Python library and command line tool for parsing pdf bank statements
Inspired by https://github.com/antonburger/pdf2csv
Banks generally send account statements in pdf format. These pdfs are often encrypted, the pdf format is difficult to extract tables from and when you finally get the table out it's in a non tidy format. This package aims to help by providing a library of functions and a set of command line tools for converting these statements into more useful formats such as csv files and pandas dataframes.
pip install pdf-statement-reader
This package uses tabula-py under the hood, which itself is a wrapper for tabula-java. You thus need to have java installed for it to work. If you have any errors complaining about java, checkout out the tabula-py page for troubleshooting advice.
In the future, we hope to move to a pure python implementation.
The package provides a command line application psr
Usage: psr [OPTIONS] COMMAND [ARGS]...
Utility for reading bank and other statements in pdf form
Options:
--help Show this message and exit.
Commands:
bulk Bulk converts all files in a folder
decrypt Decrypts a pdf file Uses pikepdf to open an encrypted pdf file...
pdf2csv Converts a pdf statement to a csv file using a given format
validate Validates the csv statement rolling balance
PDF files are notoriously difficult to extract data from. (Here's a nice blog post on why). For a really good semi-manual GUI solution, check out tabula. In fact this package uses tabula's pdf parsing library under the hood.
Since bank statements are generally of the same (if inconvenient) format, we can set up a configuration to tell the tool how to grab the data.
For each type of bank statement, the exact format will be different. A config file holds the instructions for how to process the raw pdf. For now the only config supported is for Cheque account statements from Absa bank in South Africa.
To set up a different statement, you can simply add a new config file and then tell the psr tool to use it. These config files are stored in a folder structure as follows:
config > [country code] > [bank] > [statement type].json
So for example the default config is stored in
config > za > absa > cheque.json
The config spec is a code of the form
[country code].[bank].[statement type]
Once again for the default this will be
za.absa.cheque
The configuration file itself is in JSON format. Here's the Absa cheque account one with some commentary to explain what each field does.
{
// Describes the page layout that should be scanned
"layout": {
// Default layout for all pages not otherwise defined
"default": {
// The page coordinates in containing the table in pts
// [top, left, bottom, right]
"area": [280, 27, 763, 576],
// The right x coordinate of each column in the table
"columns": [83, 264, 344, 425, 485, 570]
},
// Layout for the first page
"first": {
"area": [480, 27, 763, 576],
"columns": [83, 264, 344, 425, 485, 570]
}
},
// The columns names to be used as they exactly appear
// in the statement
"columns": {
"trans_date": "Date",
"trans_type": "Transaction Description",
"trans_detail": "Transaction Detail",
"debit": "Debit Amount",
"credit": "Credit Amount",
"balance": "Balance"
},
// The order of the columns to be output in the csv
"order": [
"trans_date",
"trans_type",
"trans_detail",
"debit",
"credit",
"balance"
],
// Specifies any cleaning operations required
"cleaning": {
// Convert these columns to numeric
"numeric": ["debit", "credit", "balance"],
// Convert these columns to date
"date": ["trans_date"],
// Use this date format to parse any date columns
"date_format": "%d/%m/%Y",
// For cases where the transaction detail is stored
// in the next line below the transaction type
"trans_detail": "below",
// Only keep the rows where these columns are populated
"dropna": ["balance"]
}
}These were the configuration options that were required for the default format. It is envisaged that as more formats are added, the list of options will grow.
A key part in setting up a new configuration is getting the page coordinates for the area and columns. The easiest way to do this is to run the tabula GUI, autodetect the page areas, save the settings as a template, then download and inspect json template file. It's not a one-to-one mapping to the psr config but hopefully it will be a good starting point.
Usage: psr decrypt [OPTIONS] INPUT_FILENAME [OUTPUT_FILENAME]
Decrypts a pdf file
Uses pikepdf to open an encrypted pdf file and then save the unencrypted
version. If no output_filename is specified then overwrites the original
file.
Options:
-p, --password TEXT The pdf encryption password. If not supplied, it will
be requested at the prompt
--help Show this message and exit.
Usage: psr pdf2csv [OPTIONS] INPUT_FILENAME [OUTPUT_FILENAME]
Converts a pdf statement to a csv file using a given format
Options:
-c, --config TEXT The configuration code defining how the file should be
parsed [default: za.absa.cheque]
--help Show this message and exit.
Usage: psr validate [OPTIONS] INPUT_FILENAME
Validates the csv statement rolling balance
Options:
-c, --config TEXT The configuration code defining how the file should be
parsed [default: za.absa.cheque]
--help Show this message and exit.
Usage: psr bulk [OPTIONS] FOLDER
Bulk converts all files in a folder
Options:
-c, --config TEXT The configuration code defining how the file
should be parsed [default: za.absa.cheque]
-p, --password TEXT The pdf encryption password. If not supplied, it
will be requested at the prompt
-d, --decrypt-suffix TEXT The suffix to append to the decrypted pdf file
when created [default: _decrypted]
-k, --keep-decrypted Keep the a copy of the decrypted file. It is
removed by default
-v, --verbose Print verbose output while running
--help Show this message and exit.