Quantize in range [1, 2^b)

Due to quantization, some scores can be 0, but our frequency encoding
(which is used for scores) assumes positive values. To fix it, we
quantize into a range starting at 1 instead.

Fixes: #572

include/pisa/linear_quantizer.hpp

@@ -1,6 +1,5 @@
 #pragma once
 
-#include <cmath>
 #include <cstdint>
 
 #include <fmt/format.h>
@@ -9,21 +8,12 @@
 namespace pisa {
 
 struct LinearQuantizer {
-    explicit LinearQuantizer(float max, std::uint8_t bits)
-        : m_max(max), m_scale(static_cast<float>((1U << bits) - 1U) / max) {
-        if (bits > 32 or bits == 0) {
-            throw std::runtime_error(fmt::format(
-                "Linear quantizer must take a number of bits between 1 and 32 but {} passed", bits
-            ));
-        }
-    }
-
-    [[nodiscard]] auto operator()(float value) const -> std::uint32_t {
-        Expects(value <= m_max);
-        return std::ceil(value * m_scale);
-    }
+    LinearQuantizer(float max, std::uint8_t bits);
+    [[nodiscard]] auto operator()(float value) const -> std::uint32_t;
+    [[nodiscard]] auto range() const noexcept -> std::uint32_t;
 
   private:
+    std::uint32_t m_range;
     float m_max;
     float m_scale;
 };

include/pisa/scorer/quantized.hpp

@@ -39,7 +39,10 @@ class QuantizingScorer {
             [this, scorer = m_scorer->term_scorer(term_id)](std::uint32_t doc, std::uint32_t freq) {
                 auto score = scorer(doc, freq);
                 assert(score >= 0.0);
-                return this->m_quantizer(score);
+                auto quantized = this->m_quantizer(score);
+                assert(quantized >= 0);
+                assert(quantized <= this->m_quantizer.range());
+                return quantized;
             };
     }
 };

src/linear_quantizer.cpp

@@ -0,0 +1,25 @@
+#include <cmath>
+
+#include "linear_quantizer.hpp"
+
+namespace pisa {
+
+LinearQuantizer::LinearQuantizer(float max, std::uint8_t bits)
+    : m_range((1U << bits) - 1U), m_max(max), m_scale(static_cast<float>(m_range - 1) / max) {
+    if (bits > 32 or bits < 2) {
+        throw std::runtime_error(fmt::format(
+            "Linear quantizer must take a number of bits between 2 and 32 but {} passed", bits
+        ));
+    }
+}
+
+auto LinearQuantizer::operator()(float value) const -> std::uint32_t {
+    Expects(0 <= value && value <= m_max);
+    return std::ceil(value * m_scale) + 1;
+}
+
+auto LinearQuantizer::range() const noexcept -> std::uint32_t {
+    return m_range;
+}
+
+}  // namespace pisa

test/test_compress.cpp

@@ -9,7 +9,7 @@
 #include "type_safe.hpp"
 #include "wand_utils.hpp"
 
-TEST_CASE("Compress block index", "[index][compress]") {
+TEST_CASE("Compress index", "[index][compress]") {
     std::string encoding = GENERATE(
         "ef",
         "single",
@@ -38,10 +38,10 @@ TEST_CASE("Compress block index", "[index][compress]") {
     );
 }
 
-TEST_CASE("Compress quantized block index", "[index][compress]") {
+TEST_CASE("Compress quantized index", "[index][compress]") {
     auto input = PISA_SOURCE_DIR "/test/test_data/test_collection";
 
-    std::string scorer = GENERATE("bm25");
+    std::string scorer = GENERATE("bm25", "qld");
     CAPTURE(scorer);
     auto scorer_params = ScorerParams(scorer);
 

test/test_linear_quantizer.cpp

@@ -8,7 +8,7 @@
 TEST_CASE("LinearQuantizer", "[scoring][unit]") {
     SECTION("construct") {
         WHEN("number of bits is 0 or 33") {
-            std::uint8_t bits = GENERATE(0, 33);
+            std::uint8_t bits = GENERATE(0, 1, 33);
             THEN("constructor fails") {
                 REQUIRE_THROWS(pisa::LinearQuantizer(10.0, bits));
             }
@@ -18,7 +18,13 @@ TEST_CASE("LinearQuantizer", "[scoring][unit]") {
         std::uint8_t bits = GENERATE(3, 8, 12, 16, 19, 32);
         float max = GENERATE(1.0, 100.0, std::numeric_limits<float>::max());
         pisa::LinearQuantizer quantizer(max, bits);
-        REQUIRE(quantizer(0) == 0);
+        REQUIRE(quantizer(0) == 1);
         REQUIRE(quantizer(max) == (1 << bits) - 1);
     }
+    SECTION("max is 0") {
+        std::uint8_t bits = GENERATE(3, 8, 12, 16, 19, 32);
+        float max = 0.0;
+        pisa::LinearQuantizer quantizer(max, bits);
+        REQUIRE(quantizer(0) == 1);
+    }
 }