MyTT Indicator Library

Use the EMA function (exponential moving average) in the mytt module on a vector:

//If mytt is not specified in the preloadModules configuration, the module must be loaded manually once in each new session.
use mytt

close = 7.2 6.97 7.08 6.74 6.49 5.9 6.26 5.9 5.35 5.63
x = EMA(close, 5)

It is often necessary to perform calculations separately for each group in a data table. The following example first creates a table that contains two stocks:

use mytt

close = 7.2 6.97 7.08 6.74 6.49 5.9 6.26 5.9 5.35 5.63 3.81 3.935 4.04 3.74 3.7 3.33 3.64 3.31 2.69 2.72
date = (2020.03.02 + 0..4 join 7..11).take(20)
symbol = take(`F,10) join take(`GPRO,10)
t = table(symbol, date, close)

Use the EMA function in the mytt module to calculate values for each stock:

update t set EMA = EMA(close, 5) context by symbol

Some functions return results with multiple columns, such as the BIAS function (bias ratio indicator).

Example:

use mytt

close = 7.2 6.97 7.08 6.74 6.49 5.9 6.26 5.9 5.35 5.63
bias1, bias2, bias3 = BIAS(close, L1 = 2, L2 = 4, L3 = 6)

Example of use in an SQL statement:

use mytt

close = 7.2 6.97 7.08 6.74 6.49 5.9 6.26 5.9 5.35 5.63 3.81 3.935 4.04 3.74 3.7 3.33 3.64 3.31 2.69 2.72
date = (2020.03.02 + 0..4 join 7..11).take(20)
symbol = take(`F,10) join take(`GPRO,10)
t = table(symbol, date, close) 
select *, BIAS(close, L1 = 2, L2 = 4, L3 = 6) as `bias1`bias2`bias3 from t context by symbol

/* output:
symbol date close bias1 bias2 bias3
------ ---------- ----- -------- -------- --------
F 2020.03.02 7.2
F 2020.03.03 6.97 -1.623
F 2020.03.04 7.08 0.783 
F 2020.03.05 6.74 -2.46 -3.68
F 2020.03.06 6.49 -1.89 -4.839 
F 2020.03.09 5.9 -4.762 -9.958 -12.333
F 2020.03.10 6.26 2.961 -1.378 -4.767
F 2020.03.11 5.9 -2.961 -3.87 -7.74
F 2020.03.12 5.35 -4.889 -8.586 -12.391
F 2020.03.13 5.63 2.55 -2.679 -4.925
GPRO 2020.03.02 3.81
GPRO 2020.03.03 3.935 1.614
GPRO 2020.03.04 4.04 1.317 
GPRO 2020.03.05 3.74 -3.856 -3.639
GPRO 2020.03.06 3.7 -0.538 -3.99 
GPRO 2020.03.09 3.33 -5.263 -10.061 -11.417
GPRO 2020.03.10 3.64 4.448 1.041 -2.435
GPRO 2020.03.11 3.31 -4.748 -5.293 -8.732
GPRO 2020.03.12 2.69 -10.333 -17.039 -20.921
GPRO 2020.03.13 2.72 0.555 -11.974 -15.833
*/

In DolphinDB:

use mytt

close = 7.2 6.97 7.08 6.74 6.49 5.9 6.26 5.9 5.35 5.63
close = take(close, 100000)
timer x = mytt::AVEDEV(close, 100)

Using the AVEDEV function in the mytt module directly on a vector of length 100,000 takes 25 ms.

The corresponding Python code is as follows:

import numpy as np
from MyTT import *
import time

close = np.array([7.2,6.97,7.08,6.74,6.49,5.9,6.26,5.9,5.35,5.63])
close = np.tile(close,10000)
start_time = time.time()
x = AVEDEV(close, 100)
print("---%s seconds---" % (time.time() - start_time))

The AVEDEV function in the Python MyTT library takes 25,000 ms, which is 1,000 times longer than the AVEDEV function in the DolphinDB mytt module. The larger the test dataset, the more significant the performance difference. This performance gap stems from DolphinDB's highly optimized C++ implementation and vectorized execution engine, which avoids the overhead of Python's iterative processing.

• The test data is daily trading data for 2,919 securities on the Shanghai Stock Exchange in 2020, filtered to include only securities with more than 120 trading days. The dataset contains 686,104 records in total.

• The calculations are performed separately for each stock based on its stock code.

• For a fair comparison, both the DolphinDB and Python test code were run in a single thread.

The test results are shown in the following table:

The test results show that the functions in the DolphinDB mytt module significantly outperform those in the Python MyTT library, with speedups reaching up to 5,520× and typical around 30×.

Core Python pandas Test Code

data.groupby("symbol").apply(lambda x: RSI(np.array(x.close), N = 24))

Core DolphinDB Test Code

RSI = select symbol, tradedate, mytt::RSI(close, N=24) as `RSI from data context by symbol

Functions with different calculation results

• CROSS_, LONGCROSS

Cause

• Floating-point precision issues

• The CROSS_ and LONGCROSS functions involve comparisons of floating-point values. Before performing the comparison, the DolphinDB mytt module applies rounding to retain 6 decimal places, whereas the Python MyTT library does not perform similar rounding. Therefore, for values that are theoretically equal but differ slightly due to floating-point representation, the two implementations may produce different comparison results, as shown below:

Functions with different calculation results

• SUM, DMI, EMV, MASS, MFI, ASI

Cause:

• If the input vector starts with null values, calculation begins from the first non-null value. The DolphinDB mytt module follows the same calculation rules as the Python MyTT library.

• For a function with a rolling or cumulative window length of k, the first k - 1 results in each group are NULL. The DolphinDB mytt module follows the same calculation rules as the Python MyTT library.

• For a function with a rolling or cumulative window length of k, if null values occur after the first non-null value in a group, the Python MyTT library returns NaN for any window that contains Nan. The DolphinDB mytt module calculates the non-null elements in the window according to the calculation rules and returns a non-null result. DolphinDB's approach ensures continuous calculation in real-time streams even when data gaps occur.

DolphinDB code and result:

close = [99.9, NULL, 84.69, 31.38, 60.9, 83.3, 97.26, 98.67]
mytt::SUM(close, 5);

[,,,,276.87, 260.27, 357.53, 371.51]

Python code and result:

close = np.array([99.9, np.nan, 84.69, 31.38, 60.9, 83.3, 97.26, 98.67])
MyTT.SUM(close,5)

array([nan, nan, nan, nan, nan,  nan, 357.53, 371.51])

Taking a sliding window sum as an example, the second element of the close vector is NULL. When calculating the fifth element (60.9), the DolphinDB mytt module looks back over the preceding five-element window, [99.9, NULL, 84.69, 31.38, 60.9], and sums the non-null elements. Therefore, the fifth element of the result vector is 276.87.

The Python MyTT library returns NaN for any window that contains NaN, so the first six elements of the resulting vector are all NaN.

Except for the differences caused by floating-point precision and null value handling described above, the percentage error of the calculation results for all other functions is less than 1e-10.

The TOPRANGE function returns, for each element, how many periods the current value has been the highest within a recent window in a sequence S. For example, TOPRANGE(High) indicates the number of days since the most recent high.

The TOPRANGE function in the mytt module differs from the TOPRANGE function in MyTT when calculating the initial data. As shown in the following figure, for the High sequence 16.95, 17.31, 17.34, the corresponding TOPRANGE results are 0, 1, and 2, indicating that each value represents a new high over the preceding 0, 1, and 2 days, respectively. MyTT starts counting only after the data has first declined and then risen, whereas the mytt module starts counting from the beginning of the data. As a result, the calculation results may differ near the beginning of the sequence.

The LOWRANGE function behaves similarly.