tsdisplay errs with input arrays y of length less than 51.

[jasmyace]

Describe the bug
Function tsdisplay, via its lag_max default value of 50, assumes that array y has at least 51 observations. A total of (at least) 51 observations is required to produce 50 lags.

Not directly specifying lag_max, when y is not of sufficient length, leads to a (perhaps) cryptic error.

To Reproduce

import pmdarima as pm
from pmdarima.utils import tsdisplay

x = pm.datasets.load_sunspots()
x51 = x[0:51]   # 51 obs.
x50 = x[0:50]   # 50 obs. 
x49 = x[0:49]   # 49 obs.
tsdisplay(x51)  # yay!  :)
tsdisplay(x50)  # oh no!  :(
tsdisplay(x49)  # oh no!  :(

Versions

System:
    python: 3.8.10 (default, May 19 2021, 11:01:55)  [Clang 10.0.0 ]
executable: /Users/Jason/opt/anaconda3/envs/python38_2/bin/python
   machine: macOS-10.14.6-x86_64-i386-64bit

Python dependencies:
        pip: 21.1.2
 setuptools: 52.0.0.post20210125
    sklearn: 0.24.2
statsmodels: 0.12.2
      numpy: 1.19.0
      scipy: 1.6.2
     Cython: 0.29.17
     pandas: 1.2.5
     joblib: 1.0.1
   pmdarima: 1.8.0

Not using the most recent version of pmdarima, but GitHub suggests most of pmdarima.utils has not undergone revisions recently.

Expected behavior
In the case that y is of length 50 or less, tsdisplay would ideally adapt to feed len(y) - 1, or an appropriate equivalent, into the internal call to plot_acf. For what it's worth, it looks like tsaplots uses np.arange to help in this area inplot_acf.

Actual behavior
Calling tsdisplay via

tsdisplay(x50)  # oh no!  :(

leads to

blahblahblah
ValueError: could not broadcast input array from shape (50) into shape (51)

Additional context
n/a

[tgsmith61591]

Hey great find and thanks so much for filing such a nicely documented issue. This is indeed a bug on our end, and it has to do with the lag_max=50 default. I will add a check to ensure lag_max is not larger than the series length; if it is, it should be an error, but it should be an error that the user can understand.

[tgsmith61591]

This is fixed in 1.8.3