openmw/scripts/osg_stats.py

#!/usr/bin/env python3
"""
osg_stats.py is a script to analyze OpenSceneGraph log. It parses given file
and builds timeseries, histograms, plots, calculate statistics for a given
set of keys over given range of frames.
"""

import click
import collections
import matplotlib.pyplot
import numpy
import statistics
import sys
import termtables
import re


@click.command()
@click.option('--print_keys', is_flag=True,
              help='Print a list of all present keys in the input file.')
@click.option('--regexp_match', is_flag=True,
              help='Use all metric that match given key. '
                   'Can be used with stats, timeseries, commulative_timeseries, hist, hist_threshold')
@click.option('--timeseries', type=str, multiple=True,
              help='Show a graph for given metric over time.')
@click.option('--commulative_timeseries', type=str, multiple=True,
              help='Show a graph for commulative sum of a given metric over time.')
@click.option('--hist', type=str, multiple=True,
              help='Show a histogram for all values of given metric.')
@click.option('--hist_ratio', nargs=2, type=str, multiple=True,
              help='Show a histogram for a ratio of two given metric (first / second). '
                   'Format: --hist_ratio <first_metric> <second_metric>.')
@click.option('--stdev_hist', nargs=2, type=str, multiple=True,
              help='Show a histogram for a standard deviation of a given metric at given scale (number). '
                   'Format: --stdev_hist <metric> <scale>.')
@click.option('--plot', nargs=3, type=str, multiple=True,
              help='Show a 2D plot for relation between two metrix (first is axis x, second is y)'
                   'using one of aggregation functions (mean, median). For example show a relation '
                   'between Physics Actors and physics_time_taken. Format: --plot <x> <y> <function>.')
@click.option('--stats', type=str, multiple=True,
              help='Print table with stats for a given metric containing min, max, mean, median etc.')
@click.option('--precision', type=int,
              help='Format floating point numbers with given precision')
@click.option('--timeseries_sum', is_flag=True,
              help='Add a graph to timeseries for a sum per frame of all given timeseries metrics.')
@click.option('--commulative_timeseries_sum', is_flag=True,
              help='Add a graph to timeseries for a sum per frame of all given commulative timeseries.')
@click.option('--stats_sum', is_flag=True,
              help='Add a row to stats table for a sum per frame of all given stats metrics.')
@click.option('--begin_frame', type=int, default=0,
              help='Start processing from this frame.')
@click.option('--end_frame', type=int, default=sys.maxsize,
              help='End processing at this frame.')
@click.option('--frame_number_name', type=str, default='FrameNumber',
              help='Frame number metric name.')
@click.option('--hist_threshold', type=str, multiple=True,
              help='Show a histogram for given metric only for frames with threshold_name metric over threshold_value.')
@click.option('--threshold_name', type=str, default='Frame duration',
              help='Frame duration metric name.')
@click.option('--threshold_value', type=float, default=1.05/60,
              help='Threshold for hist_over.')
@click.argument('path', type=click.Path(), nargs=-1)
def main(print_keys, regexp_match, timeseries, hist, hist_ratio, stdev_hist, plot, stats, precision,
         timeseries_sum, stats_sum, begin_frame, end_frame, path,
         commulative_timeseries, commulative_timeseries_sum, frame_number_name,
         hist_threshold, threshold_name, threshold_value):
    sources = {v: list(read_data(v)) for v in path} if path else {'stdin': list(read_data(None))}
    keys = collect_unique_keys(sources)
    frames, begin_frame, end_frame = collect_per_frame(
        sources=sources, keys=keys, begin_frame=begin_frame,
        end_frame=end_frame, frame_number_name=frame_number_name,
    )
    if print_keys:
        for v in keys:
            print(v)
    def matching_keys(patterns):
        if regexp_match:
            return [key for pattern in patterns for key in keys if re.search(pattern, key)]
        return keys
    if timeseries:
        draw_timeseries(sources=frames, keys=matching_keys(timeseries), add_sum=timeseries_sum,
                        begin_frame=begin_frame, end_frame=end_frame)
    if commulative_timeseries:
        draw_commulative_timeseries(sources=frames, keys=matching_keys(commulative_timeseries), add_sum=commulative_timeseries_sum,
                                    begin_frame=begin_frame, end_frame=end_frame)
    if hist:
        draw_hists(sources=frames, keys=matching_keys(hist))
    if hist_ratio:
        draw_hist_ratio(sources=frames, pairs=hist_ratio)
    if stdev_hist:
        draw_stdev_hists(sources=frames, stdev_hists=stdev_hist)
    if plot:
        draw_plots(sources=frames, plots=plot)
    if stats:
        print_stats(sources=frames, keys=matching_keys(stats), stats_sum=stats_sum, precision=precision)
    if hist_threshold:
        draw_hist_threshold(sources=frames, keys=matching_keys(hist_threshold), begin_frame=begin_frame,
                            threshold_name=threshold_name, threshold_value=threshold_value)
    matplotlib.pyplot.show()


def read_data(path):
    with open(path) if path else sys.stdin as stream:
        frame = dict()
        camera = 0
        for line in stream:
            if line.startswith('Stats Viewer'):
                if frame:
                    camera = 0
                    yield frame
                _, _, key, value = line.split(' ')
                frame = {key: int(value)}
            elif line.startswith('Stats Camera'):
                camera += 1
            elif line.startswith('    '):
                key, value = line.strip().rsplit(maxsplit=1)
                if camera:
                    key = f'{key} Camera {camera}'
                frame[key] = to_number(value)


def collect_per_frame(sources, keys, begin_frame, end_frame, frame_number_name):
    assert begin_frame < end_frame
    result = collections.defaultdict(lambda: collections.defaultdict(list))
    begin_frame = max(begin_frame, min(v[0][frame_number_name] for v in sources.values()))
    end_frame = min(end_frame, begin_frame + max(len(v) for v in sources.values()))
    for name in sources.keys():
        for key in keys:
            result[name][key] = [0] * (end_frame - begin_frame)
    for name, frames in sources.items():
        for frame in frames:
            number = frame[frame_number_name]
            if begin_frame <= number < end_frame:
                index = number - begin_frame
                for key in keys:
                    if key in frame:
                        result[name][key][index] = frame[key]
    for name in result.keys():
        for key in keys:
            result[name][key] = numpy.array(result[name][key])
    return result, begin_frame, end_frame


def collect_unique_keys(sources):
    result = set()
    for frames in sources.values():
        for frame in frames:
            for key in frame.keys():
                result.add(key)
    return sorted(result)


def draw_timeseries(sources, keys, add_sum, begin_frame, end_frame):
    fig, ax = matplotlib.pyplot.subplots()
    x = numpy.array(range(begin_frame, end_frame))
    for name, frames in sources.items():
        for key in keys:
            ax.plot(x, frames[key], label=f'{key}:{name}')
        if add_sum:
            ax.plot(x, numpy.sum(list(frames[k] for k in keys), axis=0), label=f'sum:{name}')
    ax.grid(True)
    ax.legend()
    fig.canvas.manager.set_window_title('timeseries')


def draw_commulative_timeseries(sources, keys, add_sum, begin_frame, end_frame):
    fig, ax = matplotlib.pyplot.subplots()
    x = numpy.array(range(begin_frame, end_frame))
    for name, frames in sources.items():
        for key in keys:
            ax.plot(x, numpy.cumsum(frames[key]), label=f'{key}:{name}')
        if add_sum:
            ax.plot(x, numpy.cumsum(numpy.sum(list(frames[k] for k in keys), axis=0)), label=f'sum:{name}')
    ax.grid(True)
    ax.legend()
    fig.canvas.manager.set_window_title('commulative_timeseries')


def draw_hists(sources, keys):
    fig, ax = matplotlib.pyplot.subplots()
    bins = numpy.linspace(
        start=min(min(min(v) for k, v in f.items() if k in keys) for f in sources.values()),
        stop=max(max(max(v) for k, v in f.items() if k in keys) for f in sources.values()),
        num=20,
    )
    for name, frames in sources.items():
        for key in keys:
            ax.hist(frames[key], bins=bins, label=f'{key}:{name}', alpha=1 / (len(keys) * len(sources)))
    ax.set_xticks(bins)
    ax.grid(True)
    ax.legend()
    fig.canvas.manager.set_window_title('hists')


def draw_hist_ratio(sources, pairs):
    fig, ax = matplotlib.pyplot.subplots()
    bins = numpy.linspace(
        start=min(min(min(a / b for a, b in zip(f[a], f[b])) for a, b in pairs) for f in sources.values()),
        stop=max(max(max(a / b for a, b in zip(f[a], f[b])) for a, b in pairs) for f in sources.values()),
        num=20,
    )
    for name, frames in sources.items():
        for a, b in pairs:
            ax.hist(frames[a] / frames[b], bins=bins, label=f'{a} / {b}:{name}', alpha=1 / (len(pairs) * len(sources)))
    ax.set_xticks(bins)
    ax.grid(True)
    ax.legend()
    fig.canvas.manager.set_window_title('hists_ratio')


def draw_stdev_hists(sources, stdev_hists):
    for key, scale in stdev_hists:
        scale = float(scale)
        fig, ax = matplotlib.pyplot.subplots()
        first_frames = next(v for v in sources.values())
        median = statistics.median(first_frames[key])
        stdev = statistics.stdev(first_frames[key])
        start = median - stdev / 2 * scale
        stop = median + stdev / 2 * scale
        bins = numpy.linspace(start=start, stop=stop, num=9)
        for name, frames in sources.items():
            values = [v for v in frames[key] if start <= v <= stop]
            ax.hist(values, bins=bins, label=f'{key}:{name}', alpha=1 / (len(stdev_hists) * len(sources)))
        ax.set_xticks(bins)
        ax.grid(True)
        ax.legend()
        fig.canvas.manager.set_window_title('stdev_hists')


def draw_plots(sources, plots):
    fig, ax = matplotlib.pyplot.subplots()
    for name, frames in sources.items():
        for x_key, y_key, agg in plots:
            if agg is None:
                ax.plot(frames[x_key], frames[y_key], label=f'x={x_key}, y={y_key}:{name}')
            elif agg:
                agg_f = dict(
                    mean=statistics.mean,
                    median=statistics.median,
                )[agg]
                grouped = collections.defaultdict(list)
                for x, y in zip(frames[x_key], frames[y_key]):
                    grouped[x].append(y)
                aggregated = sorted((k, agg_f(v)) for k, v in grouped.items())
                ax.plot(
                    numpy.array([v[0] for v in aggregated]),
                    numpy.array([v[1] for v in aggregated]),
                    label=f'x={x_key}, y={y_key}, agg={agg}:{name}',
                )
    ax.grid(True)
    ax.legend()
    fig.canvas.manager.set_window_title('plots')


def print_stats(sources, keys, stats_sum, precision):
    stats = list()
    for name, frames in sources.items():
        for key in keys:
            stats.append(make_stats(source=name, key=key, values=filter_not_none(frames[key]), precision=precision))
        if stats_sum:
            stats.append(make_stats(source=name, key='sum', values=sum_multiple(frames, keys), precision=precision))
    metrics = list(stats[0].keys())
    termtables.print(
        [list(v.values()) for v in stats],
        header=metrics,
        style=termtables.styles.markdown,
    )


def draw_hist_threshold(sources, keys, begin_frame, threshold_name, threshold_value):
    for name, frames in sources.items():
        indices = [n for n, v in enumerate(frames[threshold_name]) if v > threshold_value]
        numbers = [v + begin_frame for v in indices]
        x = [v for v in range(0, len(indices))]
        fig, ax = matplotlib.pyplot.subplots()
        ax.set_title(f'Frames with "{threshold_name}" > {threshold_value} ({len(indices)})')
        ax.bar(x, [frames[threshold_name][v] for v in indices], label=threshold_name, color='black', alpha=0.2)
        prev = 0
        for key in keys:
            values = [frames[key][v] for v in indices]
            ax.bar(x, values, bottom=prev, label=key)
            prev = values
        ax.hlines(threshold_value, x[0] - 1, x[-1] + 1, color='black', label='threshold', linestyles='dashed')
        ax.xaxis.set_major_locator(matplotlib.pyplot.FixedLocator(x))
        ax.xaxis.set_major_formatter(matplotlib.pyplot.FixedFormatter(numbers))
        ax.grid(True)
        ax.legend()
        fig.canvas.manager.set_window_title(f'hist_threshold:{name}')


def filter_not_none(values):
    return [v for v in values if v is not None]


def fixed_float(value, precision):
    return '{v:.{p}f}'.format(v=value, p=precision) if precision else value


def sum_multiple(frames, keys):
    result = collections.Counter()
    for key in keys:
        values = frames[key]
        for i, value in enumerate(values):
            if value is not None:
                result[i] += float(value)
    return numpy.array([result[k] for k in sorted(result.keys())])


def make_stats(source, key, values, precision):
    return collections.OrderedDict(
        source=source,
        key=key,
        number=len(values),
        min=fixed_float(min(values), precision),
        max=fixed_float(max(values), precision),
        mean=fixed_float(statistics.mean(values), precision),
        median=fixed_float(statistics.median(values), precision),
        stdev=fixed_float(statistics.stdev(values), precision),
        q95=fixed_float(numpy.quantile(values, 0.95), precision),
    )


def to_number(value):
    try:
        return int(value)
    except ValueError:
        return float(value)


if __name__ == '__main__':
    main()