summarize_ema_half_life {sparklyr.flint}

EMA half-life summarizer

Description

Calculate the exponential moving average of a time series using the half- life specified and store the result in a new column named '<column>_ema' See https://github.com/twosigma/flint/blob/master/doc/ema.md for details on different EMA implementations.

Usage

summarize_ema_half_life(
  ts_rdd,
  column,
  half_life_duration,
  window = NULL,
  time_column = "time",
  interpolation = c("previous", "linear", "current"),
  convention = c("legacy", "convolution", "core"),
  key_columns = list()
)

Arguments

ts_rdd

Timeseries RDD being summarized

column

Column to be summarized

half_life_duration

A time duration specified in string form (e.g., "1d", "1h", "15m", etc) representing the half-life duration

window

Either an R expression specifying time windows to be summarized (e.g., 'in_past("1h")' to summarize the EMA of 'column' within the time interval of [t - 1h, t] for each timestamp 't', 'in_future("5s")' to summarize EMA of 'column' within the time interval of [t, t + 5s] for each timestamp 't'), or 'NULL' to summarize EMA of 'column' within the time interval of (-inf, t] for each timestamp 't'

time_column

Name of the column containing timestamps (default: "time")

interpolation

Method used for interpolating values between two consecutive data points, must be one of "previous", "linear", and "current" (default: "previous"). See https://github.com/twosigma/flint/blob/master/doc/ema.md for details on different interpolation methods.

convention

Convolution convention, must be one of "convolution", "core", and "legacy" (default: "legacy"). See https://github.com/twosigma/flint/blob/master/doc/ema.md for details.

key_columns

Optional list of columns that will form an equivalence relation associating each record with the time series it belongs to (i.e., any 2 records having equal values in those columns will be associated with the same time series, and any 2 records having differing values in those columns are considered to be from 2 separate time series and will therefore be summarized separately) By default, 'key_colums' is empty and all records are considered to be part of a single time series.

See Also

Other summarizers: ols_regression(), summarize_avg(), summarize_corr2(), summarize_corr(), summarize_count(), summarize_covar(), summarize_dot_product(), summarize_ewma(), summarize_geometric_mean(), summarize_kurtosis(), summarize_max(), summarize_min(), summarize_nth_central_moment(), summarize_nth_moment(), summarize_product(), summarize_quantile(), summarize_skewness(), summarize_stddev(), summarize_sum(), summarize_var(), summarize_weighted_avg(), summarize_weighted_corr(), summarize_weighted_covar(), summarize_z_score()

Examples

library(sparklyr)
library(sparklyr.flint)

sc <- try_spark_connect(master = "local")

if (!is.null(sc)) {
  price_sdf <- copy_to(
    sc,
    data.frame(time = seq(1000), price = rnorm(1000))
  )
  ts <- fromSDF(price_sdf, is_sorted = TRUE, time_unit = "SECONDS")
  ts_ema <- summarize_ema_half_life(
    ts,
    column = "price",
    half_life_duration = "100s"
  )
} else {
  message("Unable to establish a Spark connection!")
}