Change the name from jaccard-strong to haussmann. NFC

doc/source/api/metrics.rst

@@ -6,10 +6,10 @@ pairwise_distances
 
 .. autofunction:: qbindiff.passes.metrics.pairwise_distances
 
-jaccard_strong
---------------
+haussmann
+---------
 
-.. autofunction:: qbindiff.passes.metrics.jaccard_strong
+.. autofunction:: qbindiff.passes.metrics.haussmann
 
 canberra_distances
 ------------------

doc/source/basicex.rst

@@ -25,7 +25,7 @@ BinExport is more flexible since it can be used with IDA, Ghidra and BinaryNinja
 
 * `-n`, `--normalize` Normalize the Call Graph by removing some of the edges/nodes that should worsen the diffing result. **WARNING:** it can potentially lead to a worse matching. To know the details of the normalization step look at {ref}`normalization`
 
-* `-d`, `--distance` Set the default distance that should be used by the features. The possible values are `canberra, correlation, cosine, euclidean, jaccard-strong`. The default one is `canberra`. To know the details of the jaccard-strong distance look here {ref}`jaccard-strong`
+* `-d`, `--distance` Set the default distance that should be used by the features. The possible values are `canberra, correlation, cosine, euclidean, haussmann`. The default one is `canberra`. To know the details of the haussmann distance look here {ref}`haussmann`
 
 * `-s`, `--sparsity-ratio` Set the density of the similarity matrix. This will loose some information (hence decrease accuracy) but it will also increase the performace. `0.999` means that the 99.9% of the matrix will be filled with zeros. The default value is `0.75`
 
@@ -65,7 +65,7 @@ Some examples are displayed below :
          -f dat \
          -f cst \
          -f addr:0.01 \
-         -d jaccard-strong -s 0.999 -sr \
+         -d haussmann -s 0.999 -sr \
          -ff bindiff -o ./result.BinDiff -vv
 
 ..  code-block:: bash

doc/source/params.rst

@@ -25,16 +25,16 @@ Most of the distance functions that QBinDiff uses come from `Scipy <https://docs
 * seuclidean
 * sqeuclidean
 
-However, some distance are unique in QBinDiff, such as the jaccard-strong distance.
+However, some distance are unique in QBinDiff, such as the haussmann distance.
 This is a experimental new metric that combines the jaccard index and the canberra distance.
 
-Jaccard-strong
-~~~~~~~~~~~~~~
+Haussmann
+~~~~~~~~~
 
 Formally it is defined as:
 
 .. math::
-   d(u, v) = \sum_{i=0}^n\frac{f(u_i, v_i)}{ | \{ i  |  u_i \neq 0 \lor v_i \neq 0 \} | }
+   d(u, v) = \sum_{i=0}^n\frac{f(u_i, v_i)}{ \lvert \{ j | u_j \neq 0 \lor v_j \neq 0 \} \rvert }
 
 .. math::
    with\ u, v \in \mathbb{R}^n

src/qbindiff/passes/fast_metrics.pyx

@@ -124,7 +124,7 @@ def sparse_canberra(floating[::1] X_data, int[:] X_indices, int[:] X_indptr,
 
             D[px, py] = d
 
-def sparse_strong_jaccard(floating[::1] X_data, int[:] X_indices, int[:] X_indptr,
+def sparse_haussmann(floating[::1] X_data, int[:] X_indices, int[:] X_indptr,
                     floating[::1] Y_data, int[:] Y_indices, int[:] Y_indptr,
                     double[:, ::1] D, double[:] w):
     """Pairwise canberra distances for CSR matrices"""

src/qbindiff/passes/metrics.py

@@ -34,7 +34,7 @@
 import sklearn.metrics
 from scipy.spatial import distance
 from scipy.sparse import issparse, csr_matrix
-from qbindiff.passes.fast_metrics import sparse_canberra, sparse_strong_jaccard
+from qbindiff.passes.fast_metrics import sparse_canberra, sparse_haussmann
 from qbindiff.types import Distance
 
 
@@ -106,16 +106,16 @@ def canberra_distances(X, Y, w=None):
     ValueError("Cannot assign weights with non-sparse matrices")
 
 
-def jaccard_strong(X, Y, w=None):
+def haussmann(X, Y, w=None):
     r"""
-    Compute a variation of the jaccard distances between the vectors in X and Y using
-    the optional array of weights w.
+    Custom distance that takes inspiration from the jaccard index and the canberra distance.
+    If computes the distance between the vectors in X and Y using the optional array of weights w.
 
     The distance function between two vector ``u`` and ``v`` is the following:
 
     .. math::
 
-        \sum_{i}\frac{f(u_i, v_i)}{ | \{ i | u_i \neq 0 \lor v_i \neq 0 \} | }
+        \sum_{i}\frac{f(u_i, v_i)}{ | \{ j | u_j \neq 0 \lor v_j \neq 0 \} | }
 
     where the function ``f`` is defined like this:
 
@@ -127,7 +127,7 @@ def jaccard_strong(X, Y, w=None):
 
     .. math::
 
-        \sum_{i}\frac{w_i * f(u_i, v_i)}{ | \{ i | u_i \neq 0 \lor v_i \neq 0 \} | }
+        \sum_{i}\frac{w_i * f(u_i, v_i)}{ | \{ j | u_j \neq 0 \lor v_j \neq 0 \} | }
 
     :param X: array-like of shape (n_samples_X, n_features)
               An array where each row is a sample and each column is a feature.
@@ -139,7 +139,7 @@ def jaccard_strong(X, Y, w=None):
               Default is None, which gives each value a weight of 1.0
 
     :return D: ndarray of shape (n_samples_X, n_samples_Y)
-               D contains the pairwise strong jaccard distances.
+               D contains the pairwise haussmann distances.
 
     When X and/or Y are CSR sparse matrices and they are not already
     in canonical format, this function modifies them in-place to
@@ -159,19 +159,19 @@ def jaccard_strong(X, Y, w=None):
             w = _validate_weights(w)
             if w.size != X.shape[1]:
                 ValueError("Weights size mismatch")
-        sparse_strong_jaccard(X.data, X.indices, X.indptr, Y.data, Y.indices, Y.indptr, D, w)
+        sparse_haussmann(X.data, X.indices, X.indptr, Y.data, Y.indices, Y.indptr, D, w)
         return D
 
     if w is None:
-        return sparse_strong_jaccard(
+        return sparse_haussmann(
             X.data, X.indices, X.indptr, Y.data, Y.indices, Y.indptr, D, None
         )
     ValueError("Cannot assign weights with non-sparse matrices")
 
 
 CUSTOM_DISTANCES = {
     Distance.canberra: canberra_distances,
-    Distance.jaccard_strong: jaccard_strong,
+    Distance.haussmann: haussmann,
 }
 
 

src/qbindiff/types.py

@@ -207,4 +207,4 @@ class Distance(IntEnum):
     canberra = 0  # doc: canberra distance
     euclidean = 1  # doc: euclidean distance
     cosine = 2  # doc: cosine distance
-    jaccard_strong = 3  # doc: custom distance
+    haussmann = 3  # doc: haussmann distance