muid 0.0.7

Memorable Unique Identifier

muid

Mine Memorable Unique Identifiers

Wait you say ... this cannot make sense!

Memorable unique identifiers are a provocative misnomer. When generating unique identifiers such as privately used keys, memorability is antithetical to uniqueness.

MUIDs might be termed "hash-memorable" identifiers. They form a subset of UUIDs whose SHA-1 hashes are memorable, in a manner in which this open source code makes precise.

Usage

As per https://muid.readthedocs.io/en/latest/ ...

Install

pip install muid

Just like uuid.uuid4() ... only it takes longer

import muid
key  = muid.muid4()  

The SHA-1 hash is memorable

print( muid.mhash(key) )    

Don't see it yet? Look closer. Here's my example:

f01dab1e-ca70-403a-a0c7-00f6c29596c4

And if that isn't clear already, then:

>>print( muid.mnemonic( key ) )
Foldable Cat  

The call muid.mnemonic uses a corpus of readable hex-like scrabble words to infer that only the first 11 characters are intended to be memorable.

Verificaton

muid.verify(key,min_len=7)

Mining

It is trivial to mine for MUIDs and the rates are way way better.

muid.mine()

This defaults to the official difficulty level:

min_len = int(requests.get("http://www.microprediction.com/config.json").json()['min_len'])

Cashing in

Currently you can sell MUID's by establishing an Algorithmia account at https://algorithmia.com/signup and then supplying MUID's to one of the following buyers.

Difficulty	Bid	Buyer
min_len=11	7c	https://algorithmia.com/algorithms/microprediction/mverify

The difficulty of 11 is only a guide and may have been updated. You can get it from:

min_len = int(requests.get("http://www.microprediction.com/config.json").json()['min_len'])

Good luck!

Example application

Memorable unique identifiers are used at www.microprediction.com to circumvent the need to maintain a lookup between user keys and public user identities. New users burn themselves a has-memorable private identity and the memorable part of the hash appears on leaderboards.

The more difficult the mnemonic (i.e. longer) the more drawdown leeway they are granted.

The benefits of using MUIDs in this context are:

• No need for central registration or key provision
• No ability to generate vast numbers of keys
• Enforced computational timeout after drawdown makes other Sybil style attacks harder.

We hope you have your own uses and would love to hear about them. Many applications can benefit from "one less join".

Discussion

https://github.com/microprediction/muid/issues or https://algorithmia.com/algorithms/microprediction/mverify/discussion

Technical details

Hash

We use a version of SHA-1 hash, namely

code = str(uuid.uuid5(uuid.NAMESPACE_DNS, key))

which takes UUID's to string representation of the hash of the UUID.

Readable hex

The image of mhash contains only strings with characters a,b,e,d,e,f plus digits. That's not such a fun game of scrabble so muid introduces the notion of readable hex strings. These are the image under the map which swaps out characters as follows:

Hex	Readable
5	s
1	l
7	t
0	o

Collisions are a non-issue

It is well appreciated that approximately 2.71 quintillion uuid4() can be generated before the risk of collision exceeds fifty percent (Wikipedia).

Collision probabilities for muid

Because MUID's are a tiny subset of UUID's, the collision probabilities per muid are higher but still small typically.

• With min_len=8 say, then there will typically be a few million attempts per muid generated. Thus one can generate a trillion muid before any there is any risk of collision in the underlying uuid's (whether used or not) because a few quintillion / a few million = one trillion.

This is a defensive calculation.

It is also a somewhat irrelevant calculation and there is a stronger reason not to worry.

Your users will be limited in their capacity to produce muid's in a way that they are not limited with uuid's. A moment's reflection should convince you that the computational capacity required to create MUID collisions over any interval of time is at least as large as the computational capacity employed to create UUID collisions.

So stop worrying and love MUIDs.

For the true worry warts ...

If you are not convinced by logic, in place of the default method=uuid.uuid4 you can supply a method of generating candidate unique identifiers with even longer string representations - for example concatenating a UUID with the last min_len characters of another independent UUID.

Scrabble fans

If you enjoy generating words using vowels a,c,e,o and consonants c,b,d,f,s,l,t then please do contribute pull requests for https://github.com/microprediction/muid/blob/master/muid/corpus.py