation # 

df = pd.read_csv('../static/files/aiddata.csv')
# df.head()


# don't actually need the first two columns:
df = df.iloc[:, 2:]

# rename the columns
df.columns = ['year', 'donor', 'recipient', 'amount', 'purpose_code', 'purpose_name']
# df.head()


# check the shape
df.shape

# close to 10K rows!

(98540, 6)

# check year range

min(df.year), max(df.year)

(1973, 2013)

Task 1 #

The first task is: for each country or region, visualize the difference between its donation and receiving, for each year. To do that, we need a dataframe that has these three columns:

Then, we can visualize the difference by year for each country or region (hereafter referred as “country”).

There are two obvious challenges:

Let’s first get donation and receiving data from the source data.

def parse_donation_and_receiving(df, var):
    """
    To get donationn or receiving data from df
    
    Input:
        - df: df
        - var: either 'donor' or 'recipient'
    
    Return:
        a dataframe
    """
    data = []
    for group in df.groupby(['year', var]):
        # total amount of donation or receiving in that year for this country
        total_amount = group[1].amount.sum()
        # the amount is so big, we use the unit of billio (10**9)
        total_amount = total_amount / 10**9
        data.append((
            # year:
            group[0][0],
            # country:
            group[0][1],
            # amount:
            total_amount
        ))
    var_name = 'donation' if var == 'donor' else 'received'
    data_df = pd.DataFrame(
        data, columns = ['year', 'country', var_name]
    )
    return data_df

# We first get the donation data
donation_df = parse_donation_and_receiving(df, 'donor')
donation_df.head()

year country donation
0 1973 Australia 0.046286
1 1973 Belgium 0.039251
2 1973 Canada 0.437928
3 1973 France 0.247190
4 1973 Germany 0.562232 
# Then we get the receiving data
receiving_df = parse_donation_and_receiving(df, 'recipient')
receiving_df.head()

year country received
0 1973 Brazil 0.312075
1 1973 Chile 0.088056
2 1973 Colombia 0.549945
3 1973 Cyprus 0.009613
4 1973 India 2.285257

First, let’s look at how many countries are there:

all_cntry = list(df.donor) + list(df.recipient)
all_cntry = list(set(all_cntry))

# there are in total 47 unique countries
len(all_cntry)

47

We have 47 unique countries, but the problem is, in some years, not all countries are present, for example, in receiving_df, only 10 countries were present in the year of 1973.

receiving_df[receiving_df.year == 1973]

year country received
0 1973 Brazil 0.312075
1 1973 Chile 0.088056
2 1973 Colombia 0.549945
3 1973 Cyprus 0.009613
4 1973 India 2.285257
5 1973 Korea 1.363707
6 1973 Kuwait 0.000325
7 1973 Saudi Arabia 0.000065
8 1973 Thailand 0.206363
9 1973 United Arab Emirates 0.000065 
# same issue for donation data
donation_df[donation_df.year == 1973]

year country donation
0 1973 Australia 0.046286
1 1973 Belgium 0.039251
2 1973 Canada 0.437928
3 1973 France 0.247190
4 1973 Germany 0.562232
5 1973 Italy 0.166719
6 1973 Japan 0.938966
7 1973 Netherlands 0.162751
8 1973 Norway 0.035875
9 1973 Sweden 0.168369
10 1973 Switzerland 0.014061
11 1973 United Kingdom 0.442579
12 1973 United States 1.553264

You might ask: what will be the issue if not all countries are in each year. Good question! The issue is that if we do not have data for all countries each year, we are not able to get the difference between donation and receiving for each country in each year very easily. To do that, we need to “normalize” the time. If a country does not appear in a year, that means it does not recieve (or donate) any money in that year.

This is how I solved the problem:

def normalize_time(data, all_cntry, var_name):
    """normlize time for donation_df and receiving_df
    
    Input: 
        - data: either donation_df or receiving_df
        - all_cntry
        - var_name: either 'donation' or 'received'
    
    Output:
        - a dataframe
    """
    dfs = []
    for group in data.groupby('year'):
        year = group[0]
        present_cntry = group[1].country.tolist()
        absent_cntry = [x for x in all_cntry if x not in present_cntry]
        absent_df = pd.DataFrame({
            'year': year,
            'country': absent_cntry,
            var_name: 0
        })
        dff = pd.concat([group[1], absent_df], ignore_index = True)
        
        # by sorting the country, we make sure that rows in the output
        # donation and receiving dataframes correspond to each other
        dff.sort_values(by='country', ascending=True, inplace=True)
        dfs.append(dff)
    
    data_df = pd.concat(dfs, ignore_index = True)
    return data_df

donation = normalize_time(donation_df, all_cntry, var_name = 'donation')
donation.head()

year country donation
0 1973 Australia 0.046286
1 1973 Austria 0.000000
2 1973 Belgium 0.039251
3 1973 Brazil 0.000000
4 1973 Canada 0.437928 
receiving = normalize_time(receiving_df, all_cntry, var_name = 'received')
receiving.head()

year country received
0 1973 Australia 0.000000
1 1973 Austria 0.000000
2 1973 Belgium 0.000000
3 1973 Brazil 0.312075
4 1973 Canada 0.000000 
# to check whether the countries in the two lists are the same
r_c = list(set(receiving.country))
d_c = list(set(donation.country))
r_c == d_c

True

Then we combine the two datasets and calculate the difference:

all_df = donation
all_df['received'] = receiving['received']
all_df['d_minus_r'] = all_df['donation'] - all_df['received']
all_df.head()

year country donation received d_minus_r
0 1973 Australia 0.046286 0.000000 0.046286
1 1973 Austria 0.000000 0.000000 0.000000
2 1973 Belgium 0.039251 0.000000 0.039251
3 1973 Brazil 0.000000 0.312075 -0.312075
4 1973 Canada 0.437928 0.000000 0.437928 
all_df.shape

(1927, 5)

To make the plot, we need to transform the format of year:

all_df['year'] = pd.to_datetime(all_df['year'], format='%Y')

alt.Chart(all_df).mark_line().encode(
    x = alt.X(
        'year',
        title = 'Year'
    ),
    y = alt.Y(
        'd_minus_r:Q',
        title = 'Donation - Received',
        axis=alt.Axis(
            titleColor='#5276A7',
            labelExpr = 'datum.value + "B `$"'
        ),
    ),
    facet = alt.Facet(
        "country:N",
        columns = 8,
        title = 'Country/Region',
        header = alt.Header(labelFontSize = 12)
    )
).properties(
    width = 120,
    height = 110,
).configure_header(
    titleColor='#5276A7',
    titleFontSize=16,
#     labelColor='red',
    labelFontSize=14
)

The problem with this plot is that with it, we are not able to see the actual donation and receiving. We can achieve this in the following way.

all_df_melted = pd.melt(all_df, 
        id_vars=['year', 'country'], 
        value_vars=['donation', 'received'],
        var_name = 'category',
        value_name = 'amount'
       )
all_df_melted.replace({
    'donation': 'Donation',
    'received': 'Received'
}, inplace = True)
all_df_melted.rename(
    columns = {
        'category': 'Category'
    }, inplace = True)
all_df_melted.head()

year country Category amount
0 1973-01-01 Australia Donation 0.046286
1 1973-01-01 Austria Donation 0.000000
2 1973-01-01 Belgium Donation 0.039251
3 1973-01-01 Brazil Donation 0.000000
4 1973-01-01 Canada Donation 0.437928

This is important: because we are going to plot both donaiton and received amount each year for each country, it is important to replace 0 with NaN. Because the figures will be very small, if we do not change 0 to NaN, then we will consider 0 as small amount of donation or received amount, which is misleading.

all_df_melted['amount'] = all_df_melted['amount'].replace(0, np.nan)
all_df_melted.head()

year country Category amount
0 1973-01-01 Australia Donation 0.046286
1 1973-01-01 Austria Donation NaN
2 1973-01-01 Belgium Donation 0.039251
3 1973-01-01 Brazil Donation NaN
4 1973-01-01 Canada Donation 0.437928 
alt.Chart(all_df_melted).mark_line().encode(
    x = alt.X(
        'year',
        title = 'Year'
    ),
    y = alt.Y(
        'amount:Q',
        title = 'Amount',
        axis=alt.Axis(
            titleColor='#5276A7',
            labelExpr = 'datum.value + "B $`"'
        ),
    ),
    color = alt.Color(
        'Category:N',
    ),
    facet = alt.Facet(
        "country:N",
        columns = 8,
        title = 'Country/Region',
        header = alt.Header(labelFontSize = 12)
    )
).properties(
    width = 120,
    height = 110,
).configure_header(
    titleColor='#5276A7',
    titleFontSize=16,
#     labelColor='red',
    labelFontSize=14
)

Task 2 #

The second task is: find the top 10 coalesced purposes (according to the amount of donation) and for these ten purposes, show how their amount changes by time.

# First, find the top ten purposes
top_purpose_df = df.groupby('purpose_name')['amount'].agg('sum').nlargest(n=10)
top_10_purposes = list(top_purpose_df.index)
top_10_purposes

['Air transport',
 'Rail transport',
 'Industrial development',
 'Power generation/non-renewable sources',
 'RESCHEDULING AND REFINANCING',
 'Import support (capital goods)',
 'Social/ welfare services',
 'Telecommunications',
 'Power generation/renewable sources',
 'Sectors not specified']

# Then select only data involving these top purposes
# We only want three columns: year, amount, and purpose name
purpose_df = df[df.purpose_name.isin(top_10_purposes)][['year', 'amount', 'purpose_name']]
# So many rows:
purpose_df.shape

(17793, 3)

purpose_df.head()

year amount purpose_name
0 1977 348718518.0 Power generation/renewable sources
2 1983 79371799.0 Rail transport
3 1984 212202942.0 Rail transport
11 1976 193588873.0 Power generation/renewable sources
19 1986 53561256.0 Power generation/renewable sources 
# For each purpose, we calculate the yearly sum:
purpose_dfs = []
for group in purpose_df.groupby('purpose_name'):
    purpose_name = group[0]
    group_df = group[1].groupby('year')['amount'].sum().to_frame()
    group_df.reset_index(inplace=True)
    group_df['purpose_name'] = group[0]
    purpose_dfs.append(group_df)
purpose = pd.concat(purpose_dfs, ignore_index = True)
purpose.head()

year amount purpose_name
0 1974 1.525672e+07 Air transport
1 1975 1.582943e+07 Air transport
2 1977 7.046280e+05 Air transport
3 1979 7.342023e+07 Air transport
4 1980 3.088275e+09 Air transport 
purpose.shape

(338, 3)

# change amount to the unit of billion
purpose['amount'] = purpose['amount'] / 10**9

# # change the format to be year
purpose['year'] = pd.to_datetime(purpose['year'], format='%Y')

max(purpose.amount)

8.344866729

# to change the purpose name
purpose_renamer_dic = {'Air transport': 'Air Transport',
 'Rail transport': 'Rail Transport',
 'Industrial development': 'Industrial Development',
 'Power generation/non-renewable sources': 'Power: Non-renewable',
 'RESCHEDULING AND REFINANCING': 'Rescheduling & Refinancing',
 'Import support (capital goods)': 'Import Support',
 'Social/ welfare services': 'Welfare Services',
 'Telecommunications': 'Telecommunications',
 'Power generation/renewable sources': 'Power: Renewable',
 'Sectors not specified': 'Not Specified'
 }

purpose.replace(purpose_renamer_dic, inplace = True)

# make the plot
alt.Chart(purpose).mark_line().encode(
    x = 'year',
    y = 'amount',
    color = 'purpose_name',
    strokeDash = 'purpose_name'
)

The problem with this chart is that it is so messy. We can visualize it in another way:

alt.Chart(purpose).mark_rect().encode(
    x = alt.X(
        'year(year)',
        title = 'Year'
    ),
    y = alt.Y(
        'purpose_name:N',
        title = 'Coalesced Purposes of Donation',
        sort=alt.EncodingSortField(field='amount', op='sum', order='descending')
    ),
    color = alt.Color(
        'amount:Q',
        title = 'Amount (in Billion $`)',
    )
).properties(
    width = 720,
    height = 300,
    title = 'Top Ten Coalesced Purposes by Total Donation Amount (1973-2013)',
).configure_axis(
    labelFontSize=11,
    titleFontSize=12
)
#DS

Last modified on 2022-10-16