Omnibox - Boost Frequency Scores Based on Number of Matching Pages

components/omnibox/browser/omnibox_field_trial.cc

Index: components/omnibox/browser/omnibox_field_trial.cc
diff --git a/components/omnibox/browser/omnibox_field_trial.cc b/components/omnibox/browser/omnibox_field_trial.cc
index cbcf6f6c652608ac671e6122705442e1a3457401..440df5a716a9337773c2d6e441ee7e22965fe2b2 100644
--- a/components/omnibox/browser/omnibox_field_trial.cc
+++ b/components/omnibox/browser/omnibox_field_trial.cc
@@ -4,6 +4,7 @@
 
 #include "components/omnibox/browser/omnibox_field_trial.h"
 
+#include <algorithm>
 #include <cmath>
 #include <string>
 
@@ -443,6 +444,29 @@ float OmniboxFieldTrial::HQPTypedValue() {
   return typed_value;
 }
 
+OmniboxFieldTrial::NumMatchesScores OmniboxFieldTrial::HQPNumMatchesScores() {
+  std::string str = variations::GetVariationParamValue(
+      kBundledExperimentFieldTrialName, kHQPNumMatchesScoresRule);
+  // The parameter is a comma-separated list of (number, value) pairs, e.g.
+  // "1:3,2:2.5,3:2,4:1.5".
+  // This is a best-effort conversion; we trust the hand-crafted parameters
+  // downloaded from the server to be perfect.  There's no need to handle
+  // errors smartly.
+  base::StringPairs kv_pairs;
+  if (!base::SplitStringIntoKeyValuePairs(str, ':', ',', &kv_pairs))
+    return NumMatchesScores{};
+  NumMatchesScores num_matches_scores(kv_pairs.size());
+  for (size_t i = 0; i < kv_pairs.size(); ++i) {
+    base::StringToSizeT(kv_pairs[i].first, &num_matches_scores[i].first);
+    // Don't allow duplicates.
+    DCHECK((i == 0) ||
+           (num_matches_scores[i].first != num_matches_scores[i - 1].first));

[Peter Kasting, 2016/12/06 05:19:30]

Wouldn't using ">" here obviate the need for the is_sorted() call below?

Or, instead, you could eliminate this DCHECK entirely and replace that with:

DCHECK(std::is_sorted(num_matches_scores.begin(), num_matches_scores.end(),
[](a, b) { return a.first < b.first; }));

Note that here a and b will need type declarations of your pair type.  It might
be easier to use a type alias in the header for this.

Doing this would also allow switching the loop to use range-based for.

[Mark P, 2016/12/08 00:21:31]

Of these two suggestions, I prefer the first one.  Did the > and removed the
is_sorted() call.

+    base::StringToDouble(kv_pairs[i].second, &num_matches_scores[i].second);
+  }
+  DCHECK(std::is_sorted(num_matches_scores.begin(), num_matches_scores.end()));
+  return num_matches_scores;
+}
+
 size_t OmniboxFieldTrial::HQPNumTitleWordsToAllow() {
   // The value of the rule is a string that encodes an integer (actually
   // size_t) containing the number of words.
@@ -548,6 +572,8 @@ const char OmniboxFieldTrial::kHQPFixFewVisitsBugRule[] = "HQPFixFewVisitsBug";
 const char OmniboxFieldTrial::kHQPFreqencyUsesSumRule[] = "HQPFreqencyUsesSum";
 const char OmniboxFieldTrial::kHQPMaxVisitsToScoreRule[] =
     "HQPMaxVisitsToScoreRule";
+const char OmniboxFieldTrial::kHQPNumMatchesScoresRule[] =
+    "HQPNumMatchesScores";
 const char OmniboxFieldTrial::kHQPNumTitleWordsRule[] = "HQPNumTitleWords";
 const char OmniboxFieldTrial::kHQPAlsoDoHUPLikeScoringRule[] =
     "HQPAlsoDoHUPLikeScoring";