зеркало из https://github.com/mozilla/rappor.git
0d8ad6d54f
- It can be used in applications, and follows the same choices as the
C++ (and Java) client libraries: MD5 to choose the bloom filter bits,
and HMAC-SHA256 with client secret for the PRR. (rappor.py)
- hash_candidates.py and rappor_sim.py use the rappor.get_bloom_bits
API, which is more efficient since it calls the hash function once
instead of 'h' times.
- Make the randomness interface a simpler "IrrRand" interface, which has
2 implementations: pure Python and C. (The PRR always uses the same
randomness.)
- regtest.sh: Instead of a boolean for 'fast_counts', we have 3 modes:
Python client, C++ client, and 'fast_counts'. The same tests can be
run against each implementation. The results are put in different
directories, so they don't overwite each other.
- Rename gen_reports.R to gen_true_values.R (since the "reports" are
typically the data transformed by RAPPOR).
- Change it to output the client and assign the cohort like this:
client,cohort,value
c1,1,v26
c1,1,v27
c2,2,v8
c2,2,v9
The cohort is stable per client, but the value is different.
- Add unit test
- tests/rappor_sim.py: New test protocol. Read true_values CSV input.
The CSV output contains the bloom filter, PRR, and IRR for easier
debugging.
- compare_dist.R loads the true values and computes the histograms,
rather than rappor_sim.py having to do it. This makes it easier to
use the test harness with C++ and other clients.
- demo.sh: quickly run equivalent tests for each mode.
- regtest.sh: show the implementation tested
|
||
|---|---|---|
| .. | ||
| cpp | ||
| python | ||