decoy

Decoy

Documenting and experimenting on the Decoy Selection Algorithm.

Strategy

• New unit tests shall be written, that cover the happy path and edge cases of the decoy algorithm
• The nevralgic parts of the system shall be first experimentally probed and the results shall be gathered
• An alternative Python implementation shall be prepared, since Python offers many statistical analysis and visualization tools. The correspondence between this and the original implementation shall be confirmed via the Kolmogorov-Smirnov test
• Based on the Python implementation, and R implementation shall be created and its results are confirmed via the same statistical tests
• With the R implementation at hand, a mathematical expression of the decoy algo shall be derived

Results

Except for the last point, namely the mathematical expression, the research project already delivers satisfying results.

Visualization

Gamma pickers

Comparison of the final Gamma Picker's result. The differences may result at least from the different random samplers chosen in both implementations.

Gamma distributions

The point of this comparison is to make sure, that the parameters of the distribution aren't misinterpreted across the two APIs.

Unit tests

Usage

Plots

Use these steps to recreate the above plots:

BRANCH=mrl-decoy-descr-tests

git clone --recursive https://github.com/mj-xmr/monero-mj.git
cd monero-mj
git pull origin $BRANCH
git checkout -b $BRANCH
make release-all 	# Build with tests, but don't run them
cd build/Linux/$BRANCH/release  # Linux or whatever host system
tests/unit_tests/unit_tests --gtest_filter=decoy.*  # Run the relevant tests only and generate the data
cd ..

git clone --recursive https://github.com/mj-xmr/monero-mrl-mj.git
cd monero-mrl-mj
decoy/python/mrl_decoy_plot.py
decoy/python/mrl_decoy_reimpl.py

Statistical tests

In order to generate an abundence of test data, being able to perform statistical tests on them, just run the following script:

cd decoy/python
bash mrl-decoy-loop.sh

As a result of running this script, there are going to be sets of data generated in: /tmp/monero/decoy. These data can be compared against each other via:

# Run Kolmogorov-Smirnov test:
./mrl_decoy_ks.py \
-d1 /tmp/monero/decoy/decoy-1/python/picks_raw_py_mul_length_100000.csv \
-d2 /tmp/monero/decoy/decoy-2/cpp/mrl_pick_mul_length_100000.csv

./mrl_decoy_ks.py \
-d1 /tmp/monero/decoy/decoy-1/python/picks_raw_py_mul_length_5.csv \
-d2 /tmp/monero/decoy/decoy-2/cpp/mrl_pick_mul_length_5.csv

... and so on. The number suffix of the data files is a multiple of the minimum length, at which the algorithm may start to deliver expected results.