Command Line Tool¶
clkhash includes a command line tool which can be used to interact without writing Python code.
The primary use case is to encode personally identifiable data from a csv into Cryptographic Longterm Keys.
The command line tool can be accessed in two equivalent ways:
- Using the
clkutilscript which gets added to your path during installation. - directly running the python module with
python -m clkhash.
A list of valid commands can be listed with the --help argument:
$ clkutil --help
Usage: clkutil [OPTIONS] COMMAND [ARGS]...
This command line application allows a user to hash their data into
cryptographic longterm keys for use in private comparison.
This tool can also interact with a entity matching service; creating new
mappings, uploading locally hashed data, watching progress, and retrieving
results.
Example:
clkutil hash private_data.csv secretkey1 secretkey2 schema.json
output-clks.json
All rights reserved Confidential Computing 2016.
Options:
--version Show the version and exit.
--help Show this message and exit.
Commands:
benchmark carry out a local benchmark
create create a run on the entity service
create-project create a linkage project on the entity service
delete delete a run on the anonlink entity service
delete-project delete a project on the anonlink entity service
describe show distribution of clk popcounts
generate generate random pii data for testing
generate-default-schema get the default schema used in generated random PII
hash generate hashes from local PII data
results fetch results from entity service
status get status of entity service
upload upload hashes to entity service
validate-schema validate linkage schema
Command specific help¶
The clkutil tool has help pages for all commands built in - simply append --help
to the command.
Hashing¶
The command line tool clkutil can be used to hash a csv file of personally identifiable information.
The tool needs to be provided with keys and a Linkage Schema; it will output a file containing
json serialized hashes.
$ clkutil hash --help
Usage: clkutil hash [OPTIONS] PII_CSV KEYS... SCHEMA CLK_JSON
Process data to create CLKs
Given a file containing CSV data as PII_CSV, and a JSON document defining
the expected schema, verify the schema, then hash the data to create CLKs
writing them as JSON to CLK_JSON. Note the CSV file should contain a
header row - however this row is not used by this tool.
It is important that the keys are only known by the two data providers.
Two words should be provided. For example:
$clkutil hash pii.csv horse staple pii-schema.json clk.json
Use "-" for CLK_JSON to write JSON to stdout.
Options:
-q, --quiet Quiet any progress messaging
--no-header Don't skip the first row
--check-header BOOLEAN If true, check the header against the schema
--validate BOOLEAN If true, validate the entries against the schema
--help Show this message and exit.
Example¶
Assume a csv (fake-pii.csv) contains rows like the following:
0,Libby Slemmer,1933/09/13,F
1,Garold Staten,1928/11/23,M
2,Yaritza Edman,1972/11/30,F
It can be hashed using clkutil with:
$ clkutil hash --schema simple-schema.json fake-pii.csv horse staple clk.json
Where:
horse stapleis the two part secret key that both participants will use to hash their data.simple-schema.jsonis a Linkage Schema describing how to hash the csv. E.g, ignore the first column, use bigram tokens of the name, use positional unigrams of the date of birth etc.clk.jsonis the output file.
Describing¶
Users can inspect the distribution of the number of bits set in CLKs by using the describe command.
$ clkutil describe --help
Usage: clkutil describe [OPTIONS] CLK_JSON
show distribution of clk's popcounts
Options:
--help Show this message and exit.
Example¶
$ clkutil describe example_clks_a.json
339| oo
321| ooo
303| ooo
285| ooo o
268| oooooo
250| oooooooo
232| oooooooo
214| ooooooooo
196| o ooooooooo o
179| o ooooooooooo
161| oooooooooooooo
143| ooooooooooooooo
125| oooooooooooooooo
107| oooooooooooooooooo
90| ooooooooooooooooooooo
72| oooooooooooooooooooooo
54| oooooooooooooooooooooooo
36| ooooooooooooooooooooooooooo
18| oooooooooooooooooooooooooooooooo
1| o o ooooooooooooooooooooooooooooooooooooooooooooooooooo oo
------------------------------------------------------------
4 4 4 4 4 4 4 4 5 5 5 5 5 5 5 5 5 6 6 6 6 6 6 6 6 6 7 7 7 7
1 2 3 4 5 6 7 9 0 1 2 3 4 5 7 8 9 0 1 2 3 5 6 7 8 9 0 1 3 4
0 1 2 4 5 7 8 0 1 2 4 5 7 8 0 1 2 4 5 7 8 0 1 2 4 5 7 8 0 1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4 8 3 7 1 6 0 4 9 3 7 2 6 0 5 9 3 8 2 6 1 5 9 4 8 2 7 1 5
-------------------------
| Summary |
-------------------------
| observations: 5000 |
| min value: 410.000000 |
| mean : 601.571600 |
| max value: 753.000000 |
-------------------------
Note
It is an indication of problems in the hashing if the distribution is skewed towards no bits set or all bits set. Consult the Tutorial for CLI tool clkhash for further details.
Data Generation¶
The command line tool has a generate command for generating fake pii data.
$ clkutil generate --help
Usage: clkutil generate [OPTIONS] [SIZE] OUTPUT
Generate fake PII data for testing
Options:
-s, --schema FILENAME
--help Show this message and exit.
$ clkutil generate 1000 fake-pii-out.csv
$ head -n 4 fake-pii-out.csv
INDEX,NAME freetext,DOB YYYY/MM/DD,GENDER M or F
0,Libby Slemmer,1933/09/13,F
1,Garold Staten,1928/11/23,M
2,Yaritza Edman,1972/11/30,F
A corresponding hashing schema can be generated as well:
$ clkutil generate-default-schema schema.json
$ cat schema.json
{
"version": 1,
"clkConfig": {
"l": 1024,
"k": 30,
"hash": {
"type": "doubleHash"
},
"kdf": {
"type": "HKDF",
"hash": "SHA256",
"salt": "SCbL2zHNnmsckfzchsNkZY9XoHk96P/G5nUBrM7ybymlEFsMV6PAeDZCNp3rfNUPCtLDMOGQHG4pCQpfhiHCyA==",
"info": "c2NoZW1hX2V4YW1wbGU=",
"keySize": 64
}
},
"features": [
{
"identifier": "INDEX",
"format": {
"type": "integer"
},
"hashing": {
"ngram": 1,
"weight": 0
}
},
{
"identifier": "NAME freetext",
"format": {
"type": "string",
"encoding": "utf-8",
"case": "mixed",
"minLength": 3
},
"hashing": {
"ngram": 2,
"weight": 0.5
}
},
{
"identifier": "DOB YYYY/MM/DD",
"format": {
"type": "string",
"encoding": "ascii",
"description": "Numbers separated by slashes, in the year, month, day order",
"pattern": "(?:\\d\\d\\d\\d/\\d\\d/\\d\\d)\\Z"
},
"hashing": {
"ngram": 1,
"positional": true
}
},
{
"identifier": "GENDER M or F",
"format": {
"type": "enum",
"values": ["M", "F"]
},
"hashing": {
"ngram": 1,
"weight": 2
}
}
]
}
Benchmark¶
A quick hashing benchmark can be carried out to determine the rate at which the current machine can generate 10000 clks from a simple schema (data as generated above):
python -m clkhash.cli benchmark
generating CLKs: 100% 10.0K/10.0K [00:01<00:00, 7.72Kclk/s, mean=521, std=34.7]
10000 hashes in 1.350489 seconds. 7.40 KH/s
As a rule of thumb a single modern core will hash around 1M entities in about 20 minutes.
Note
Hashing speed is effected by the number of features and the corresponding schema. Thus these numbers will, in general, not be a good predictor for the performance of a specific use-case.
The output shows a running mean and std deviation of the generated clks’ popcounts. This can be used as a basic sanity check - ensure the CLK’s popcount is not around 0 or 1024.
Interaction with Entity Service¶
There are several commands that interact with a REST api for carrying out privacy preserving linking. These commands are:
- status
- create-project
- create
- upload
- results
See also the Tutorial for CLI.
