Pandas: Compute the half-yearly average of monthly collected house sales

Question

Using data from From the Zillow research data site mainly city level. the data structure is 6 columns contain city related information and the remaining 245 columns contain the monthly sale price. I've used the code below to display a sample of the data

import pandas as pd
from tabulate import tabulate 

df = pd.read_csv("City_Zhvi_AllHomes.csv")
c = df.columns.tolist()
cols = c[:7] 
cols.append(c[-1]) 
print (tabulate(df[cols].iloc[23:29], headers = 'keys', tablefmt = 'orgtbl'))

The above code will print a sample as shown below:

|    |   RegionID | RegionName    | State   | Metro         | CountyName   |   SizeRank |   1996-04 |   2016-08 |
|----+------------+---------------+---------+---------------+--------------+------------+-----------+-----------|
| 23 |       5976 | Milwaukee     | WI      | Milwaukee     | Milwaukee    |         24 |     68100 |     99500 |
| 24 |       7481 | Tucson        | AZ      | Tucson        | Pima         |         25 |     91500 |    153000 |
| 25 |      13373 | Portland      | OR      | Portland      | Multnomah    |         26 |    121100 |    390500 |
| 26 |      33225 | Oklahoma City | OK      | Oklahoma City | Oklahoma     |         27 |     64900 |    130500 |
| 27 |      40152 | Omaha         | NE      | Omaha         | Douglas      |         28 |     88900 |    143800 |
| 28 |      23429 | Albuquerque   | NM      | Albuquerque   | Bernalillo   |         29 |    115400 |    172000 |

Part of df is a time series, the trick here is to separate the time dependant columns from the rest, use pandas resample and to_datetime

Suppose we are only interested in summarising the sales for the years 1998-2000 That will enable us to select the columns

# seperate time columns and convert their names to datetime
tdf = df[df.columns[6:]].rename(columns=pd.to_datetime)

# find the columns in the period 1998-2000
cols = tdf.columns
sel_cols = cols[(cols > '1997-12-31') & (cols < '2000')]

# select the columns, resample on columns
# calculate the mean 
# rename the columns the way we like

mdf = tdf[sel_cols].resample('6M',axis=1).mean().rename(
    columns=lambda x: '{:}${:}'.format(x.year, [1, 2][x.quarter > 2]))

# reattach non-time columns
mdf[df.columns[:6]] = df[df.columns[:6]]

print (tabulate(mdf[mdf.columns[0:9]].iloc[
       23:29], headers='keys', tablefmt='orgtbl'))

The above code will print a sample as shown below:

|    |   1998$1 |   1998$2 |   1999$1 |   1999$2 |   2000$1 |   RegionID | RegionName    | State   | Metro         |
|----+----------+----------+----------+----------+----------+------------+---------------+---------+---------------|
| 23 |    71900 |  72483.3 |  72616.7 |  74266.7 |    75920 |       5976 | Milwaukee     | WI      | Milwaukee     |
| 24 |    94200 |  95133.3 |  96533.3 |  99100   |   100600 |       7481 | Tucson        | AZ      | Tucson        |
| 25 |   139000 | 141900   | 145233   | 148900   |   151980 |      13373 | Portland      | OR      | Portland      |
| 26 |    68500 |  69616.7 |  72016.7 |  73616.7 |    74900 |      33225 | Oklahoma City | OK      | Oklahoma City |
| 27 |    98200 |  99250   | 103367   | 109083   |   112160 |      40152 | Omaha         | NE      | Omaha         |
| 28 |   121000 | 122050   | 122833   | 123633   |   124420 |      23429 | Albuquerque   | NM      | Albuquerque   |

The question is:

The last column from the resample result, has the year "2000" despite the selection using <'2000', why?

EDIT: Just for fun, I include a more "pandorable" method of doing the above

import pandas as pd

housing = pd.read_csv('City_Zhvi_AllHomes.csv',
    index_col=list(range(6))).filter(
    regex='199[8-9]-[0-1][0-9]').rename(
    columns=pd.to_datetime).resample('2Q',
                    closed='left',axis=1).mean().rename(
    columns=lambda x: str(x.to_period('2Q')).replace(
        'Q','$').replace('2','1').replace('4','2')).reset_index()

This delivers the desired outcome, printout of housing.iloc[23:27,4:] is shown below

|    | CountyName   |   SizeRank |   1998$1 |   1998$2 |   1999$1 |   1999$2 |
|----+--------------+------------+----------+----------+----------+----------|
| 23 | Milwaukee    |         24 |  72366.7 |  72583.3 |  73916.7 |  75750   |
| 24 | Pima         |         25 |  94883.3 |  96183.3 |  98783.3 | 100450   |
| 25 | Multnomah    |         26 | 141167   | 144733   | 148183   | 151767   |
| 26 | Oklahoma     |         27 |  69300   |  71550   |  73466.7 |  74766.7 |

You can add `axis=1` instead `tdf[sel_cols].T.resample('6M').mean().T.rename` use `tdf[sel_cols].resample('6M', axis=1).mean().rename`. — jezrael, Jan 15 '17 at 13:31
The more simplest answer is provided for this conversion is here, https://stackoverflow.com/questions/40497199/how-to-convert-monthly-data-to-quarterly-in-pandas. Also look for converting the 'Q' to 'q' here https://stackoverflow.com/questions/42330848/convert-pandas-period-period-type-column-names-to-lowercase — pushya, Jan 27 '18 at 02:55

score 6 · Accepted Answer · answered Jan 15 '17 at 21:11

Consider using the closed argument of pandas' resample to decide:

Which side of bin interval is closed

Below uses left where 6 month cut-offs are 6/30 and 12/31 instead of 1/1 and 7/1 which yielded the 2000 value:

mdf = tdf[sel_cols].T.resample('6M', closed='left').mean().T.rename(
    columns=lambda x: '{:}${:}'.format(x.year, [1, 2][x.quarter > 2]))

mdf[df.columns[:6]] = df[df.columns[:6]]
print(mdf.head())

#           1998$1         1998$2         1999$1         1999$2  RegionID    RegionName State                           Metro    CountyName  SizeRank
# 0            NaN            NaN            NaN            NaN      6181      New York    NY                        New York        Queens         1
# 1  169183.333333  179166.666667  189116.666667  198466.666667     12447   Los Angeles    CA  Los Angeles-Long Beach-Anaheim   Los Angeles         2
# 2  117700.000000  121666.666667  125550.000000  133000.000000     17426       Chicago    IL                         Chicago          Cook         3
# 3   50550.000000   50650.000000   51150.000000   51866.666667     13271  Philadelphia    PA                    Philadelphia  Philadelphia         4
# 4   97583.333333  101083.333333  104816.666667  108566.666667     40326       Phoenix    AZ                         Phoenix      Maricopa         5

print(mdf[mdf['Metro'].isin(['Milwaukee', 'Tucson', 'Portland', 'Oklahoma City', 'Omaha', 'Albuquerque'])].head())

#            1998$1         1998$2         1999$1         1999$2  RegionID   RegionName State        Metro  CountyName  SizeRank
# 23   72366.666667   72583.333333   73916.666667   75750.000000      5976    Milwaukee    WI    Milwaukee   Milwaukee        24
# 24   94883.333333   96183.333333   98783.333333  100450.000000      7481       Tucson    AZ       Tucson        Pima        25
# 25  141166.666667  144733.333333  148183.333333  151766.666667     13373     Portland    OR     Portland   Multnomah        26
# 26   98950.000000  102450.000000  108016.666667  112116.666667     40152        Omaha    NE        Omaha     Douglas        27
# 27  121816.666667  122666.666667  123550.000000  124333.333333     23429  Albuquerque    NM  Albuquerque  Bernalillo        28

By the way consider reshaping the data with melt to long format, aggregate by half year, then pivot_table back to wide format. Admittedly, performance is lowered here but arguably more readable (mainly the Half_Year string concatenation is the bottleneck). You do get a long form dataset for other aggregations and/or modeling:

import pandas as pd
import datetime as dt
import numpy as np

# MELT (WIDE --> LONG)
idcols = ['RegionID', 'RegionName', 'State', 'Metro']
mdf = pd.melt(df, id_vars=idcols + ['CountyName', 'SizeRank'], var_name='Year_Month', value_name='Sale_Amt').reset_index()

# CALCULATE HALF_YEAR STRING
mdf['Year_Month'] = pd.to_datetime(mdf['Year_Month'])
mdf['Half_Year'] = mdf['Year_Month'].dt.year.astype(str) + '$' + np.where(mdf['Year_Month'].dt.month <= 6, 1, 2).astype(str)

# FILTER DATASET BY DATE INTERVAL
mdf = mdf[mdf['Year_Month'].between('1998-01-01', '1999-12-31')]

#  GROUP BY AGGREGATION OF HOUSE SALES
mdf = mdf.groupby(idcols + ['Half_Year'])['Sale_Amt'].mean().reset_index()

# PIVOT (LONG --> WIDE)
pvtdf = mdf.pivot_table(index=idcols, columns='Half_Year', values='Sale_Amt', aggfunc=sum).reset_index()

Output

metros = ['Milwaukee', 'Tucson', 'Portland', 'Oklahoma City', 'Omaha', 'Albuquerque']
print(pvtdf[(pvtdf['RegionName'].isin(metros)) &  (pvtdf['Metro'].isin(metros))])

# Half_Year  RegionID   RegionName State        Metro         1998$1         1998$2         1999$1         1999$2
# 430            5976    Milwaukee    WI    Milwaukee   72366.666667   72583.333333   73916.666667   75750.000000
# 680            7481       Tucson    AZ       Tucson   94883.333333   96183.333333   98783.333333  100450.000000
# 1584          13373     Portland    OR     Portland  141166.666667  144733.333333  148183.333333  151766.666667
# 2923          23429  Albuquerque    NM  Albuquerque  121816.666667  122666.666667  123550.000000  124333.333333
# 5473          40152        Omaha    NE        Omaha   98950.000000  102450.000000  108016.666667  112116.666667

regarding melt, performance is usually more important than readability specially when a large dataset is involved, the dataset here is over 10,000 so melt might be ok, but if the size increase beyond that, it would be impractical — sgDysregulation, Jan 18 '17 at 16:55
I found this very useful, I also end up getting strange dates until I added the closed='left' parameter. However I don't understand how it works, could you elaborate a little bit more about this?: "cut-offs are 6/30 and 12/31 instead of 1/1 and 7/1", what is the meaning of these fractions? — Abimael Domínguez, Jan 24 '22 at 05:17
@AbimaelDomínguez, those fractions are dates in US month-first format: `MM/YY`. Using `6M` period in `resample` which slightly changes according to the `closed` parameter. — Parfait, Jan 24 '22 at 22:44

jezrael · Answer 2 · 2017-01-15T14:51:36.113

You can use groupby by new column names with custom format, because if use resample by 6M it return 2000 (it looks like bug):

c = df.columns.tolist()
cols = c[:7] 
cols.append(c[-1]) 
print (cols)
['RegionID', 'RegionName', 'State', 'Metro', 'CountyName', 'SizeRank', '1996-04', '2016-11']

#set to index all NOT date columns
tdf = df.set_index(df.columns.tolist()[:6])
tdf.columns = pd.to_datetime(tdf.columns).to_period('M')

# find the columns in the period 1998-2000
cols = tdf.columns
sel_cols = cols[(cols > '1997-12') & (cols < '2000')]
print (sel_cols)
PeriodIndex(['1998-01', '1998-02', '1998-03', '1998-04', '1998-05', '1998-06',
             '1998-07', '1998-08', '1998-09', '1998-10', '1998-11', '1998-12',
             '1999-01', '1999-02', '1999-03', '1999-04', '1999-05', '1999-06',
             '1999-07', '1999-08', '1999-09', '1999-10', '1999-11', '1999-12'],
            dtype='period[M]', freq='M')

#change format columns
a = tdf[sel_cols].rename(columns=lambda x: '{:}${:}'.format(x.year, [1, 2][x.quarter > 2]))
print (a.columns)
Index(['1998$1', '1998$1', '1998$1', '1998$1', '1998$1', '1998$1', '1998$2',
       '1998$2', '1998$2', '1998$2', '1998$2', '1998$2', '1999$1', '1999$1',
       '1999$1', '1999$1', '1999$1', '1999$1', '1999$2', '1999$2', '1999$2',
       '1999$2', '1999$2', '1999$2'],
      dtype='object')

#groupby by a with new format
mdf = a.groupby(a.columns, axis=1).mean().reset_index()

print (mdf.head())
   RegionID    RegionName State                           Metro    CountyName  \
0      6181      New York    NY                        New York        Queens   
1     12447   Los Angeles    CA  Los Angeles-Long Beach-Anaheim   Los Angeles   
2     17426       Chicago    IL                         Chicago          Cook   
3     13271  Philadelphia    PA                    Philadelphia  Philadelphia   
4     40326       Phoenix    AZ                         Phoenix      Maricopa   

   SizeRank         1998$1         1998$2         1999$1         1999$2  
0         1            NaN            NaN            NaN            NaN  
1         2  169183.333333  179166.666667  189116.666667  198466.666667  
2         3  117700.000000  121666.666667  125550.000000  133000.000000  
3         4   50550.000000   50650.000000   51150.000000   51866.666667  
4         5   97583.333333  101083.333333  104816.666667  108566.666667

Pandas: Compute the half-yearly average of monthly collected house sales

2 Answers2