pandas data mining from Eurostat

Question

I'm starting a work to analyse data from Stats Institutions like Eurostat using python, and so pandas. I found out there are two methods to get data from Eurostat.

• pandas_datareader: it seems very easy to use but I found some problems to get some specific data
• pandasdmx: I've found it a bit complicated but it seems a promising solution, but documentation is poor

I use a free Azure notebook, online service, but I don't think it will complicate more my situation.

Let me explain the problems for pandas_datareader. According to the pandas documentation, in the section API, there is this short documented package and it works. Apart from the shown example, that nicely works, a problem arises about other tables. For example, I can get data about European house price, which ID table is prc_hpi_a with this simple code:

import pandas_datareader.data as web
import datetime
df = web.DataReader('prc_hpi_a', 'eurostat')

But the table has three types of data about dwellings: TOTAL, EXISTING and NEW. I got only Existing dwellings and I don't know how to get the other ones. Do you have a solution for these types of filtering.

Secondly there is the path using pandasdmx. Here it is more complicated. My idea is to upload all data to a pandas DataFrame, and then I can analyse as I want. Easy to say, but I've not find many tutorials that explain this passage: upload data to pandas structures. For example, I found this tutorial, but I'm stuck to the first step, that is instantiate a client:

import pandasdmx
from pandasdmx import client
#estat=client('Eurostat', 'milk.db')

and it returns:

--------------------------------------------------------------------------- ImportError Traceback (most recent call last) in () 1 import pandasdmx ----> 2 from pandasdmx import client 3 estat=client('Eurostat', 'milk.db')

ImportError: cannot import name 'client'

What's the problem here? I've looked around but no answer to this problem

I also followed this tutorial:

from pandasdmx import Request
estat = Request('ESTAT')
metadata = estat.datastructure('DSD_une_rt_a').write()
metadata.codelist.iloc[8:18]
resp = estat.data('une_rt_a', key={'GEO': 'EL+ES+IE'}, params={'startPeriod': '2007'})
data = resp.write(s for s in resp.data.series if s.key.AGE == 'TOTAL')
data.columns.names
data.columns.levels
data.loc[:, ('PC_ACT', 'TOTAL', 'T')]

I got the data, but my purpose is to upload them to a pandas structure (Series, DataFrame, etc..), so I can handle easily according to my work. How to do that? Actually I did with this working line (below the previous ones):

s=pd.DataFrame(data)

But it doesn't work if I try to get other data tables. Let me explain with another example about the Harmonized Index Current Price table:

estat = Request('ESTAT')
metadata = estat.datastructure('DSD_prc_hicp_midx').write()
resp = estat.data('prc_hicp_midx')
data = resp.write(s for s in resp.data.series if s.key.COICOP == 'CP00')

It returns an error here, that is:

--------------------------------------------------------------------------- AttributeError Traceback (most recent call last) in () 2 metadata = estat.datastructure('DSD_prc_hicp_midx').write() 3 resp = estat.data('prc_hicp_midx') ----> 4 data = resp.write(s for s in resp.data.series if s.key.COICOP == 'CP00') 5 #metadata.codelist 6 #data.loc[:, ('TOTAL', 'INX_Q','EA', 'Q')]

~/anaconda3_501/lib/python3.6/site-packages/pandasdmx/api.py in getattr(self, name) 622 Make Message attributes directly readable from Response instance 623 ''' --> 624 return getattr(self.msg, name) 625 626 def _init_writer(self, writer):

AttributeError: 'DataMessage' object has no attribute 'data'

Why does it do not get data now? What's wrong now?

I lost almost a day looking around for some clear examples and explanations. Do you have some to propose? Is there a full and clear documentation? I found also this page with other examples, explaining the use of categorical schemes, but it is not for Eurostat (as explained at some point)

Both methods could work, apart from some explained issues, but I need also a suggestion to have a definitely method to use, to query Eurostat but also many other institutions like OECD, World Bank, etc... Could you guide me to a definitive and working solution, even if it is different for each institution?

Answer 1

That's my definitive answer to my question that works for each type of data collected from Eurostat. I post here because it can be useful for many.

Let me propose some examples. They produce three pandas series (EU_unempl,EU_GDP,EU_intRates) with data and correct time indexes

#----Unemployment Rate---------
dataEU_unempl=pd.read_json('http://ec.europa.eu/eurostat/wdds/rest/data/v2.1/json/en/ei_lmhr_m?geo=EA&indic=LM-UN-T-TOT&s_adj=NSA&unit=PC_ACT',typ='series',orient='table',numpy=True) #,typ='DataFrame',orient='table'
x=[]
for i in range(int(sorted(dataEU_unempl['value'].keys())[0]),1+int(sorted(dataEU_unempl['value'].keys(),reverse=True)[0])):
    x=numpy.append(x,dataEU_unempl['value'][str(i)])
EU_unempl=pd.Series(x,index=pd.date_range((pd.to_datetime((sorted(dataEU_unempl['dimension']['time']['category']['index'].keys())[(sorted(int(v) for v in dataEU_unempl['value'].keys())[0])]),format='%YM%M')), periods=len(x), freq='M')) #'1/1993'


#----GDP---------
dataEU_GDP=pd.read_json('http://ec.europa.eu/eurostat/wdds/rest/data/v2.1/json/en/namq_10_gdp?geo=EA&na_item=B1GQ&s_adj=NSA&unit=CP_MEUR',typ='series',orient='table',numpy=True) #,typ='DataFrame',orient='table'
x=[]
for i in range((sorted(int(v) for v in dataEU_GDP['value'].keys())[0]),1+(sorted((int(v) for v in dataEU_GDP['value'].keys()),reverse=True))[0]):
    x=numpy.append(x,dataEU_GDP['value'][str(i)])
EU_GDP=pd.Series(x,index=pd.date_range((pd.Timestamp(sorted(dataEU_GDP['dimension']['time']['category']['index'].keys())[(sorted(int(v) for v in dataEU_GDP['value'].keys())[0])])), periods=len(x), freq='Q'))


#----Money market interest rates---------
dataEU_intRates=pd.read_json('http://ec.europa.eu/eurostat/wdds/rest/data/v2.1/json/en/irt_st_m?geo=EA&intrt=MAT_ON',typ='series',orient='table',numpy=True) #,typ='DataFrame',orient='table'
x=[]
for i in range((sorted(int(v) for v in dataEU_intRates['value'].keys())[0]),1+(sorted((int(v) for v in dataEU_intRates['value'].keys()),reverse=True))[0]):
    x=numpy.append(x,dataEU_intRates['value'][str(i)])
EU_intRates=pd.Series(x,index=pd.date_range((pd.to_datetime((sorted(dataEU_intRates['dimension']['time']['category']['index'].keys())[(sorted(int(v) for v in dataEU_intRates['value'].keys())[0])]),format='%YM%M')), periods=len(x), freq='M'))

Answer 2

The general solution is to not rely on overly-specific APIs like datareader and instead go to the source. You can use datareader's source code as inspiration and as a guide for how to do it. But ultimately when you need to get data from a source, you may want to directly access that source and load the data.

One very popular tool for HTTP APIs is requests. You can easily use it to load JSON data from any website or HTTP(S) service. Once you have the JSON, you can load it into Pandas. Because this solution is based on general-purpose building blocks, it is applicable to virtually any data source on the Web (as opposed to e.g. pandaSDMX, which is only applicable to SDMX data sources).

Answer 3

Load with read_csv and multiple separators

The problem with eurostat data from the bulk download repository is that they are tab separated files where the first 3 columns are separated by commas. Pandas read_csv() can deal with mulitple separators as a regex if you specify engine="python". This works for some data sets, but the OP's dataset also contains flags, which cannot be ignored in the last column.

# Load the house price index from the Eurostat bulk download facility
import pandas
code = "prc_hpi_a"
url = f"https://ec.europa.eu/eurostat/estat-navtree-portlet-prod/BulkDownloadListing?sort=1&file=data%2F{code}.tsv.gz"                                                                                                                        # Pandas.read_csv could almost read it directly with a multiple separator
df = pandas.read_csv(url, sep=",|\t| [^ ]?\t", na_values=":", engine="python")

# But the last column is a character column instead of a numeric because of the
# presence of a flag ": c" illustrated in the last line of the table extract
# below

# purchase,unit,geo\time\t  2006\t  2005
#   DW_EXST,I10_A_AVG,AT\t     :\t     :
#   DW_EXST,I10_A_AVG,BE\t 83.86\t 75.16
#   DW_EXST,I10_A_AVG,BG\t 87.81\t 76.56
#   DW_EXST,I10_A_AVG,CY\t     :\t     :
#   DW_EXST,I10_A_AVG,CZ\t     :\t     :
#   DW_EXST,I10_A_AVG,DE\t100.80\t101.10
#   DW_EXST,I10_A_AVG,DK\t113.85\t 91.79
#   DW_EXST,I10_A_AVG,EE\t156.23\t 98.69
#   DW_EXST,I10_A_AVG,ES\t109.68\t     :
#   DW_EXST,I10_A_AVG,FI\t   : c\t   : c

Load with the eurostat package

There is also a python package called eurostat which makes it possible to search and load data set from the bulk facility into pandas data frames. Load 2 different monthly exchange rate data sets:

import eurostat
df1 = eurostat.get_data_df(code)

The table of content of the bulk download facility can be read with

toc_url = "https://ec.europa.eu/eurostat/estat-navtree-portlet-prod/BulkDownloadListing?sort=1&file=table_of_contents_en.txt"
toc2 = pandas.read_csv(toc_url, sep="\t")
# Remove white spaces at the beginning and end of strings
toc2 = toc2.applymap(lambda x: x.strip() if isinstance(x, str) else x)

or with

toc = eurostat.get_toc_df()
toc0 = (eurostat.subset_toc_df(toc, "exchange"))

The last line searches for the datasets that have "exchange" in their title

Reshape to long format

It might be useful to reshape the eurostat data to long format with

    if any(df.columns.str.contains("time")):
        time_column = df.columns[df.columns.str.contains("time")][-1]
        # Id columns are before the time columns
        id_columns = df.loc[:, :time_column].columns
        df = df.melt(id_vars=id_columns, var_name="period", value_name="value")
        # Remove "\time" from the rightmost column of the index
        df = df.rename(columns=lambda x: re.sub(r"\\time", "", x))