使用Rust ta库计算EMA时如何限制next迭代器的输入数据数量

我正在使用ta库计算指数移动平均线（Exponential Moving Average），目前我通过next迭代器传入每个close收盘价进行计算。我希望将EMA的输入值限制为201个（计算EMA200时，基于201个输入值和例如500个输入值得到的输出结果存在差异），请问是否有方法可以实现该需求？

原有代码 #

use ta::indicators::ExponentialMovingAverage as EMA;
use ta::Next;

pub fn main() {
    let mut reader = csv::Reader::from_path("datafile.csv").unwrap();

    let mut ema = EMA::new(200).unwrap();
    
    for record in reader.deserialize() {
        let (timestamp, open, high, low, close, volume): (String, f64, f64, f64, f64, f64) = record.unwrap();

        let ema_val = ema.next(close);

        println!("{} - ema: {}", timestamp, ema_val);
    }
}

数据源格式示例 #

timestamp,open,high,low,close,volume
2017-01-03,757.919983,758.760010,747.700012,753.669983,3521100
2017-01-04,758.390015,759.679993,754.200012,757.179993,2510500
2017-01-05,761.549988,782.400024,760.260010,780.450012,5830100
2017-01-06,782.359985,799.440002,778.479980,795.989990,5986200
2017-01-09,798.000000,801.770020,791.770020,796.919983,3440100
2017-01-10,796.599976,798.000000,789.539978,795.900024,2558400
2017-01-11,793.659973,799.500000,789.510010,799.020020,2992800
2017-01-12,800.309998,814.130005,799.500000,813.640015,4873900
2017-01-13,814.320007,821.650024,811.400024,817.140015,3791900
2017-01-17,815.700012,816.000000,803.440002,809.719971,3659400
2017-01-18,809.500000,811.729980,804.270020,807.479980,2354200
2017-01-19,810.000000,813.510010,807.320007,809.039978,2540800
2017-01-20,815.280029,816.020020,806.260010,808.330017,3376200
2017-01-23,806.799988,818.500000,805.080017,817.880005,2797500
2017-01-24,822.000000,823.989990,814.500000,822.440002,2971700
2017-01-25,825.789978,837.419983,825.289978,836.520020,3922600
2017-01-26,835.530029,843.840027,833.000000,839.150024,3586300
2017-01-27,839.000000,839.700012,829.440002,835.770020,2998700
2017-01-30,833.000000,833.500000,816.380005,830.380005,3747300
2017-01-31,823.750000,826.989990,819.559998,823.479980,3137200

差异是EMA的算法特性导致的：标准EMA计算会保留所有历史输入的加权影响，仅权重随时间衰减，因此输入500个和201个数据得到的结果自然存在区别。要实现仅用最近201个输入计算EMA，可以通过固定长度滑动窗口+每次重置EMA实例的方式实现，修改后代码如下：

use ta::indicators::ExponentialMovingAverage as EMA;
use ta::Next;
use std::collections::VecDeque;

pub fn main() {
    let mut reader = csv::Reader::from_path("datafile.csv").unwrap();
    // 初始化容量为201的双向队列作为滑动窗口，存储最近201个收盘价
    let mut close_window = VecDeque::with_capacity(201);
    const EMA_PERIOD: usize = 200;

    for record in reader.deserialize() {
        let (timestamp, _, _, _, close, _): (String, f64, f64, f64, f64, f64) = record.unwrap();
        // 新收盘价加入窗口，超出长度则弹出最早的历史数据
        close_window.push_back(close);
        if close_window.len() > 201 {
            close_window.pop_front();
        }

        let ema_val = if close_window.len() == 201 {
            // 窗口填满后，新建EMA实例重新计算当前窗口所有数据的EMA值
            let mut ema = EMA::new(EMA_PERIOD).unwrap();
            close_window.iter().map(|&price| ema.next(price)).last().unwrap()
        } else {
            // 窗口未填满时返回空值，也可根据需求调整为普通EMA计算逻辑
            f64::NAN
        };

        if !ema_val.is_nan() {
            println!("{} - ema: {}", timestamp, ema_val);
        }
    }
}

补充说明 #

• 窗口未满阶段的处理可以按需调整，不需要跳过的话可以直接用原有逻辑计算输出。
• 该方案时间复杂度为O(n*201)，普通行情数据集运行无压力，不需要额外优化。如果数据量极大，可以提前预计算EMA权重减少重复计算，日常使用无需考虑。

内容的提问来源于stack exchange，提问作者AlgoQ