piker/piker/ui/_flows.py

925 lines
28 KiB
Python

# piker: trading gear for hackers
# Copyright (C) Tyler Goodlet (in stewardship for pikers)
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <https://www.gnu.org/licenses/>.
'''
High level streaming graphics primitives.
This is an intermediate layer which associates real-time low latency
graphics primitives with underlying FSP related data structures for fast
incremental update.
'''
from __future__ import annotations
from functools import partial
from typing import (
Optional,
Callable,
)
import msgspec
import numpy as np
from numpy.lib import recfunctions as rfn
import pyqtgraph as pg
from PyQt5.QtGui import QPainterPath
from PyQt5.QtCore import (
# Qt,
QLineF,
# QSizeF,
QRectF,
# QPointF,
)
from ..data._sharedmem import (
ShmArray,
# open_shm_array,
)
from .._profile import pg_profile_enabled, ms_slower_then
from ._ohlc import (
BarItems,
gen_qpath,
)
from ._curve import (
FastAppendCurve,
# step_path_arrays_from_1d,
)
from ._compression import (
# ohlc_flatten,
ds_m4,
)
from ..log import get_logger
log = get_logger(__name__)
# class FlowsTable(msgspec.Struct):
# '''
# Data-AGGRegate: high level API onto multiple (categorized)
# ``Flow``s with high level processing routines for
# multi-graphics computations and display.
# '''
# flows: dict[str, np.ndarray] = {}
# @classmethod
# def from_token(
# cls,
# shm_token: tuple[
# str,
# str,
# tuple[str, str],
# ],
# ) -> Renderer:
# shm = attach_shm_array(token)
# return cls(shm)
def rowarr_to_path(
rows_array: np.ndarray,
x_basis: np.ndarray,
flow: Flow,
) -> QPainterPath:
# TODO: we could in theory use ``numba`` to flatten
# if needed?
# to 1d
y = rows_array.flatten()
return pg.functions.arrayToQPath(
# these get passed at render call time
x=x_basis[:y.size],
y=y,
connect='all',
finiteCheck=False,
path=flow.path,
)
def mk_ohlc_flat_copy(
ohlc_shm: ShmArray,
# XXX: we bind this in currently..
# x_basis: np.ndarray,
# vr: Optional[slice] = None,
) -> tuple[np.ndarray, np.ndarray]:
'''
Return flattened-non-copy view into an OHLC shm array.
'''
ohlc = ohlc_shm._array[['open', 'high', 'low', 'close']]
# if vr:
# ohlc = ohlc[vr]
# x = x_basis[vr]
unstructured = rfn.structured_to_unstructured(
ohlc,
copy=False,
)
# breakpoint()
y = unstructured.flatten()
# x = x_basis[:y.size]
return y
class Flow(msgspec.Struct): # , frozen=True):
'''
(Financial Signal-)Flow compound type which wraps a real-time
shm array stream with displayed graphics (curves, charts)
for high level access and control as well as efficient incremental
update.
The intention is for this type to eventually be capable of shm-passing
of incrementally updated graphics stream data between actors.
'''
name: str
plot: pg.PlotItem
graphics: pg.GraphicsObject
_shm: ShmArray
is_ohlc: bool = False
render: bool = True # toggle for display loop
gy: Optional[ShmArray] = None
gx: Optional[np.ndarray] = None
_iflat_last: int = 0
_iflat_first: int = 0
_last_uppx: float = 0
_in_ds: bool = False
_graphics_tranform_fn: Optional[Callable[ShmArray, np.ndarray]] = None
# map from uppx -> (downsampled data, incremental graphics)
_src_r: Optional[Renderer] = None
_render_table: dict[
Optional[int],
tuple[Renderer, pg.GraphicsItem],
] = {}
# TODO: hackery to be able to set a shm later
# but whilst also allowing this type to hashable,
# likely will require serializable token that is used to attach
# to the underlying shm ref after startup?
# _shm: Optional[ShmArray] = None # currently, may be filled in "later"
# last read from shm (usually due to an update call)
_last_read: Optional[np.ndarray] = None
# cache of y-range values per x-range input.
_mxmns: dict[tuple[int, int], tuple[float, float]] = {}
@property
def shm(self) -> ShmArray:
return self._shm
# TODO: remove this and only allow setting through
# private ``._shm`` attr?
@shm.setter
def shm(self, shm: ShmArray) -> ShmArray:
print(f'{self.name} DO NOT SET SHM THIS WAY!?')
self._shm = shm
def maxmin(
self,
lbar,
rbar,
) -> tuple[float, float]:
'''
Compute the cached max and min y-range values for a given
x-range determined by ``lbar`` and ``rbar``.
'''
rkey = (lbar, rbar)
cached_result = self._mxmns.get(rkey)
if cached_result:
return cached_result
shm = self.shm
if shm is None:
mxmn = None
else: # new block for profiling?..
arr = shm.array
# build relative indexes into shm array
# TODO: should we just add/use a method
# on the shm to do this?
ifirst = arr[0]['index']
slice_view = arr[
lbar - ifirst:
(rbar - ifirst) + 1
]
if not slice_view.size:
mxmn = None
else:
if self.is_ohlc:
ylow = np.min(slice_view['low'])
yhigh = np.max(slice_view['high'])
else:
view = slice_view[self.name]
ylow = np.min(view)
yhigh = np.max(view)
mxmn = ylow, yhigh
if mxmn is not None:
# cache new mxmn result
self._mxmns[rkey] = mxmn
return mxmn
def view_range(self) -> tuple[int, int]:
'''
Return the indexes in view for the associated
plot displaying this flow's data.
'''
vr = self.plot.viewRect()
return int(vr.left()), int(vr.right())
def datums_range(self) -> tuple[
int, int, int, int, int, int
]:
'''
Return a range tuple for the datums present in view.
'''
l, r = self.view_range()
# TODO: avoid this and have shm passed
# in earlier.
if self.shm is None:
# haven't initialized the flow yet
return (0, l, 0, 0, r, 0)
array = self.shm.array
index = array['index']
start = index[0]
end = index[-1]
lbar = max(l, start)
rbar = min(r, end)
return (
start, l, lbar, rbar, r, end,
)
def read(self) -> tuple[
int, int, np.ndarray,
int, int, np.ndarray,
]:
# read call
array = self.shm.array
indexes = array['index']
ifirst = indexes[0]
ilast = indexes[-1]
ifirst, l, lbar, rbar, r, ilast = self.datums_range()
# get read-relative indices adjusting
# for master shm index.
lbar_i = max(l, ifirst) - ifirst
rbar_i = min(r, ilast) - ifirst
# TODO: we could do it this way as well no?
# to_draw = array[lbar - ifirst:(rbar - ifirst) + 1]
in_view = array[lbar_i: rbar_i + 1]
return (
# abs indices + full data set
ifirst, ilast, array,
# relative indices + in view datums
lbar_i, rbar_i, in_view,
)
def update_graphics(
self,
use_vr: bool = True,
render: bool = True,
array_key: Optional[str] = None,
profiler: Optional[pg.debug.Profiler] = None,
**kwargs,
) -> pg.GraphicsObject:
'''
Read latest datums from shm and render to (incrementally)
render to graphics.
'''
profiler = profiler or pg.debug.Profiler(
msg=f'Flow.update_graphics() for {self.name}',
disabled=not pg_profile_enabled(),
gt=ms_slower_then,
delayed=True,
)
# shm read and slice to view
read = (
xfirst, xlast, array,
ivl, ivr, in_view,
) = self.read()
profiler('read src shm data')
if (
not in_view.size
or not render
):
return self.graphics
graphics = self.graphics
if isinstance(graphics, BarItems):
# if no source data renderer exists create one.
r = self._src_r
if not r:
# OHLC bars path renderer
r = self._src_r = Renderer(
flow=self,
# TODO: rename this to something with ohlc
draw_path=gen_qpath,
last_read=read,
)
ds_curve_r = Renderer(
flow=self,
# just swap in the flat view
# data_t=lambda array: self.gy.array,
last_read=read,
draw_path=partial(
rowarr_to_path,
x_basis=None,
),
)
curve = FastAppendCurve(
name='OHLC',
color=graphics._color,
)
curve.hide()
self.plot.addItem(curve)
# baseline "line" downsampled OHLC curve that should
# kick on only when we reach a certain uppx threshold.
self._render_table[0] = (
ds_curve_r,
curve,
)
dsc_r, curve = self._render_table[0]
# do checks for whether or not we require downsampling:
# - if we're **not** downsampling then we simply want to
# render the bars graphics curve and update..
# - if insteam we are in a downsamplig state then we to
x_gt = 6
uppx = curve.x_uppx()
in_line = should_line = curve.isVisible()
if (
should_line
and uppx < x_gt
):
print('FLIPPING TO BARS')
should_line = False
elif (
not should_line
and uppx >= x_gt
):
print('FLIPPING TO LINE')
should_line = True
profiler(f'ds logic complete line={should_line}')
# do graphics updates
if should_line:
fields = ['open', 'high', 'low', 'close']
if self.gy is None:
# create a flattened view onto the OHLC array
# which can be read as a line-style format
shm = self.shm
# flat = self.gy = self.shm.unstruct_view(fields)
self.gy = self.shm.ustruct(fields)
first = self._iflat_first = self.shm._first.value
last = self._iflat_last = self.shm._last.value
# write pushed data to flattened copy
self.gy[first:last] = rfn.structured_to_unstructured(
self.shm.array[fields]
)
# generate an flat-interpolated x-domain
self.gx = (
np.broadcast_to(
shm._array['index'][:, None],
(
shm._array.size,
# 4, # only ohlc
self.gy.shape[1],
),
) + np.array([-0.5, 0, 0, 0.5])
)
assert self.gy.any()
# print(f'unstruct diff: {time.time() - start}')
# profiler('read unstr view bars to line')
# start = self.gy._first.value
# update flatted ohlc copy
(
iflat_first,
iflat,
ishm_last,
ishm_first,
) = (
self._iflat_first,
self._iflat_last,
self.shm._last.value,
self.shm._first.value
)
# check for shm prepend updates since last read.
if iflat_first != ishm_first:
# write newly prepended data to flattened copy
self.gy[
ishm_first:iflat_first
] = rfn.structured_to_unstructured(
self.shm._array[fields][ishm_first:iflat_first]
)
self._iflat_first = ishm_first
# # flat = self.gy = self.shm.unstruct_view(fields)
# self.gy = self.shm.ustruct(fields)
# # self._iflat_last = self.shm._last.value
# # self._iflat_first = self.shm._first.value
# # do an update for the most recent prepend
# # index
# iflat = ishm_first
to_update = rfn.structured_to_unstructured(
self.shm._array[iflat:ishm_last][fields]
)
self.gy[iflat:ishm_last][:] = to_update
profiler('updated ustruct OHLC data')
# slice out up-to-last step contents
y_flat = self.gy[ishm_first:ishm_last]
x_flat = self.gx[ishm_first:ishm_last]
# update local last-index tracking
self._iflat_last = ishm_last
# reshape to 1d for graphics rendering
y = y_flat.reshape(-1)
x = x_flat.reshape(-1)
profiler('flattened ustruct OHLC data')
# do all the same for only in-view data
y_iv_flat = y_flat[ivl:ivr]
x_iv_flat = x_flat[ivl:ivr]
y_iv = y_iv_flat.reshape(-1)
x_iv = x_iv_flat.reshape(-1)
profiler('flattened ustruct in-view OHLC data')
# legacy full-recompute-everytime method
# x, y = ohlc_flatten(array)
# x_iv, y_iv = ohlc_flatten(in_view)
# profiler('flattened OHLC data')
curve.update_from_array(
x,
y,
x_iv=x_iv,
y_iv=y_iv,
view_range=(ivl, ivr), # hack
profiler=profiler,
# should_redraw=False,
# NOTE: already passed through by display loop?
# do_append=uppx < 16,
**kwargs,
)
curve.show()
profiler('updated ds curve')
else:
# render incremental or in-view update
# and apply ouput (path) to graphics.
path, last = r.render(
read,
only_in_view=True,
)
graphics.path = path
graphics.draw_last(last)
# NOTE: on appends we used to have to flip the coords
# cache thought it doesn't seem to be required any more?
# graphics.setCacheMode(QtWidgets.QGraphicsItem.NoCache)
# graphics.setCacheMode(QtWidgets.QGraphicsItem.DeviceCoordinateCache)
# graphics.prepareGeometryChange()
graphics.update()
if (
not in_line
and should_line
):
# change to line graphic
log.info(
f'downsampling to line graphic {self.name}'
)
graphics.hide()
# graphics.update()
curve.show()
curve.update()
elif in_line and not should_line:
log.info(f'showing bars graphic {self.name}')
curve.hide()
graphics.show()
graphics.update()
# update our pre-downsample-ready data and then pass that
# new data the downsampler algo for incremental update.
# graphics.update_from_array(
# array,
# in_view,
# view_range=(ivl, ivr) if use_vr else None,
# **kwargs,
# )
# generate and apply path to graphics obj
# graphics.path, last = r.render(
# read,
# only_in_view=True,
# )
# graphics.draw_last(last)
else:
# ``FastAppendCurve`` case:
array_key = array_key or self.name
uppx = graphics.x_uppx()
if graphics._step_mode and self.gy is None:
self._iflat_first = self.shm._first.value
# create a flattened view onto the OHLC array
# which can be read as a line-style format
shm = self.shm
# fields = ['index', array_key]
i = shm._array['index'].copy()
out = shm._array[array_key].copy()
self.gx = np.broadcast_to(
i[:, None],
(i.size, 2),
) + np.array([-0.5, 0.5])
# self.gy = np.broadcast_to(
# out[:, None], (out.size, 2),
# )
self.gy = np.empty((len(out), 2), dtype=out.dtype)
self.gy[:] = out[:, np.newaxis]
# start y at origin level
self.gy[0, 0] = 0
if graphics._step_mode:
(
iflat_first,
iflat,
ishm_last,
ishm_first,
) = (
self._iflat_first,
self._iflat_last,
self.shm._last.value,
self.shm._first.value
)
il = max(iflat - 1, 0)
# check for shm prepend updates since last read.
if iflat_first != ishm_first:
print(f'prepend {array_key}')
# i_prepend = self.shm._array['index'][
# ishm_first:iflat_first]
y_prepend = self.shm._array[array_key][
ishm_first:iflat_first
]
y2_prepend = np.broadcast_to(
y_prepend[:, None], (y_prepend.size, 2),
)
# write newly prepended data to flattened copy
self.gy[ishm_first:iflat_first] = y2_prepend
self._iflat_first = ishm_first
append_diff = ishm_last - iflat
if append_diff:
# slice up to the last datum since last index/append update
# new_x = self.shm._array[il:ishm_last]['index']
new_y = self.shm._array[il:ishm_last][array_key]
new_y2 = np.broadcast_to(
new_y[:, None], (new_y.size, 2),
)
self.gy[il:ishm_last] = new_y2
profiler('updated step curve data')
# print(
# f'append size: {append_diff}\n'
# f'new_x: {new_x}\n'
# f'new_y: {new_y}\n'
# f'new_y2: {new_y2}\n'
# f'new gy: {gy}\n'
# )
# update local last-index tracking
self._iflat_last = ishm_last
# slice out up-to-last step contents
x_step = self.gx[ishm_first:ishm_last+2]
# shape to 1d
x = x_step.reshape(-1)
y_step = self.gy[ishm_first:ishm_last+2]
lasts = self.shm.array[['index', array_key]]
last = lasts[array_key][-1]
y_step[-1] = last
# shape to 1d
y = y_step.reshape(-1)
# s = 6
# print(f'lasts: {x[-2*s:]}, {y[-2*s:]}')
profiler('sliced step data')
# do all the same for only in-view data
ys_iv = y_step[ivl:ivr+1]
xs_iv = x_step[ivl:ivr+1]
y_iv = ys_iv.reshape(ys_iv.size)
x_iv = xs_iv.reshape(xs_iv.size)
# print(
# f'ys_iv : {ys_iv[-s:]}\n'
# f'y_iv: {y_iv[-s:]}\n'
# f'xs_iv: {xs_iv[-s:]}\n'
# f'x_iv: {x_iv[-s:]}\n'
# )
profiler('flattened ustruct in-view OHLC data')
# legacy full-recompute-everytime method
# x, y = ohlc_flatten(array)
# x_iv, y_iv = ohlc_flatten(in_view)
# profiler('flattened OHLC data')
x_last = array['index'][-1]
y_last = array[array_key][-1]
graphics._last_line = QLineF(
x_last - 0.5, 0,
x_last + 0.5, 0,
)
graphics._last_step_rect = QRectF(
x_last - 0.5, 0,
x_last + 0.5, y_last,
)
# graphics.update()
graphics.update_from_array(
x=x,
y=y,
x_iv=x_iv,
y_iv=y_iv,
view_range=(ivl, ivr) if use_vr else None,
draw_last=False,
slice_to_head=-2,
should_redraw=bool(append_diff),
# NOTE: already passed through by display loop?
# do_append=uppx < 16,
**kwargs
)
# graphics.reset_cache()
# print(
# f"path br: {graphics.path.boundingRect()}\n",
# # f"fast path br: {graphics.fast_path.boundingRect()}",
# f"last rect br: {graphics._last_step_rect}\n",
# f"full br: {graphics._br}\n",
# )
else:
x = array['index']
y = array[array_key]
x_iv = in_view['index']
y_iv = in_view[array_key]
# graphics.draw_last(x, y)
profiler(f'draw last segment {array_key}')
graphics.update_from_array(
x=x,
y=y,
x_iv=x_iv,
y_iv=y_iv,
view_range=(ivl, ivr) if use_vr else None,
# NOTE: already passed through by display loop?
# do_append=uppx < 16,
**kwargs
)
return graphics
def xy_downsample(
x,
y,
px_width,
uppx,
x_spacer: float = 0.5,
) -> tuple[np.ndarray, np.ndarray]:
# downsample whenever more then 1 pixels per datum can be shown.
# always refresh data bounds until we get diffing
# working properly, see above..
bins, x, y = ds_m4(
x,
y,
px_width=px_width,
uppx=uppx,
log_scale=bool(uppx)
)
# flatten output to 1d arrays suitable for path-graphics generation.
x = np.broadcast_to(x[:, None], y.shape)
x = (x + np.array(
[-x_spacer, 0, 0, x_spacer]
)).flatten()
y = y.flatten()
return x, y
class Renderer(msgspec.Struct):
flow: Flow
# called to render path graphics
draw_path: Callable[np.ndarray, QPainterPath]
# called on input data but before any graphics format
# conversions or processing.
data_t: Optional[Callable[ShmArray, np.ndarray]] = None
data_t_shm: Optional[ShmArray] = None
# called on the final data (transform) output to convert
# to "graphical data form" a format that can be passed to
# the ``.draw()`` implementation.
graphics_t: Optional[Callable[ShmArray, np.ndarray]] = None
graphics_t_shm: Optional[ShmArray] = None
# path graphics update implementation methods
prepend_fn: Optional[Callable[QPainterPath, QPainterPath]] = None
append_fn: Optional[Callable[QPainterPath, QPainterPath]] = None
# last array view read
last_read: Optional[np.ndarray] = None
# output graphics rendering, the main object
# processed in ``QGraphicsObject.paint()``
path: Optional[QPainterPath] = None
# def diff(
# self,
# latest_read: tuple[np.ndarray],
# ) -> tuple[np.ndarray]:
# # blah blah blah
# # do diffing for prepend, append and last entry
# return (
# to_prepend
# to_append
# last,
# )
def render(
self,
new_read,
# only render datums "in view" of the ``ChartView``
only_in_view: bool = False,
) -> list[QPainterPath]:
'''
Render the current graphics path(s)
There are (at least) 3 stages from source data to graphics data:
- a data transform (which can be stored in additional shm)
- a graphics transform which converts discrete basis data to
a `float`-basis view-coords graphics basis. (eg. ``ohlc_flatten()``,
``step_path_arrays_from_1d()``, etc.)
- blah blah blah (from notes)
'''
# do full source data render to path
(
xfirst, xlast, array,
ivl, ivr, in_view,
) = self.last_read
if only_in_view:
array = in_view
# # get latest data from flow shm
# self.last_read = (
# xfirst, xlast, array, ivl, ivr, in_view
# ) = new_read
if self.path is None or only_in_view:
# redraw the entire source data if we have either of:
# - no prior path graphic rendered or,
# - we always intend to re-render the data only in view
# data transform: convert source data to a format
# expected to be incrementally updates and later rendered
# to a more graphics native format.
if self.data_t:
array = self.data_t(array)
# maybe allocate shm for data transform output
# if self.data_t_shm is None:
# fshm = self.flow.shm
# shm, opened = maybe_open_shm_array(
# f'{self.flow.name}_data_t',
# # TODO: create entry for each time frame
# dtype=array.dtype,
# readonly=False,
# )
# assert opened
# shm.push(array)
# self.data_t_shm = shm
elif self.path:
print(f'inremental update not supported yet {self.flow.name}')
# TODO: do incremental update
# prepend, append, last = self.diff(self.flow.read())
# do path generation for each segment
# and then push into graphics object.
hist, last = array[:-1], array[-1]
# call path render func on history
self.path = self.draw_path(hist)
self.last_read = new_read
return self.path, last