Real tail of stream (last n elements)

Start with the Stream.from_periodic example and trace the accumulate(...).to_dataframe path used to retain DataFrame batches. The issue names no source file or test, so first determine where stream accumulation and window behavior are implemented. Done means agreeing on the API and semantics for exposing the last n elements, then verifying that behavior for the shown DataFrame workflow.

To my knowledge there is currently no way to show (or plot) the last n elements of a Streamz DataFrame.

I think this is a very useful function for debugging, but also for slower Streamz, for example CFD results from OpenFOAM.

Here is a naive implementation:

def tail(length):
    def tail_func(x, y):
        x = x.append(y, ignore_index=True)
        return x.iloc[-length:].reset_index(drop=True)
    return tail_func

Here is an example:
(thanks for the feedback @martindurant)

import pandas as pd
from streamz import Stream
import numpy as np

names = ["Mike", "Tim", "Anna", "Kim", "Andy"]

def emitter():
    n = np.random.randint(1, 5)
    return {"name": np.random.choice(names, n), "age": np.random.randint(18, 32, n)}

stream = Stream.from_periodic(emitter, 1)
df1 = pd.DataFrame({"name": ["test"], "age": [40]})
out = stream.map(pd.DataFrame).accumulate(tail(6), start=pd.DataFrame()).to_dataframe(df1)
# out.stream.sink(print)  # optional, or some other output

stream.start()

# if using a Jupyter Lab
out

This could either be the bare representation of window or could replace the tail function. I think the name tail suits this function better than the current one. What do you think?

Implementation wise one could be much more efficient by using .loc and round robin replacing values and saving the order somewhere else, but for a first go, I would stick by the naive version.