huaxingao commented on code in PR #297: URL: https://github.com/apache/datafusion-comet/pull/297#discussion_r1578498812
########## core/src/execution/datafusion/expressions/variance.rs: ########## @@ -0,0 +1,259 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +//! Defines physical expressions that can evaluated at runtime during query execution + +use std::{any::Any, sync::Arc}; + +use crate::execution::datafusion::expressions::{stats::StatsType, utils::down_cast_any_ref}; +use arrow::{ + array::{ArrayRef, Float64Array}, + compute::cast, + datatypes::{DataType, Field}, +}; +use datafusion::logical_expr::Accumulator; +use datafusion_common::{downcast_value, DataFusionError, Result, ScalarValue}; +use datafusion_physical_expr::{expressions::format_state_name, AggregateExpr, PhysicalExpr}; + +/// VAR_SAMP and VAR_POP aggregate expression +/// The implementation mostly is the same as the DataFusion's implementation. The reason +/// we have our own implementation is that DataFusion has UInt64 for state_field `count`, +/// while Spark has Double for count. Also we have added `null_on_divide_by_zero` +/// to be consistent with Spark's implementation. +#[derive(Debug)] +pub struct Variance { + name: String, + expr: Arc<dyn PhysicalExpr>, + stats_type: StatsType, + null_on_divide_by_zero: bool, +} + +impl Variance { + /// Create a new VARIANCE aggregate function + pub fn new( + expr: Arc<dyn PhysicalExpr>, + name: impl Into<String>, + data_type: DataType, + stats_type: StatsType, + null_on_divide_by_zero: bool, + ) -> Self { + // the result of variance just support FLOAT64 data type. + assert!(matches!(data_type, DataType::Float64)); + Self { + name: name.into(), + expr, + stats_type, + null_on_divide_by_zero, + } + } +} + +impl AggregateExpr for Variance { + /// Return a reference to Any that can be used for downcasting + fn as_any(&self) -> &dyn Any { + self + } + + fn field(&self) -> Result<Field> { + Ok(Field::new(&self.name, DataType::Float64, true)) + } + + fn create_accumulator(&self) -> Result<Box<dyn Accumulator>> { + Ok(Box::new(VarianceAccumulator::try_new( + self.stats_type, + self.null_on_divide_by_zero, + )?)) + } + + fn create_sliding_accumulator(&self) -> Result<Box<dyn Accumulator>> { + Ok(Box::new(VarianceAccumulator::try_new( + self.stats_type, + self.null_on_divide_by_zero, + )?)) + } + + fn state_fields(&self) -> Result<Vec<Field>> { + Ok(vec![ + Field::new( + format_state_name(&self.name, "count"), + DataType::Float64, + true, + ), + Field::new( + format_state_name(&self.name, "mean"), + DataType::Float64, + true, + ), + Field::new(format_state_name(&self.name, "m2"), DataType::Float64, true), + ]) + } + + fn expressions(&self) -> Vec<Arc<dyn PhysicalExpr>> { + vec![self.expr.clone()] + } + + fn name(&self) -> &str { + &self.name + } +} + +impl PartialEq<dyn Any> for Variance { + fn eq(&self, other: &dyn Any) -> bool { + down_cast_any_ref(other) + .downcast_ref::<Self>() + .map(|x| { + self.name == x.name && self.expr.eq(&x.expr) && self.stats_type == x.stats_type + }) + .unwrap_or(false) + } +} + +/// An accumulator to compute variance +#[derive(Debug)] +pub struct VarianceAccumulator { + m2: f64, + mean: f64, + count: f64, + stats_type: StatsType, + null_on_divide_by_zero: bool, +} + +impl VarianceAccumulator { + /// Creates a new `VarianceAccumulator` + pub fn try_new(s_type: StatsType, null_on_divide_by_zero: bool) -> Result<Self> { + Ok(Self { + m2: 0_f64, + mean: 0_f64, + count: 0_f64, + stats_type: s_type, + null_on_divide_by_zero, + }) + } + + pub fn get_count(&self) -> f64 { + self.count + } + + pub fn get_mean(&self) -> f64 { + self.mean + } + + pub fn get_m2(&self) -> f64 { + self.m2 + } +} + +impl Accumulator for VarianceAccumulator { + fn state(&mut self) -> Result<Vec<ScalarValue>> { + Ok(vec![ + ScalarValue::from(self.count), + ScalarValue::from(self.mean), + ScalarValue::from(self.m2), + ]) + } + + fn update_batch(&mut self, values: &[ArrayRef]) -> Result<()> { + let values = &cast(&values[0], &DataType::Float64)?; Review Comment: Removed. Thanks! I have the same casting in covariance. There are a few problems I need to fix in covariance 1. remove the unnecessary cast 2. add `null_on_divide_by_zero` 3. combine `CovSample` and `CovPopulation` in `expr.proto` I will have a PR to fix these problems. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
