tsfeatures（）无意义错误（需要至少两个非NA值进行插值，但所有值均为非NA）

Question

tsfeatures() says that approx() fails due to needing at least two non-NA values to interpolate. I can't find any NAs in my time series or the time series index, and I can construct a toy example with no NA values, ordered and unique indices, and still reproduce the error.

There appear to be a number of previous questions concerning R's tsfeatures() function, but I can't find one that explains why I'm receiving the error:

Error in approx(idx, x[idx], tt, rule = 2) : 
  need at least two non-NA values to interpolate

# datetimes has type POSIXct
Browse[3]> class(datetimes[1:3])
[1] "POSIXct" "POSIXt"

# the first three date times are unique and ordered
Browse[3]> datetimes[1:3]
[1] "2019-06-18 19:59:48 PDT" "2019-06-18 20:00:48 PDT" "2019-06-18 20:01:48 PDT"

# create our time series
Browse[3]> temp = xts(x = c(0, 0, 0), order.by = datetimes[1:3])
Browse[3]> temp
                    [,1]
2019-06-18 19:59:48    0
2019-06-18 20:00:48    0
2019-06-18 20:01:48    0

# try to get our features:
Browse[3]> tsfeatures(temp)
Error in approx(idx, x[idx], tt, rule = 2) : 
  need at least two non-NA values to interpolate

Edit: My teammate and I discovered that the error message is displayed if the time series consists of entirely the same value:

Browse[3]> tsfeatures(xts(x = c(1, 1, 1), order.by = datetimes[1:3]))
Error in approx(idx, x[idx], tt, rule = 2) : 
  need at least two non-NA values to interpolate

Browse[3]> tsfeatures(xts(x = c(1, 2, 1), order.by = datetimes[1:3]))
# A tibble: 1 x 16
  frequency nperiods seasonal_period    trend spike linearity curvature e_acf1 e_acf10 entropy x_acf1 x_acf10 diff1_acf1 diff1_acf10 diff2_acf1 diff2_acf10
      <dbl>    <dbl>           <dbl>    <dbl> <dbl>     <dbl>     <dbl>  <dbl>   <dbl>   <dbl>  <dbl>   <dbl>      <dbl>       <dbl>      <dbl>       <dbl>
1    0.0167        0          0.0167 1.11e-16 0.333 -6.04e-32  3.49e-32 -0.667      NA      NA -0.667      NA       -0.5        0.25         NA           0

Can someone explain the error message?

Answer 1

The error is due to the fact that tsfeatures() automatically scales the time series by default. When you pass a constant series, scaling reduces the series to NaN s. This is now prevented in the current github version -- fixed here .