A toolbox and library of ETL, data quality, and data analysis tools
It’s neither an enterprise ETL tool, nor an enterprise analysis, reporting, or data mining tool. It’s intended to be an easily-adopted tool for technical analysts that combines the most useful subset of data transformation and analysis capabilities necessary to do 80% of the work. Its open source python codebase allows it to be easily extended to with custom code to handle that always challenging last 20%.
Current Status: Strong support for easy analysis, simple transformations of csv files, ability to create data dictionaries, change detection, and emerging data quality capabilities.
More info is on the DataGristle wiki here: https://github.com/kenfar/DataGristle/wiki
~ $ pip install datagristle ~
Splits a csv file into two separate files based on how records pass or fail validation checks: - Field count - checks the number of fields in each record against the number required. The correct number of fields can be provided in an argument or will default to using the number from the first record. - Schema - uses csv file requirements defined in a json-schema file for quality checking. These requirements include the number of fields, and for each field - the type, min & max length, min & max value, whether or not it can be blank, existance within a list of valid values, and finally compliance with a regex pattern. The output can just be the return code (0 for success, 1+ for errors), can be some high level statistics, or can be the csv input records split between good and erroneous files. Output can also be limited to a random subset. Examples: $ gristle_validator sample.csv -f 3 Prints all valid input rows to stdout, prints all records with other than 3 fields to stderr along with an extra final field that describes the error. $ gristle_validator sample.csv Prints all valid input rows to stdout, prints all records with other than the same number of fields found on the first record to stderr along with an extra final field that describes the error. $ gristle_validator sample.csv -d '|' --hasheader Same comparison as above, but in this case the file was too small or complex for the pgm to automatically determine csv dialect, so we had to explicitly give that info to program. $ gristle_validator sample.csv --outgood sample_good.csv --outerr sample_err.csv Same comparison as above, but explicitly splits good and bad data into separate files. $ gristle_validator sample.csv --randomout 1 Same comparison as above, but only writes a random 1% of data out. $ gristle_validator sample.csv --silent Same comparison as above, but writes nothing out. Exit code can be used to determine if any bad records were found. $ gristle_validator sample.csv --validschema sample_schema.csv The above command checks both field count as well as validations described in the sample_schema.csv file. Here's an example of what that file might look like: items: - title: rowid blank: False required: True dg_type: integer dg_minimum: 1 dg_maximum: 60 - title: start_date blank: False minLength: 8 maxLength: 10 pattern: '[0-9]*/[0-9]*/[1-2][0-9][0-9][0-9]' - title: location blank: False minLength: 2 maxLength: 2 enum: ['ny','tx','ca','fl','wa','ga','al','mo']
Extracts subsets of input files based on user-specified columns and rows. The input csv file can be piped into the program through stdin or identified via a command line option. The output will default to stdout, or redirected to a filename via a command line option. The columns and rows are specified using python list slicing syntax - so individual columns or rows can be listed as can ranges. Inclusion or exclusion logic can be used - and even combined. Examples: $ gristle_slicer sample.csv Prints all rows and columns $ gristle_slicer sample.csv -c":5, 10:15" -C 13 Prints columns 0-4 and 10,11,12,14 for all records $ gristle_slicer sample.csv -C:-1 Prints all columns except for the last for all records $ gristle_slicer sample.csv -c:5 -r-100 Prints columns 0-4 for the last 100 records $ gristle_slicer sample.csv -c:5 -r-100 -d'|' --quoting=quote_all Prints columns 0-4 for the last 100 records, csv dialect info (delimiter, quoting) provided manually) $ cat sample.csv | gristle_slicer -c:5 -r-100 -d'|' --quoting=quote_all Prints columns 0-4 for the last 100 records, csv dialect info (delimiter, quoting) provided manually)
Creates a frequency distribution of values from columns of the input file and prints it out in columns - the first being the unique key and the last being the count of occurances. Examples: $ gristle_freaker sample.csv -d '|' -c 0 Creates two columns from the input - the first with unique keys from column 0, the second with a count of how many times each exists. $ gristle_freaker sample.csv -d '|' -c 0 --sortcol 1 --sortorder forward --writelimit 25 In addition to what was described in the first example, this example adds sorting of the output by count ascending and just prints the first 25 entries. $ gristle_freaker sample.csv -d '|' -c 0 --sampling_rate 3 --sampling_method interval In addition to what was described in the first example, this example adds a sampling in which it only references every third record. $ gristle_freaker sample.csv -d '|' -c 0,1 Creates three columns from the input - the first two with unique key combinations from columns 0 & 1, the third with the number of times each combination exists. $ gristle_freaker sample.csv -d '|' -c -1 Creates two columns from the input - the first with unique keys from the last column of the file (negative numbers wrap), then a second with the number of times each exists. $ gristle_freaker sample.csv -d '|' --columntype all Creates two columns from the input - all columns combined into a key, then a second with the number of times each combination exists. $ gristle_freaker sample.csv -d '|' --columntype each Unlike the other examples, this one performs a separate analysis for every single column of the file. Each analysis produces three columns from the input - the first is a column number, second is a unique value from the column, and the third is the number of times that value appeared. This output is repeated for each column.
Displays a single record of a file, one field per line, with field names displayed as labels to the left of the field values. Also allows simple navigation between records. Examples: $ gristle_viewer sample.csv -r 3 Presents the third record in the file with one field per line and field names from the header record as labels in the left column. $ gristle_viewer sample.csv -r 3 -d '|' -q quote_none In addition to what was described in the first example this adds explicit csv dialect overrides.
Analyzes the structures and contents of csv files in the end producing a report of its findings. It is intended to speed analysis of csv files by automating the most common and frequently-performed analysis tasks. It's useful in both understanding the format and data and quickly spotting issues. Examples: $ gristle_determinator japan_station_radiation.csv This command will analyze a file with radiation measurements from various Japanese radiation stations. File Structure: format type: csv field cnt: 4 record cnt: 100 has header: True delimiter: csv quoting: False skipinitialspace: False quoting: QUOTE_NONE doublequote: False quotechar: " lineterminator: '\n' escapechar: None Field Analysis Progress: Analyzing field: 0 Analyzing field: 1 Analyzing field: 2 Analyzing field: 3 Fields Analysis Results: ------------------------------------------------------ Name: station_id Field Number: 0 Wrong Field Cnt: 0 Type: timestamp Min: 1010000001 Max: 1140000006 Unique Values: 99 Known Values: 99 Top Values not shown - all values are unique ------------------------------------------------------ Name: datetime_utc Field Number: 1 Wrong Field Cnt: 0 Type: timestamp Min: 2011-02-28 15:00:00 Max: 2011-02-28 15:00:00 Unique Values: 1 Known Values: 1 Top Values: 2011-02-28 15:00:00 x 99 occurrences ------------------------------------------------------ Name: sa Field Number: 2 Wrong Field Cnt: 0 Type: integer Min: -999 Max: 52 Unique Values: 35 Known Values: 35 Mean: 2.45454545455 Median: 38.0 Variance: 31470.2681359 Std Dev: 177.398613681 Top Values: 41 x 7 occurrences 42 x 7 occurrences 39 x 6 occurrences 37 x 5 occurrences 46 x 5 occurrences 17 x 4 occurrences 38 x 4 occurrences 40 x 4 occurrences 45 x 4 occurrences 44 x 4 occurrences ------------------------------------------------------ Name: ra Field Number: 3 Wrong Field Cnt: 0 Type: integer Min: -888 Max: 0 Unique Values: 2 Known Values: 2 Mean: -556.121212121 Median: -888.0 Variance: 184564.833792 Std Dev: 429.610095077 Top Values: -888 x 62 occurrences 0 x 37 occurrences
gristle_differ compares two files, typically an old and a new file, based on explicit keys in a way that is far more accurate than diff. It can also compare just subsets of columns, and perform post-delta transforms to populate fields with static values, values from other fields, variables from the command line, or incrementing sequence numbers. Examples: $ gristle_differ file0.dat file1.dat --key-cols '0, 2' --ignore_cols '19,22,33' - Sorts both files on columns 0 & 2 - Dedupes both files on column 0 - Compares all fields except fields 19,22, and 23 - Automatically determines the csv dialect - Produces the following files: - file1.dat.insert - file1.dat.delete - file1.dat.same - file1.dat.chgnew - file1.dat.chgold $ gristle_differ file0.dat file1.dat --key-cols '0' --compare_cols '1,2,3,4,5,6,7' -d '|' - Sorts both files on columns 0 - Dedupes both files on column 0 - Compares fields 1,2,3,4,5,6,7 - Uses '|' as the field delimiter - Produces the same output file names as example 1. $ gristle_differ file0.dat file1.dat --config-fn ./foo.yml \ --variables batchid:919 --variables pkid:82304 - Produces the same output file names as example 1. - But in this case it gets the majority of its configuration items from the config file ('foo.yml'). This could include key columns, comparison columns, ignore columns, post-delta transformations, and other information. - The two variables options are used to pass in user-defined variables that can be referenced by the post-delta transformations. The batchid will get copied into a batch_id column for every file, and the pkid is a sequence that will get incremented and used for new rows in the insert, delete and chgnew files.
Gristle_metadata provides a command-line interface to the metadata database. It's mostly useful for scripts, but also useful for occasional direct command-line access to the metadata. Examples: $ gristle_metadata --table schema --action list Prints a list of all rows for the schema table. $ gristle_metadata --table element --action put --prompt Allows the user to input a row into the element table and prompts the user for all fields necessary.
Gristle_md_reporter allows the user to create data dictionary reports that combine information about the collection and fields along with field value descriptions and frequencies. Examples: $ gristle_md_reporter --report datadictionary --collection_id 2 Prints a data dictionary report of collection_id 2. $ gristle_md_reporter --report datadictionary --collection_name presidents Prints a data dictionary report of the president collection. $ gristle_md_reporter --report datadictionary --collection_id 2 --field_id 3 Prints a data dictionary report of the president collection, only shows field-level information for field_id 3.
Gristle_dir_merger consolidates directory structures of files. Is both fast and flexible with a variety of options for choosing which file to use based on full (name and md5) and partial matches (name only) . Examples $ gristle_dir_merger /tmp/foo /data/foo - Compares source of /tmp/foo to dest of /data/foo. - Files will be consolidated into /data/foo, and deleted from /tmp/foo. - Comparison will be: match-on-name-and-md5 (default) - Full matches will use: keep_dest (default) - Partial matches will use: keep_newest (default) - Bottom line: this is what you normally want. $ gristle_dir_merger /tmp/foo /data/foo --dry-run - Same as the first example - except it only prints what it would do without actually doing it. - Bottom line: this is a good step to take prior to running it for real. $ gristle_dir_merger /tmp/foo /data/foo -r - Same as the first example - except it runs recursively through the directories. $ gristle_dir_merger /tmp/foo /data/foo --on-partial-match keep-biggest - Comparison will be: match-on-name-and-md5 (default) - Full matches will use: keep_dest (default) - Partial matches will use: keep_biggest (override) - Bottom line: this is a good combo if you know that some files have been modified on both source & dest, and newest isn't the best. $ gristle_dir_merger /tmp/foo /data/foo --match-on-name-only --on-full-match keep-source - Comparison will be: match-on-name-only (override) - Full matches will use: keep_source (override) - Bottom line: this is a good way to go if you have files that have changed in both directories, but always want to use the source files.