Rework overlay pin technique: "align to first"

As part of solving a final bullet-issue in #455, which is specifically
a case:
- with N > 2 curves, one of which is the "major" dispersion curve" and
  the others are "minors",
- we can run into a scenario where some minor curve which gets pinned to
  the major (due to the original "pinning technique" -> "align to
  major") at some `P(t)` which is *not* the major's minimum / maximum
  due to the minor having a smaller/shorter support and thus,
- requires that in order to show then max/min on the minor curve we have
  to expand the range of the major curve as well but,
- that also means any previously scaled (to the major) minor curves need
  to be adjusted as well or they'll not be pinned to the major the same
  way!

I originally was trying to avoid doing the recursive iteration back
through all previously scaled minor curves and instead decided to try
implementing the "per side" curve dispersion detection (as was
originally attempted when first starting this work). The idea is to
decide which curve's up or down "swing in % returns" would determine the
global y-range *on that side*. Turns out I stumbled on the "align to
first" technique in the process: "for each overlay curve we align its
earliest sample (in time) to the same level of the earliest such sample
for whatever is deemed the major (directionally disperse) curve in
view".

I decided (with help) that this "pin to first" approach/style is equally
as useful and maybe often more so when wanting to view support-disjoint
time series:

- instead of compressing the y-range on "longer series which have lesser
  sigma" to make whatever "shorter but larger-sigma series" pin to it at
  an intersect time step, this instead will expand the price ranges
  based on the earliest time step in each series.
- the output global-returns-overlay-range for any N-set of series is equal to
  the same in the previous "pin to intersect time" technique.
- the only time this technique seems less useful is for overlaying
  market feeds which have the same destination asset but different
  source assets (eg. btceur and btcusd on the same chart since if one
  of the series is shorter it will always be aligned to the earliest
  datum on the longer instead of more naturally to the intersect sample
  level as was in the previous approach).

As such I'm going to keep this technique as discovered and will later
add back optional support for the "align to intersect" approach from
previous (which will again require detecting the highest dispersion
curve direction-agnostic) and pin all minors to the price level at which
they start on the major.

Further details of the implementation rework in
`.interact_graphics_cycle()` include:

- add `intersect_from_longer()` to detect and deliver a common datum
  from 2 series which are different in length: the first time-index
  sample in the longer.
- Rewrite the drafted `OverlayT` to only compute (inversed log-returns)
  transforms for a single direction and use 2 instances, one for each
  direction inside the `Viz`-overlay iteration loop.
- do all dispersion-per-side major curve detection in the first pass of
  all `Viz`s on a plot, instead updating the `OverlayT` instances for
  each side and compensating for any length mismatch and
  rescale-to-minor cases in each loop cycle.
log_linearized_curve_overlays
Tyler Goodlet 2023-02-16 15:23:56 -05:00
parent d5ba26cfaf
commit 5f470d6122
1 changed files with 298 additions and 243 deletions

View File

@ -155,11 +155,11 @@ async def handle_viewmode_kb_inputs(
}
):
import tractor
god = order_mode.godw
feed = order_mode.feed
chart = order_mode.chart
vlm_chart = chart.linked.subplots['volume']
dvlm_pi = vlm_chart._vizs['dolla_vlm'].plot
god = order_mode.godw # noqa
feed = order_mode.feed # noqa
chart = order_mode.chart # noqa
vlm_chart = chart.linked.subplots['volume'] # noqa
dvlm_pi = vlm_chart._vizs['dolla_vlm'].plot # noqa
await tractor.breakpoint()
# SEARCH MODE #
@ -360,49 +360,6 @@ async def handle_viewmode_mouse(
view.order_mode.submit_order()
class OverlayT(Struct):
'''
An overlay co-domain range transformer.
Used to translate and apply a range from one y-range
to another based on a returns logarithm:
R(ymn, ymx, yref) = (ymx - yref)/yref
which gives the log-scale multiplier, and
ymx_t = yref * (1 + R)
which gives the inverse to translate to the same value
in the target co-domain.
'''
viz: Viz # viz with largest measured dispersion
mx: float = 0
mn: float = float('inf')
up_swing: float = 0
down_swing: float = 0
disp: float = 0
def loglin_from_range(
self,
y_ref: float, # reference value for dispersion metric
mn: float, # min y in target log-lin range
mx: float, # max y in target log-lin range
offset: float, # y-offset to start log-scaling from
) -> tuple[float, float]:
r_up = (mx - y_ref) / y_ref
r_down = (mn - y_ref) / y_ref
ymn = offset * (1 + r_down)
ymx = offset * (1 + r_up)
return ymn, ymx
class ChartView(ViewBox):
'''
Price chart view box with interaction behaviors you'd expect from
@ -1048,7 +1005,6 @@ class ChartView(ViewBox):
np.ndarray, # in-view array
],
] = {}
major_in_view: np.ndarray = None
# ONLY auto-yrange the viz mapped to THIS view box
if not do_overlay_scaling:
@ -1072,12 +1028,22 @@ class ChartView(ViewBox):
# don't iterate overlays, just move to next chart
continue
for name, viz in chart._vizs.items():
# create a group overlay log-linearized y-range transform to
# track and eventually inverse transform all overlay curves
# to a common target max dispersion range.
dnt = OverlayT()
upt = OverlayT()
if debug_print:
print(
f'UX GRAPHICS CYCLE: {viz.name}@{chart_name}'
)
if debug_print:
print(
f'BEGIN UX GRAPHICS CYCLE: @{chart_name}\n'
+
'#'*100
+
'\n'
)
for name, viz in chart._vizs.items():
out = _maybe_calc_yrange(
viz,
@ -1119,7 +1085,6 @@ class ChartView(ViewBox):
# charts besides OHLC?
else:
ymn, ymx = yrange
# print(f'adding {viz.name} to overlay')
# determine start datum in view
arr = viz.shm.array
@ -1128,36 +1093,169 @@ class ChartView(ViewBox):
log.warning(f'{viz.name} not in view?')
continue
row_start = arr[read_slc.start - 1]
# row_start = arr[read_slc.start - 1]
row_start = arr[read_slc.start]
if viz.is_ohlc:
y_start = row_start['open']
y_ref = row_start['open']
else:
y_start = row_start[viz.name]
y_ref = row_start[viz.name]
profiler(f'{viz.name}@{chart_name} MINOR curve median')
overlay_table[viz.plot.vb] = (
viz,
y_start,
y_ref,
ymn,
ymx,
read_slc,
in_view,
)
# find curve with max dispersion
disp = abs(ymx - ymn) / y_start
key = 'open' if viz.is_ohlc else viz.name
start_t = in_view[0]['time']
r_down = (ymn - y_ref) / y_ref
r_up = (ymx - y_ref) / y_ref
msg = (
f'### {viz.name}@{chart_name} ###\n'
f'y_ref: {y_ref}\n'
f'down disp: {r_down}\n'
f'up disp: {r_up}\n'
)
profiler(msg)
if debug_print:
print(msg)
# track the "major" curve as the curve with most
# dispersion.
if (
dnt.rng is None
or (
r_down < dnt.rng
and r_down < 0
)
):
dnt.viz = viz
dnt.rng = r_down
dnt.in_view = in_view
dnt.start_t = in_view[0]['time']
major_mn = ymn
msg = f'NEW DOWN: {viz.name}@{chart_name} r:{r_down}\n'
profiler(msg)
if debug_print:
print(msg)
else:
# minor in the down swing range so check that if
# we apply the current rng to the minor that it
# doesn't go outside the current range for the major
# otherwise we recompute the minor's range (when
# adjusted for it's intersect point to be the new
# major's range.
intersect = intersect_from_longer(
dnt.start_t,
dnt.in_view,
start_t,
in_view,
)
profiler(f'{viz.name}@{chart_name} intersect by t')
if intersect:
longer_in_view, _t, i = intersect
scaled_mn = dnt.apply_rng(y_ref)
if scaled_mn > ymn:
# after major curve scaling we detected
# the minor curve is still out of range
# so we need to adjust the major's range
# to include the new composed range.
y_maj_ref = longer_in_view[key]
new_major_ymn = (
y_maj_ref
*
(1 + r_down)
)
# rewrite the major range to the new
# minor-pinned-to-major range and mark
# the transform as "virtual".
msg = (
f'EXPAND DOWN bc {viz.name}@{chart_name}\n'
f'y_start epoch time @ {_t}:\n'
f'y_maj_ref @ {_t}: {y_maj_ref}\n'
f'R: {dnt.rng} -> {r_down}\n'
f'MN: {major_mn} -> {new_major_ymn}\n'
)
dnt.rng = r_down
major_mn = dnt.y_val = new_major_ymn
profiler(msg)
if debug_print:
print(msg)
if (
upt.rng is None
or (
r_up > upt.rng
and r_up > 0
)
):
upt.rng = r_up
upt.viz = viz
upt.in_view = in_view
upt.start_t = in_view[0]['time']
major_mx = ymx
msg = f'NEW UP: {viz.name}@{chart_name} r:{r_up}\n'
profiler(msg)
if debug_print:
print(msg)
else:
intersect = intersect_from_longer(
upt.start_t,
upt.in_view,
start_t,
in_view,
)
profiler(f'{viz.name}@{chart_name} intersect by t')
if intersect:
longer_in_view, _t, i = intersect
scaled_mx = upt.apply_rng(y_ref)
if scaled_mx < ymx:
# after major curve scaling we detected
# the minor curve is still out of range
# so we need to adjust the major's range
# to include the new composed range.
y_maj_ref = longer_in_view[key]
new_major_ymx = (
y_maj_ref
*
(1 + r_up)
)
# rewrite the major range to the new
# minor-pinned-to-major range and mark
# the transform as "virtual".
msg = (
f'EXPAND UP bc {viz.name}@{chart_name}:\n'
f'y_maj_ref @ {_t}: {y_maj_ref}\n'
f'R: {upt.rng} -> {r_up}\n'
f'MX: {major_mx} -> {new_major_ymx}\n'
)
upt.rng = r_up
major_mx = upt.y_val = new_major_ymx
profiler(msg)
print(msg)
# find curve with max dispersion
disp = abs(ymx - ymn) / y_ref
if disp > mx_disp:
major_viz = viz
mx_disp = disp
major_mn = ymn
major_mx = ymx
major_in_view = in_view
profiler(f'{viz.name}@{chart_name} set new major')
profiler(f'{viz.name}@{chart_name} MINOR curve scale')
@ -1203,6 +1301,15 @@ class ChartView(ViewBox):
profiler(f'<{chart_name}>.interact_graphics_cycle({name})')
# if a minor curves scaling brings it "outside" the range of
# the major curve (in major curve co-domain terms) then we
# need to rescale the major to also include this range. The
# below placeholder denotes when this occurs.
# group_mxmn: None | tuple[float, float] = None
# TODO: probably re-write this loop as a compiled cpython or
# numba func.
# conduct "log-linearized multi-plot" scalings for all groups
for (
view,
@ -1216,169 +1323,7 @@ class ChartView(ViewBox):
)
) in overlay_table.items():
# we use the ymn/mx verbatim from the major curve
# (i.e. the curve measured to have the highest
# dispersion in view).
if viz is major_viz:
ymn = y_min
ymx = y_max
continue
else:
key = 'open' if viz.is_ohlc else viz.name
# handle case where major and minor curve(s) have
# a disjoint x-domain (one curve is smaller in
# length then the other):
# - find the highest (time) index common to both
# curves.
# - slice out the first "intersecting" y-value from
# both curves for use in log-linear scaling such
# that the intersecting y-value is used as the
# reference point for scaling minor curve's
# y-range based on the major curves y-range.
# get intersection point y-values for both curves
minor_in_view_start = minor_in_view[0]
minor_i_start = minor_in_view_start['index']
minor_i_start_t = minor_in_view_start['time']
major_in_view_start = major_in_view[0]
major_i_start = major_in_view_start['index']
major_i_start_t = major_in_view_start['time']
y_major_intersect = major_in_view_start[key]
y_minor_intersect = minor_in_view_start[key]
profiler(f'{viz.name}@{chart_name} intersect detection')
tdiff = (major_i_start_t - minor_i_start_t)
if debug_print:
print(
f'{major_viz.name} time diff with minor:\n'
f'maj:{major_i_start_t}\n'
'-\n'
f'min:{minor_i_start_t}\n'
f'=> {tdiff}\n'
)
# major has later timestamp adjust minor
if tdiff > 0:
slc = slice_from_time(
arr=minor_in_view,
start_t=major_i_start_t,
stop_t=major_i_start_t,
)
y_minor_intersect = minor_in_view[slc.start][key]
profiler(f'{viz.name}@{chart_name} intersect by t')
# minor has later timestamp adjust major
elif tdiff < 0:
slc = slice_from_time(
arr=major_in_view,
start_t=minor_i_start_t,
stop_t=minor_i_start_t,
)
y_major_intersect = major_in_view[slc.start][key]
profiler(f'{viz.name}@{chart_name} intersect by t')
if debug_print:
print(
f'major_i_start: {major_i_start}\n'
f'major_i_start_t: {major_i_start_t}\n'
f'minor_i_start: {minor_i_start}\n'
f'minor_i_start_t: {minor_i_start_t}\n'
)
# TODO: probably write this as a compile cpython or
# numba func.
# compute directional (up/down) y-range
# % swing/dispersion starting at the reference index
# determined by the above indexing arithmetic.
y_ref = y_major_intersect
if not y_ref:
log.warning(
f'BAD y_major_intersect?!: {y_major_intersect}'
)
# breakpoint()
r_up = (major_mx - y_ref) / y_ref
r_down = (major_mn - y_ref) / y_ref
minor_y_start = y_minor_intersect
ymn = minor_y_start * (1 + r_down)
ymx = minor_y_start * (1 + r_up)
profiler(f'{viz.name}@{chart_name} SCALE minor')
# XXX: handle out of view cases where minor curve
# now is outside the range of the major curve. in
# this case we then re-scale the major curve to
# include the range missing now enforced by the
# minor (now new major for this *side*). Note this
# is side (up/down) specific.
new_maj_mxmn: None | tuple[float, float] = None
if y_max > ymx:
y_ref = y_minor_intersect
r_up_minor = (y_max - y_ref) / y_ref
y_maj_ref = y_major_intersect
new_maj_ymx = y_maj_ref * (1 + r_up_minor)
new_maj_mxmn = (major_mn, new_maj_ymx)
if debug_print:
print(
f'{view.name} OUT OF RANGE:\n'
'--------------------\n'
f'y_max:{y_max} > ymx:{ymx}\n'
)
ymx = y_max
profiler(f'{viz.name}@{chart_name} re-SCALE major UP')
if y_min < ymn:
y_ref = y_minor_intersect
r_down_minor = (y_min - y_ref) / y_ref
y_maj_ref = y_major_intersect
new_maj_ymn = y_maj_ref * (1 + r_down_minor)
new_maj_mxmn = (
new_maj_ymn,
new_maj_mxmn[1] if new_maj_mxmn else major_mx
)
if debug_print:
print(
f'{view.name} OUT OF RANGE:\n'
'--------------------\n'
f'y_min:{y_min} < ymn:{ymn}\n'
)
ymn = y_min
profiler(
f'{viz.name}@{chart_name} re-SCALE major DOWN'
)
if new_maj_mxmn:
if debug_print:
print(
f'RESCALE MAJOR {major_viz.name}:\n'
f'previous: {(major_mn, major_mx)}\n'
f'new: {new_maj_mxmn}\n'
)
major_mn, major_mx = new_maj_mxmn
if debug_print:
print(
f'{view.name} APPLY group mxmn\n'
'--------------------\n'
f'y_minor_intersect: {y_minor_intersect}\n'
f'y_major_intersect: {y_major_intersect}\n'
f'scaled ymn: {ymn}\n'
f'scaled ymx: {ymx}\n'
f'scaled mx_disp: {mx_disp}\n'
)
key = 'open' if viz.is_ohlc else viz.name
if (
isinf(ymx)
@ -1389,32 +1334,47 @@ class ChartView(ViewBox):
)
continue
ymn = dnt.apply_rng(y_start)
ymx = upt.apply_rng(y_start)
# NOTE XXX: we have to set each curve's range once (and
# ONLY ONCE) here since we're doing this entire routine
# inside of a single render cycle (and apparently calling
# `ViewBox.setYRange()` multiple times within one only takes
# the first call as serious...) XD
view._set_yrange(
yrange=(ymn, ymx),
)
profiler(f'{viz.name}@{chart_name} log-SCALE minor')
# NOTE XXX: we have to set the major curve's range once (and
# only once) here since we're doing this entire routine
# inside of a single render cycle (and apparently calling
# `ViewBox.setYRange()` multiple times within one only takes
# the first call as serious...) XD
if debug_print:
print(
f'Scale MAJOR {major_viz.name}:\n'
f'scaled mx_disp: {mx_disp}\n'
f'previous: {(major_mn, major_mx)}\n'
f'new: {new_maj_mxmn}\n'
)
major_viz.plot.vb._set_yrange(
yrange=(major_mn, major_mx),
)
profiler(f'{viz.name}@{chart_name} log-SCALE major')
# major_mx, major_mn = new_maj_mxmn
if debug_print:
print(
'------------------------------\n'
f'LOGLIN SCALE CYCLE: {viz.name}@{chart_name}\n'
f'UP MAJOR C: {upt.viz.name} with disp: {upt.rng}\n'
f'DOWN MAJOR C: {dnt.viz.name} with disp: {dnt.rng}\n'
f'y_start: {y_start}\n'
f'y min: {y_min}\n'
f'y max: {y_max}\n'
f'T scaled ymn: {ymn}\n'
f'T scaled ymx: {ymx}\n'
'------------------------------\n'
)
# profiler(f'{viz.name}@{chart_name} log-SCALE major')
# major_mx, major_mn = group_mxmn
# vrs = major_viz.plot.vb.viewRange()
# if vrs[1][0] > major_mn:
# breakpoint()
if debug_print:
print(
f'END UX GRAPHICS CYCLE: @{chart_name}\n'
+
'#'*100
+
'\n'
)
if not do_linked_charts:
return
@ -1466,3 +1426,98 @@ def _maybe_calc_yrange(
read_slc,
yrange_kwargs,
)
class OverlayT(Struct):
'''
An overlay co-domain range transformer.
Used to translate and apply a range from one y-range
to another based on a returns logarithm:
R(ymn, ymx, yref) = (ymx - yref)/yref
which gives the log-scale multiplier, and
ymx_t = yref * (1 + R)
which gives the inverse to translate to the same value
in the target co-domain.
'''
start_t: float | None = None
viz: Viz = None
# % "range" computed from some ref value to the mn/mx
rng: float | None = None
in_view: np.ndarray | None = None
# pinned-minor curve modified mn and max for the major dispersion
# curve due to one series being shorter and the pin + scaling from
# that pin point causing the original range to have to increase.
y_val: float | None = None
def apply_rng(
self,
y_start: float, # reference value for dispersion metric
) -> float:
return y_start * (1 + self.rng)
# def loglin_from_range(
# self,
# y_ref: float, # reference value for dispersion metric
# mn: float, # min y in target log-lin range
# mx: float, # max y in target log-lin range
# offset: float, # y-offset to start log-scaling from
# ) -> tuple[float, float]:
# r_up = (mx - y_ref) / y_ref
# r_down = (mn - y_ref) / y_ref
# ymn = offset * (1 + r_down)
# ymx = offset * (1 + r_up)
# return ymn, ymx
def intersect_from_longer(
start_t_first: float,
in_view_first: np.ndarray,
start_t_second: float,
in_view_second: np.ndarray,
) -> np.ndarray:
tdiff = start_t_first - start_t_second
if tdiff == 0:
return False
i: int = 0
# first time series has an "earlier" first time stamp then the 2nd.
# aka 1st is "shorter" then the 2nd.
if tdiff > 0:
longer = in_view_second
find_t = start_t_first
i = 1
# second time series has an "earlier" first time stamp then the 1st.
# aka 2nd is "shorter" then the 1st.
elif tdiff < 0:
longer = in_view_first
find_t = start_t_second
i = 0
slc = slice_from_time(
arr=longer,
start_t=find_t,
stop_t=find_t,
)
return (
longer[slc.start],
find_t,
i,
)