Display loop mega-cleanup

The most important changes include:
- iterating the new `Flow` type and updating graphics
- adding detailed profiling
- increasing the min uppx before graphics updates are throttled
- including the L1 spread in y-range calcs so that you never have the
  bid/ask go "out of view"..
- pass around `Flow`s instead of shms
- drop all the old prototyped downsampling code
big_data_lines
Tyler Goodlet 2022-04-07 14:11:01 -04:00
parent 7c615a403b
commit ee831baeb3
1 changed files with 59 additions and 108 deletions

View File

@ -30,7 +30,6 @@ import numpy as np
import tractor
import trio
import pyqtgraph as pg
from PyQt5.QtCore import QLineF
from .. import brokers
from ..data.feed import open_feed
@ -73,13 +72,20 @@ _tick_groups = {
}
# TODO: delegate this to each `Flow.maxmin()` which includes
# caching and further we should implement the following stream based
# approach, likely with ``numba``:
# https://arxiv.org/abs/cs/0610046
# https://github.com/lemire/pythonmaxmin
def chart_maxmin(
chart: ChartPlotWidget,
ohlcv_shm: ShmArray,
vlm_chart: Optional[ChartPlotWidget] = None,
) -> tuple[
tuple[int, int, int, int],
float,
float,
float,
@ -88,11 +94,6 @@ def chart_maxmin(
Compute max and min datums "in view" for range limits.
'''
# TODO: implement this
# https://arxiv.org/abs/cs/0610046
# https://github.com/lemire/pythonmaxmin
# array = chart._arrays[chart.name]
array = ohlcv_shm.array
ifirst = array[0]['index']
@ -105,18 +106,23 @@ def chart_maxmin(
chart.default_view()
return (last_bars_range, 0, 0, 0)
mx, mn = np.nanmax(in_view['high']), np.nanmin(in_view['low'])
# TODO: when we start using line charts, probably want to make
# this an overloaded call on our `DataView
# sym = chart.name
# mx, mn = np.nanmax(in_view[sym]), np.nanmin(in_view[sym])
mx, mn = (
np.nanmax(in_view['high']),
np.nanmin(in_view['low'],)
)
mx_vlm_in_view = 0
if vlm_chart:
mx_vlm_in_view = np.max(in_view['volume'])
mx_vlm_in_view = np.max(
in_view['volume']
)
return last_bars_range, mx, max(mn, 0), mx_vlm_in_view
return (
last_bars_range,
mx,
max(mn, 0), # presuming price can't be negative?
mx_vlm_in_view,
)
@dataclass
@ -272,8 +278,9 @@ async def graphics_update_loop(
chart.default_view()
# main loop
# main real-time quotes update loop
async for quotes in stream:
ds.quotes = quotes
quote_period = time.time() - last_quote
quote_rate = round(
@ -311,36 +318,40 @@ def graphics_update_cycle(
trigger_all: bool = False, # flag used by prepend history updates
) -> None:
# TODO: eventually optimize this whole graphics stack with ``numba``
# hopefully XD
chart = ds.chart
profiler = pg.debug.Profiler(
msg=f'Graphics loop cycle for: `{chart.name}`',
disabled=True, # not pg_profile_enabled(),
gt=1/12 * 1e3,
# gt=ms_slower_then,
)
# unpack multi-referenced components
chart = ds.chart
vlm_chart = ds.vlm_chart
l1 = ds.l1
ohlcv = ds.ohlcv
array = ohlcv.array
vars = ds.vars
tick_margin = vars['tick_margin']
update_uppx = 5
update_uppx = 6
for sym, quote in ds.quotes.items():
# compute the first available graphic's x-units-per-pixel
xpx = vlm_chart.view.x_uppx()
# NOTE: vlm may be written by the ``brokerd`` backend
# event though a tick sample is not emitted.
# TODO: show dark trades differently
# https://github.com/pikers/piker/issues/116
# NOTE: this used to be implemented in a dedicated
# "increment tas": ``check_for_new_bars()`` but it doesn't
# "increment task": ``check_for_new_bars()`` but it doesn't
# make sense to do a whole task switch when we can just do
# this simple index-diff and all the fsp sub-curve graphics
# are diffed on each draw cycle anyway; so updates to the
@ -364,60 +375,6 @@ def graphics_update_cycle(
profiler('maxmin call')
liv = r > i_step # the last datum is in view
# compute the first available graphic's x-units-per-pixel
xpx = vlm_chart.view.xs_in_px()
# print(f'vlm xpx {xpx}')
in_view = chart.in_view(ohlcv.array)
if lbar != rbar:
# view box width in pxs
w = chart.view.boundingRect().width()
# TODO: a better way to get this?
# i would guess the esiest way is to just
# get the ``.boundingRect()`` of the curve
# in view but maybe there's something smarter?
# Currently we're just mapping the rbar, lbar to
# pixels via:
cw = chart.view.mapViewToDevice(QLineF(lbar, 0, rbar, 0)).length()
# is this faster?
# cw = chart.mapFromView(QLineF(lbar, 0 , rbar, 0)).length()
profiler(
f'view width pxs: {w}\n'
f'curve width pxs: {cw}\n'
f'sliced in view: {in_view.size}'
)
# compress bars to m4 line(s) if uppx is high enough
# if in_view.size > cw:
# from ._compression import ds_m4, hl2mxmn
# mxmn, x = hl2mxmn(in_view)
# profiler('hl tracer')
# nb, x, y = ds_m4(
# x=x,
# y=mxmn,
# # TODO: this needs to actually be the width
# # in pixels of the visible curve since we don't
# # want to downsample any 'zeros' around the curve,
# # just the values that make up the curve graphic,
# # i think?
# px_width=cw,
# )
# profiler(
# 'm4 downsampled\n'
# f' ds bins: {nb}\n'
# f' x.shape: {x.shape}\n'
# f' y.shape: {y.shape}\n'
# f' x: {x}\n'
# f' y: {y}\n'
# )
# assert y.size == mxmn.size
# don't real-time "shift" the curve to the
# left unless we get one of the following:
if (
@ -435,7 +392,9 @@ def graphics_update_cycle(
if vlm_chart:
# always update y-label
ds.vlm_sticky.update_from_data(*array[-1][['index', 'volume']])
ds.vlm_sticky.update_from_data(
*array[-1][['index', 'volume']]
)
if (
(xpx < update_uppx or i_diff > 0)
@ -464,17 +423,17 @@ def graphics_update_cycle(
if (
mx_vlm_in_view != vars['last_mx_vlm']
):
# print(f'mx vlm: {last_mx_vlm} -> {mx_vlm_in_view}')
yrange = (0, mx_vlm_in_view * 1.375)
vlm_chart.view._set_yrange(
yrange=yrange,
)
# print(f'mx vlm: {last_mx_vlm} -> {mx_vlm_in_view}')
vars['last_mx_vlm'] = mx_vlm_in_view
for curve_name, flow in vlm_chart._flows.items():
update_fsp_chart(
vlm_chart,
flow.shm,
flow,
curve_name,
array_key=curve_name,
)
@ -529,7 +488,6 @@ def graphics_update_cycle(
# current) tick first order as an optimization where we only
# update from the last tick from each type class.
# last_clear_updated: bool = False
# for typ, tick in reversed(lasts.items()):
# update ohlc sampled price bars
if (
@ -541,7 +499,7 @@ def graphics_update_cycle(
array,
)
# iterate in FIFO order per frame
# iterate in FIFO order per tick-frame
for typ, tick in lasts.items():
price = tick.get('price')
@ -612,42 +570,34 @@ def graphics_update_cycle(
if (
(mx > vars['last_mx']) or (mn < vars['last_mn'])
and not chart._static_yrange == 'axis'
and r > i_step # the last datum is in view
):
# print(f'new y range: {(mn, mx)}')
chart.view._set_yrange(
yrange=(mn, mx),
# TODO: we should probably scale
# the view margin based on the size
# of the true range? This way you can
# slap in orders outside the current
# L1 (only) book range.
# range_margin=0.1,
)
main_vb = chart.view
if (
main_vb._ic is None
or not main_vb._ic.is_set()
):
main_vb._set_yrange(
# TODO: we should probably scale
# the view margin based on the size
# of the true range? This way you can
# slap in orders outside the current
# L1 (only) book range.
# range_margin=0.1,
yrange=(mn, mx),
)
vars['last_mx'], vars['last_mn'] = mx, mn
# run synchronous update on all derived fsp subplots
for name, subchart in ds.linked.subplots.items():
if name == 'volume':
# run synchronous update on all linked flows
for curve_name, flow in chart._flows.items():
# TODO: should the "main" (aka source) flow be special?
if curve_name == chart.data_key:
continue
update_fsp_chart(
subchart,
subchart._shm,
# XXX: do we really needs seperate names here?
name,
array_key=name,
)
subchart.cv._set_yrange()
# TODO: all overlays on all subplots..
# run synchronous update on all derived overlays
for curve_name, flow in chart._flows.items():
update_fsp_chart(
chart,
flow.shm,
flow,
curve_name,
array_key=curve_name,
)
@ -743,6 +693,7 @@ async def display_symbol_data(
# TODO: a data view api that makes this less shit
chart._shm = ohlcv
chart._flows[chart.data_key].shm = ohlcv
# NOTE: we must immediately tell Qt to show the OHLC chart
# to avoid a race where the subplots get added/shown to
@ -799,7 +750,7 @@ async def display_symbol_data(
# that it isn't double rendered in the display loop
# above since we do a maxmin calc on the volume data to
# determine if auto-range adjustements should be made.
linked.subplots.pop('volume', None)
# linked.subplots.pop('volume', None)
# TODO: make this not so shit XD
# close group status