John Belmonte

After spending many weeks on this last year, I'm fairly confident saying 1) numba does not operate in its fast mode with python arrays, and 2) numpy by itself will...

shows clearly that numba only works well with numpy arrays: ```python import numba import numpy as np def argmin_diff(a): i, i_m = 1, -1 m = inf last_item = a[0]...

> There is something wrong here, because the bisect code in distogram is a copy of the stdlib read below that stdlib function definition: ```python # Overwrite above definitions with...

> Otherwise I have to convert the bins to a value-only list In the implementations I've seen, (point, count) tuple works fine and gives the most performance opportunities in pure...

My `ApproximateHistogram` in Python using only stdlib is now contained within a performance timer project I've published: https://github.com/belm0/perf-timer/blob/063ef693702a9a8dc227d7be14b89c096c643873/src/perf_timer/_histogram.py

So using stdlib bisect / tuples made a huge difference, or the old numbers were mistaken? ```diff Interpreter Operation Numpy Req/s ============ ========== ======= ========== - CPython 3.7 update no...

It looks like the laptop numbers need to be updated? ``` On a i7-9800X Intel CPU, performances are: ============ ========== ======= ========== Interpreter Operation Numpy Req/s ============ ========== ======= ==========...

> Yes, for CPython this is a huge bump. @MainRo would you consider adding me to the README credits as I helped make this package about 10x faster and produce...

Thank you for your suggestion and use case explanation. Besides caching that you mention, using a JIT is another case where the first run (and perhaps later runs if the...

Though it's a roundabout way, histogram can provide "min"-- it's the same as 0 percentile. Here you will get a report of min, median, and max: ```python PerfTimer('foo', observer=HistogramObserver, quantiles=(0,...