Struct ArrayOfDoublesSketch 

pub struct ArrayOfDoublesSketch { /* private fields */ }

A mutable, update-only ArrayOfDoubles Tuple sketch: estimates the number of distinct keys added via update_*, and carries a fixed-width array of f64 values per retained key, summed on collision. Build one with ArrayOfDoublesSketchBuilder.

Call Self::compact to produce an immutable, serializable super::CompactArrayOfDoublesSketch snapshot for storage, transmission, or use as input to a set operation.

Implementations

impl ArrayOfDoublesSketch

pub fn update_u64( &mut self, key: u64, values: &[f64], ) -> Result<(), SketchError>

Adds a u64 key with its associated values. Returns SketchError::InvalidConfig unless values.len() == self.get_num_values().

Examples found in repository

examples/bench_tuple_update.rs (line 68)

fn bench_distinct(items: u64) {
    let mut sketch = ArrayOfDoublesSketchBuilder::new()
        .lg_k(LG_K)
        .num_values(NUM_VALUES)
        .build()
        .expect("builder rejected fixed valid parameters");

    let start = Instant::now();
    for key in 0..items {
        sketch
            .update_u64(key, &VALUES)
            .expect("update rejected a correctly-sized value slice");
    }
    let elapsed = start.elapsed();

    // Reading the estimate keeps the loop above from being optimised out.
    report("distinct", items, elapsed, sketch.get_estimate());
}

fn bench_hot(items: u64) {
    let mut sketch = ArrayOfDoublesSketchBuilder::new()
        .lg_k(LG_K)
        .num_values(NUM_VALUES)
        .build()
        .expect("builder rejected fixed valid parameters");

    let start = Instant::now();
    for i in 0..items {
        sketch
            .update_u64(i % HOT_KEY_SPACE, &VALUES)
            .expect("update rejected a correctly-sized value slice");
    }
    let elapsed = start.elapsed();

    report("hot", items, elapsed, sketch.get_estimate());
}

examples/tuple.rs (line 22)

fn main() {
    // Two sketches of user IDs, each carrying [sessions, revenue] per user.
    let mut day1 = ArrayOfDoublesSketchBuilder::new()
        .lg_k(12)
        .num_values(2)
        .build()
        .unwrap();
    for id in 0..10_000u64 {
        day1.update_u64(id, &[1.0, 2.50]).unwrap();
    }

    let mut day2 = ArrayOfDoublesSketchBuilder::new()
        .lg_k(12)
        .num_values(2)
        .build()
        .unwrap();
    for id in 5_000..15_000u64 {
        day2.update_u64(id, &[1.0, 4.00]).unwrap();
    }

    println!("Day 1 unique users (estimate): {:.0}", day1.get_estimate());
    println!("Day 2 unique users (estimate): {:.0}", day2.get_estimate());
    println!("Values per entry: {}", day1.get_num_values());

    // Union: unique users across both days, with per-user values summed for
    // anyone who appeared on both.
    let mut union = ArrayOfDoublesUnionBuilder::new()
        .lg_k(12)
        .num_values(2)
        .build()
        .unwrap();
    union.update(&day1).unwrap();
    union.update(&day2).unwrap();
    let combined = union.get_result(true);
    println!(
        "Total unique users (union estimate): {:.0}",
        combined.get_estimate()
    );

    // Per-entry access is what distinguishes Tuple sketches from HLL/Theta/CPC:
    // scale the retained sample's revenue back up by 1/theta to estimate the
    // full population total.
    let retained_revenue: f64 = combined.entries().map(|(_, values)| values[1]).sum();
    println!(
        "Estimated total revenue: {:.2} (from {} retained entries, theta = {:.4})",
        retained_revenue / combined.get_theta(),
        combined.get_num_retained(),
        combined.get_theta()
    );

    // Intersection: users who came back on day 2.
    let mut intersection = ArrayOfDoublesIntersection::new(2).unwrap();
    intersection.update(&day1).unwrap();
    intersection.update(&day2).unwrap();
    match intersection.get_result(true) {
        Ok(returning) => println!(
            "Returning users (intersection estimate): {:.0}",
            returning.get_estimate()
        ),
        Err(e) => println!("No intersection result: {e}"),
    }

    // A-not-b: users who only came on day 1.
    let a_not_b = ArrayOfDoublesAnotB::new();
    let day1_only = a_not_b.compute(&day1, &day2, true).unwrap();
    println!(
        "Day-1-only users (a-not-b estimate): {:.0}",
        day1_only.get_estimate()
    );

    // Jaccard similarity of the two days' audiences.
    let similarity = array_of_doubles_jaccard_similarity(&day1, &day2).unwrap();
    println!(
        "Jaccard similarity: {:.3} (range [{:.3}, {:.3}])",
        similarity.estimate, similarity.lower_bound, similarity.upper_bound
    );

    // Serialize a compact sketch for storage/transmission, then restore it.
    let compact = day1.compact(true);
    let bytes = compact.serialize();
    println!("Serialized day-1 sketch: {} bytes", bytes.len());
    let restored = CompactArrayOfDoublesSketch::deserialize(&bytes).unwrap();
    println!(
        "Restored estimate: {:.0} ({} values per entry)",
        restored.get_estimate(),
        restored.get_num_values()
    );
}

pub fn update_i64( &mut self, key: i64, values: &[f64], ) -> Result<(), SketchError>

Adds an i64 key with its associated values. See Self::update_u64.

pub fn update_u32( &mut self, key: u32, values: &[f64], ) -> Result<(), SketchError>

Adds a u32 key with its associated values. See Self::update_u64.

pub fn update_i32( &mut self, key: i32, values: &[f64], ) -> Result<(), SketchError>

Adds an i32 key with its associated values. See Self::update_u64.

pub fn update_u16( &mut self, key: u16, values: &[f64], ) -> Result<(), SketchError>

Adds a u16 key with its associated values. See Self::update_u64.

pub fn update_i16( &mut self, key: i16, values: &[f64], ) -> Result<(), SketchError>

Adds an i16 key with its associated values. See Self::update_u64.

pub fn update_u8(&mut self, key: u8, values: &[f64]) -> Result<(), SketchError>

Adds a u8 key with its associated values. See Self::update_u64.

pub fn update_i8(&mut self, key: i8, values: &[f64]) -> Result<(), SketchError>

Adds an i8 key with its associated values. See Self::update_u64.

pub fn update_f64( &mut self, key: f64, values: &[f64], ) -> Result<(), SketchError>

Adds an f64 key with its associated values. See Self::update_u64.

pub fn update_str( &mut self, key: &str, values: &[f64], ) -> Result<(), SketchError>

Adds a string key with its associated values. See Self::update_u64.

pub fn update_bytes( &mut self, key: &[u8], values: &[f64], ) -> Result<(), SketchError>

Adds an arbitrary byte-slice key with its associated values. See Self::update_u64.

pub fn trim(&mut self)

Removes retained entries in excess of the nominal size k, lowering the theta threshold to do so.

Note that this does shift Self::get_estimate — trimming lowers theta, and the estimate is derived from the retained count and theta together. Upstream only guarantees the excess entries are dropped.

pub fn reset(&mut self)

Resets this sketch to its initial, empty state. num_values is preserved.

pub fn get_estimate(&self) -> f64

Returns the current estimate of the number of distinct keys added.

Examples found in repository

examples/bench_tuple_update.rs (line 74)

fn bench_distinct(items: u64) {
    let mut sketch = ArrayOfDoublesSketchBuilder::new()
        .lg_k(LG_K)
        .num_values(NUM_VALUES)
        .build()
        .expect("builder rejected fixed valid parameters");

    let start = Instant::now();
    for key in 0..items {
        sketch
            .update_u64(key, &VALUES)
            .expect("update rejected a correctly-sized value slice");
    }
    let elapsed = start.elapsed();

    // Reading the estimate keeps the loop above from being optimised out.
    report("distinct", items, elapsed, sketch.get_estimate());
}

fn bench_hot(items: u64) {
    let mut sketch = ArrayOfDoublesSketchBuilder::new()
        .lg_k(LG_K)
        .num_values(NUM_VALUES)
        .build()
        .expect("builder rejected fixed valid parameters");

    let start = Instant::now();
    for i in 0..items {
        sketch
            .update_u64(i % HOT_KEY_SPACE, &VALUES)
            .expect("update rejected a correctly-sized value slice");
    }
    let elapsed = start.elapsed();

    report("hot", items, elapsed, sketch.get_estimate());
}

examples/tuple.rs (line 34)

fn main() {
    // Two sketches of user IDs, each carrying [sessions, revenue] per user.
    let mut day1 = ArrayOfDoublesSketchBuilder::new()
        .lg_k(12)
        .num_values(2)
        .build()
        .unwrap();
    for id in 0..10_000u64 {
        day1.update_u64(id, &[1.0, 2.50]).unwrap();
    }

    let mut day2 = ArrayOfDoublesSketchBuilder::new()
        .lg_k(12)
        .num_values(2)
        .build()
        .unwrap();
    for id in 5_000..15_000u64 {
        day2.update_u64(id, &[1.0, 4.00]).unwrap();
    }

    println!("Day 1 unique users (estimate): {:.0}", day1.get_estimate());
    println!("Day 2 unique users (estimate): {:.0}", day2.get_estimate());
    println!("Values per entry: {}", day1.get_num_values());

    // Union: unique users across both days, with per-user values summed for
    // anyone who appeared on both.
    let mut union = ArrayOfDoublesUnionBuilder::new()
        .lg_k(12)
        .num_values(2)
        .build()
        .unwrap();
    union.update(&day1).unwrap();
    union.update(&day2).unwrap();
    let combined = union.get_result(true);
    println!(
        "Total unique users (union estimate): {:.0}",
        combined.get_estimate()
    );

    // Per-entry access is what distinguishes Tuple sketches from HLL/Theta/CPC:
    // scale the retained sample's revenue back up by 1/theta to estimate the
    // full population total.
    let retained_revenue: f64 = combined.entries().map(|(_, values)| values[1]).sum();
    println!(
        "Estimated total revenue: {:.2} (from {} retained entries, theta = {:.4})",
        retained_revenue / combined.get_theta(),
        combined.get_num_retained(),
        combined.get_theta()
    );

    // Intersection: users who came back on day 2.
    let mut intersection = ArrayOfDoublesIntersection::new(2).unwrap();
    intersection.update(&day1).unwrap();
    intersection.update(&day2).unwrap();
    match intersection.get_result(true) {
        Ok(returning) => println!(
            "Returning users (intersection estimate): {:.0}",
            returning.get_estimate()
        ),
        Err(e) => println!("No intersection result: {e}"),
    }

    // A-not-b: users who only came on day 1.
    let a_not_b = ArrayOfDoublesAnotB::new();
    let day1_only = a_not_b.compute(&day1, &day2, true).unwrap();
    println!(
        "Day-1-only users (a-not-b estimate): {:.0}",
        day1_only.get_estimate()
    );

    // Jaccard similarity of the two days' audiences.
    let similarity = array_of_doubles_jaccard_similarity(&day1, &day2).unwrap();
    println!(
        "Jaccard similarity: {:.3} (range [{:.3}, {:.3}])",
        similarity.estimate, similarity.lower_bound, similarity.upper_bound
    );

    // Serialize a compact sketch for storage/transmission, then restore it.
    let compact = day1.compact(true);
    let bytes = compact.serialize();
    println!("Serialized day-1 sketch: {} bytes", bytes.len());
    let restored = CompactArrayOfDoublesSketch::deserialize(&bytes).unwrap();
    println!(
        "Restored estimate: {:.0} ({} values per entry)",
        restored.get_estimate(),
        restored.get_num_values()
    );
}

pub fn get_lower_bound(&self, num_std_dev: u8) -> Result<f64, SketchError>

Returns the lower bound of the confidence interval around Self::get_estimate, for the given number of standard deviations (1, 2, or 3, corresponding to roughly 67%, 95%, and 99% confidence). Returns SketchError::InvalidConfig for any other value.

pub fn get_upper_bound(&self, num_std_dev: u8) -> Result<f64, SketchError>

Returns the upper bound of the confidence interval around Self::get_estimate. See Self::get_lower_bound for the meaning of num_std_dev.

pub fn is_empty(&self) -> bool

Returns true if no keys have been added to this sketch.

pub fn is_estimation_mode(&self) -> bool

Returns true if this sketch’s theta threshold is below 1.0 (i.e. it has begun sampling and Self::get_estimate is a statistical estimate rather than an exact count).

pub fn is_ordered(&self) -> bool

Returns true if this sketch’s retained entries are sorted by hash value.

pub fn get_theta(&self) -> f64

Returns the current theta threshold (1.0 until sampling begins).

pub fn get_num_retained(&self) -> u32

Returns the number of entries currently retained by this sketch.

pub fn get_num_values(&self) -> u8

Returns the fixed number of f64 values each retained entry carries, as configured at build time.

Examples found in repository

examples/tuple.rs (line 36)

fn main() {
    // Two sketches of user IDs, each carrying [sessions, revenue] per user.
    let mut day1 = ArrayOfDoublesSketchBuilder::new()
        .lg_k(12)
        .num_values(2)
        .build()
        .unwrap();
    for id in 0..10_000u64 {
        day1.update_u64(id, &[1.0, 2.50]).unwrap();
    }

    let mut day2 = ArrayOfDoublesSketchBuilder::new()
        .lg_k(12)
        .num_values(2)
        .build()
        .unwrap();
    for id in 5_000..15_000u64 {
        day2.update_u64(id, &[1.0, 4.00]).unwrap();
    }

    println!("Day 1 unique users (estimate): {:.0}", day1.get_estimate());
    println!("Day 2 unique users (estimate): {:.0}", day2.get_estimate());
    println!("Values per entry: {}", day1.get_num_values());

    // Union: unique users across both days, with per-user values summed for
    // anyone who appeared on both.
    let mut union = ArrayOfDoublesUnionBuilder::new()
        .lg_k(12)
        .num_values(2)
        .build()
        .unwrap();
    union.update(&day1).unwrap();
    union.update(&day2).unwrap();
    let combined = union.get_result(true);
    println!(
        "Total unique users (union estimate): {:.0}",
        combined.get_estimate()
    );

    // Per-entry access is what distinguishes Tuple sketches from HLL/Theta/CPC:
    // scale the retained sample's revenue back up by 1/theta to estimate the
    // full population total.
    let retained_revenue: f64 = combined.entries().map(|(_, values)| values[1]).sum();
    println!(
        "Estimated total revenue: {:.2} (from {} retained entries, theta = {:.4})",
        retained_revenue / combined.get_theta(),
        combined.get_num_retained(),
        combined.get_theta()
    );

    // Intersection: users who came back on day 2.
    let mut intersection = ArrayOfDoublesIntersection::new(2).unwrap();
    intersection.update(&day1).unwrap();
    intersection.update(&day2).unwrap();
    match intersection.get_result(true) {
        Ok(returning) => println!(
            "Returning users (intersection estimate): {:.0}",
            returning.get_estimate()
        ),
        Err(e) => println!("No intersection result: {e}"),
    }

    // A-not-b: users who only came on day 1.
    let a_not_b = ArrayOfDoublesAnotB::new();
    let day1_only = a_not_b.compute(&day1, &day2, true).unwrap();
    println!(
        "Day-1-only users (a-not-b estimate): {:.0}",
        day1_only.get_estimate()
    );

    // Jaccard similarity of the two days' audiences.
    let similarity = array_of_doubles_jaccard_similarity(&day1, &day2).unwrap();
    println!(
        "Jaccard similarity: {:.3} (range [{:.3}, {:.3}])",
        similarity.estimate, similarity.lower_bound, similarity.upper_bound
    );

    // Serialize a compact sketch for storage/transmission, then restore it.
    let compact = day1.compact(true);
    let bytes = compact.serialize();
    println!("Serialized day-1 sketch: {} bytes", bytes.len());
    let restored = CompactArrayOfDoublesSketch::deserialize(&bytes).unwrap();
    println!(
        "Restored estimate: {:.0} ({} values per entry)",
        restored.get_estimate(),
        restored.get_num_values()
    );
}

pub fn entries(&self) -> impl Iterator<Item = (u64, Vec<f64>)>

Iterates the retained entries as (hash, values) pairs, where values.len() == self.get_num_values().

The entries are copied out of C++ in two FFI calls up front (cxx cannot hand back a live C++ iterator), so each item owns its Vec rather than borrowing from the sketch. Iteration order is unspecified for an update sketch; compact it with ordered = true for hash-ordered iteration.

pub fn compact(&self, ordered: bool) -> CompactArrayOfDoublesSketch

Produces an immutable, serializable super::CompactArrayOfDoublesSketch snapshot of this sketch’s current state. If ordered is true, the snapshot’s entries are sorted by hash value.

Examples found in repository

examples/tuple.rs (line 92)

fn main() {
    // Two sketches of user IDs, each carrying [sessions, revenue] per user.
    let mut day1 = ArrayOfDoublesSketchBuilder::new()
        .lg_k(12)
        .num_values(2)
        .build()
        .unwrap();
    for id in 0..10_000u64 {
        day1.update_u64(id, &[1.0, 2.50]).unwrap();
    }

    let mut day2 = ArrayOfDoublesSketchBuilder::new()
        .lg_k(12)
        .num_values(2)
        .build()
        .unwrap();
    for id in 5_000..15_000u64 {
        day2.update_u64(id, &[1.0, 4.00]).unwrap();
    }

    println!("Day 1 unique users (estimate): {:.0}", day1.get_estimate());
    println!("Day 2 unique users (estimate): {:.0}", day2.get_estimate());
    println!("Values per entry: {}", day1.get_num_values());

    // Union: unique users across both days, with per-user values summed for
    // anyone who appeared on both.
    let mut union = ArrayOfDoublesUnionBuilder::new()
        .lg_k(12)
        .num_values(2)
        .build()
        .unwrap();
    union.update(&day1).unwrap();
    union.update(&day2).unwrap();
    let combined = union.get_result(true);
    println!(
        "Total unique users (union estimate): {:.0}",
        combined.get_estimate()
    );

    // Per-entry access is what distinguishes Tuple sketches from HLL/Theta/CPC:
    // scale the retained sample's revenue back up by 1/theta to estimate the
    // full population total.
    let retained_revenue: f64 = combined.entries().map(|(_, values)| values[1]).sum();
    println!(
        "Estimated total revenue: {:.2} (from {} retained entries, theta = {:.4})",
        retained_revenue / combined.get_theta(),
        combined.get_num_retained(),
        combined.get_theta()
    );

    // Intersection: users who came back on day 2.
    let mut intersection = ArrayOfDoublesIntersection::new(2).unwrap();
    intersection.update(&day1).unwrap();
    intersection.update(&day2).unwrap();
    match intersection.get_result(true) {
        Ok(returning) => println!(
            "Returning users (intersection estimate): {:.0}",
            returning.get_estimate()
        ),
        Err(e) => println!("No intersection result: {e}"),
    }

    // A-not-b: users who only came on day 1.
    let a_not_b = ArrayOfDoublesAnotB::new();
    let day1_only = a_not_b.compute(&day1, &day2, true).unwrap();
    println!(
        "Day-1-only users (a-not-b estimate): {:.0}",
        day1_only.get_estimate()
    );

    // Jaccard similarity of the two days' audiences.
    let similarity = array_of_doubles_jaccard_similarity(&day1, &day2).unwrap();
    println!(
        "Jaccard similarity: {:.3} (range [{:.3}, {:.3}])",
        similarity.estimate, similarity.lower_bound, similarity.upper_bound
    );

    // Serialize a compact sketch for storage/transmission, then restore it.
    let compact = day1.compact(true);
    let bytes = compact.serialize();
    println!("Serialized day-1 sketch: {} bytes", bytes.len());
    let restored = CompactArrayOfDoublesSketch::deserialize(&bytes).unwrap();
    println!(
        "Restored estimate: {:.0} ({} values per entry)",
        restored.get_estimate(),
        restored.get_num_values()
    );
}

Trait Implementations

impl ArrayOfDoublesInput for ArrayOfDoublesSketch

fn get_num_values(&self) -> u8

The fixed number of f64 values each of this sketch’s retained entries carries. Set operations require all operands to agree.

impl Send for ArrayOfDoublesSketch