introduce ColumnCombValueStats

optd-cost-model/Cargo.toml

@@ -11,4 +11,4 @@ arrow-schema = "53.2.0"
 datafusion-expr = "32.0.0"
 ordered-float = "4.0"
 chrono = "0.4"
-
+serde_json = "1.0"

optd-cost-model/src/cost/mod.rs

@@ -4,3 +4,4 @@ pub mod agg;
 pub mod filter;
 pub mod join;
 pub mod limit;
+pub mod stats;

optd-cost-model/src/cost/stats.rs

@@ -0,0 +1,43 @@
+use serde::{Deserialize, Serialize};
+
+use crate::common::values::Value;
+
+pub type ColumnCombValue = Vec<Option<Value>>;
+
+/// Ideally, MostCommonValues would have trait bounds for Serialize and Deserialize. However, I have
+/// not figured out how to both have Deserialize as a trait bound and utilize the Deserialize
+/// macro, because the Deserialize trait involves lifetimes.
+pub trait MostCommonValues: 'static + Send + Sync {
+    // it is true that we could just expose freq_over_pred() and use that for freq() and
+    // total_freq() however, freq() and total_freq() each have potential optimizations (freq()
+    // is O(1) instead of     O(n) and total_freq() can be cached)
+    // additionally, it makes sense to return an Option<f64> for freq() instead of just 0 if value
+    // doesn't exist thus, I expose three different functions
+    fn freq(&self, value: &ColumnCombValue) -> Option<f64>;
+    fn total_freq(&self) -> f64;
+    fn freq_over_pred(&self, pred: Box<dyn Fn(&ColumnCombValue) -> bool>) -> f64;
+
+    // returns the # of entries (i.e. value + freq) in the most common values structure
+    fn cnt(&self) -> usize;
+}
+
+/// A more general interface meant to perform the task of a histogram.
+///
+/// This more general interface is still compatible with histograms but allows
+/// more powerful statistics like TDigest.
+///
+/// Ideally, Distribution would have trait bounds for Serialize and Deserialize.
+/// However, I have not figured out how to both have Deserialize as a trait bound
+/// and utilize the Deserialize macro, because the Deserialize trait involves lifetimes.
+pub trait Distribution: 'static + Send + Sync {
+    // Give the probability of a random value sampled from the distribution being <= `value`
+    fn cdf(&self, value: &Value) -> f64;
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+pub struct ColumnCombValueStats<M: MostCommonValues, D: Distribution> {
+    pub mcvs: M,          // Does NOT contain full nulls.
+    pub distr: Option<D>, // Does NOT contain mcvs; optional.
+    pub ndistinct: u64,   // Does NOT contain full nulls.
+    pub null_frac: f64,   // % of full nulls.
+}