Merge pull request #4 from dataiku/fix/remove-inner-products

Remove every inner product based similarity metrics
dataiku · Nov 25, 2020 · 2a6c8b8 · 2a6c8b8
2 parents 63f8037 + 2e42eb7
commit 2a6c8b8
Show file tree

Hide file tree

Showing 3 changed files with 2 additions and 12 deletions.
diff --git a/custom-recipes/similarity-search-index/recipe.json b/custom-recipes/similarity-search-index/recipe.json
@@ -122,10 +122,6 @@
                 {
                     "label": "Hamming",
                     "value": "hamming"
-                },
-                {
-                    "label": "Dot",
-                    "value": "dot"
                 }
             ],
             "defaultValue": "angular",
@@ -148,10 +144,6 @@
                     "label": "Exact Search for L2",
                     "value": "IndexFlatL2"
                 },
-                {
-                    "label": "Exact Search for Inner Product",
-                    "value": "IndexFlatIP"
-                },
                 {
                     "label": "Locality-Sensitive Hashing",
                     "value": "IndexLSH"

diff --git a/python-lib/dku_param_loading.py b/python-lib/dku_param_loading.py
@@ -90,7 +90,7 @@ def load_indexing_recipe_params() -> Dict:
     modeling_params["expert"] = bool(recipe_config.get("expert"))
     if modeling_params["algorithm"] == "annoy":
         modeling_params["annoy_metric"] = recipe_config.get("annoy_metric")
-        if modeling_params["annoy_metric"] not in {"angular", "euclidean", "manhattan", "hamming", "dot"}:
+        if modeling_params["annoy_metric"] not in {"angular", "euclidean", "manhattan", "hamming"}:
             raise PluginParamValidationError(f"Invalid Annoy distance metric: {modeling_params['annoy_metric']}")
         modeling_params["annoy_num_trees"] = recipe_config.get("annoy_num_trees")
         if not isinstance(modeling_params["annoy_num_trees"], int):
@@ -99,7 +99,7 @@ def load_indexing_recipe_params() -> Dict:
             raise PluginParamValidationError("Number of trees must be above 1")
     elif modeling_params["algorithm"] == "faiss":
         modeling_params["faiss_index_type"] = recipe_config.get("faiss_index_type")
-        if modeling_params["faiss_index_type"] not in {"IndexFlatL2", "IndexFlatIP", "IndexLSH"}:
+        if modeling_params["faiss_index_type"] not in {"IndexFlatL2", "IndexLSH"}:
             raise PluginParamValidationError(f"Invalid FAISS index type: {modeling_params['faiss_index_type']}")
         modeling_params["faiss_lsh_num_bits"] = recipe_config.get("faiss_lsh_num_bits")
         if not isinstance(modeling_params["faiss_lsh_num_bits"], int):

diff --git a/python-lib/nearest_neighbor/faiss.py b/python-lib/nearest_neighbor/faiss.py
@@ -21,8 +21,6 @@ def __init__(self, num_dimensions: int, **kwargs):
         self.faiss_lsh_num_bits = int(kwargs.get("faiss_lsh_num_bits", 4))
         if self.faiss_index_type == "IndexFlatL2":
             self.index = faiss.IndexFlatL2(self.num_dimensions)
-        elif self.faiss_index_type == "IndexFlatIP":
-            self.index = faiss.IndexFlatIP(self.num_dimensions)
         elif self.faiss_index_type == "IndexLSH":
             self.index = faiss.IndexLSH(self.num_dimensions, self.faiss_lsh_num_bits)
         else: