Lazily import pandas to speedup non-pandas use

[hugovk]

I'm using pytablewriter in some CLIs, and noticed they were sometimes slow so I profiled it.

Here's how to use python -X importtime and tuna to identify bottlenecks: https://medium.com/alan/how-we-improved-our-python-backend-start-up-time-2c33cd4873c8

For example with this test script:

from pytablewriter import MarkdownTableWriter

def main():
    writer = MarkdownTableWriter(
        table_name="example_table",
        headers=["int", "float", "str", "bool", "mix", "time"],
        value_matrix=[
            [0,   0.1,      "hoge", True,   0,      "2017-01-01 03:04:05+0900"],
            [2,   "-2.23",  "foo",  False,  None,   "2017-12-23 45:01:23+0900"],
            [3,   0,        "bar",  "true",  "inf", "2017-03-03 33:44:55+0900"],
            [-10, -9.9,     "",     "FALSE", "nan", "2017-01-01 00:00:00+0900"],
        ],
    )
    return writer.dumps()

Then run:

python -m pip install tuna
python -X importtime -c "import pandastest; pandastest.main()" 2> pandastest-master.log
tuna pandastest-master.log

Shows most of the import time comes from pandas:

It takes almost half a second to import pandas, 78.1% of the total 0.63s time.

This is not surprising, pandas and its dependency NumPy are big libraries.

However with this PR, if we lazily import pandas, that is, only import it when needed, we get a big speedup for all the non-pandas use cases, which covers a lot of pytablewriter's formats:

Now it only takes 0.164s for the whole program, a huge improvement over the 0.63s before, and very noticeable on the command line.

Another quick before and after comparison:

$ time python pandastest.py
python3 pandastest.py  0.95s user 2.02s system 432% cpu 0.687 total
$ time python pandastest.py
python3 pandastest.py  0.10s user 0.03s system 57% cpu 0.223 total

[thombashi]

Great analysis.
Thank you for your contribution.