Spaces:
Running
Running
File size: 9,354 Bytes
c9ef990 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 | """Plotly figures for the TiRex-2 demo, with a consistent brand palette."""
from __future__ import annotations
import numpy as np
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
from plotly.subplots import make_subplots
import plotly.colors as pc
from tirex2.plotting import (
COVARIATE_COLORS,
plot_covariate,
plot_forecast,
)
from core import ForecastResult
COLOR_PALETTE = px.colors.qualitative.G10
# Brand-ish, colour-blind-friendly palette (shared with the covariate example).
C_HISTORY = COLOR_PALETTE[0] # navy - observed history
C_TRUTH = COLOR_PALETTE[0] # black - ground truth (dashed)
C_MEDIAN = COLOR_PALETTE[1] # orange - forecast median
C_MULTI = COLOR_PALETTE[2] # green - multivariate forecast (covariate lab)
PLOTLY_TEMPLATE = "plotly_white"
# colour-blind-friendly, skipping first three which are similar to history and forecast
C_COVARIATES = COLOR_PALETTE[3:]
# Shared typography / chrome so charts match the app's Inter-based styling.
FONT = dict(family="Inter, -apple-system, BlinkMacSystemFont, sans-serif",
color="#16202e", size=11)
C_GRID = "#eef2f7"
def increase_color_brightness(hex_color: str, factor: float) -> str:
"""Increase brightness of a hex color by a given factor."""
rgb = pc.hex_to_rgb(hex_color)
brighter_rgb = pc.find_intermediate_color(rgb, pc.hex_to_rgb("#FFFFFF"), factor)
return pc.label_rgb(brighter_rgb)
def hex_to_rgba(hex_color: str, alpha: float) -> str:
r, g, b = pc.hex_to_rgb(hex_color)
return f"rgba({r}, {g}, {b}, {alpha})"
def _polish_axes(fig) -> None:
"""Lighten gridlines and drop the heavy axis chrome for a cleaner look."""
fig.update_xaxes(showgrid=True, gridcolor=C_GRID, zeroline=False,
showline=False, ticks="", title_font=dict(size=12, color="#8a93a3"))
fig.update_yaxes(showgrid=True, gridcolor=C_GRID, zeroline=False,
showline=False, ticks="", title_font=dict(size=12, color="#8a93a3"))
def _iter_covariate_rows(result: ForecastResult, x):
"""Yield ``(label, x, y)`` per covariate, aligned to the context origin.
The full covariate history is returned (never trimmed) so no data is dropped; the
view is narrowed later purely via the shared x-axis range. Past covariates carry
only history, so their x-values stop at the forecast start; future covariates run
through the horizon.
"""
ts = result.timeseries
if result.cov_mode == "past":
arrays = ts.past_covariates if ts is not None else None
elif result.cov_mode == "future":
arrays = ts.future_covariates if ts is not None else None
else:
arrays = None
if arrays is None:
return
labels = result.cov_names or []
x = np.asarray(x)
for i, cov in enumerate(np.asarray(arrays, dtype=np.float32)):
xc = x[:len(cov)]
yc = cov[:len(xc)]
label = labels[i] if i < len(labels) else f"Covariate {i + 1}"
yield label, xc, yc
def _as_positions(x):
"""Return integer plot positions plus optional date labels.
The tirex2 plotting primitives crash on a datetime x-axis when a series is absent
(they compare a ``Timestamp`` against ``np.inf``). Feeding them plain positions and
relabelling the axis with dates sidesteps that while keeping a readable time axis.
"""
x = np.asarray(x)
is_datetime = x.dtype.kind == "M" or (
x.dtype == object and len(x) and isinstance(x[0], pd.Timestamp)
)
if is_datetime:
return np.arange(len(x)), pd.to_datetime(x)
return x, None
def _apply_date_ticks(fig, positions, labels) -> None:
"""Relabel the (numeric) x-axis with ~8 formatted date ticks."""
n = min(8, len(labels))
if n < 2:
return
idx = np.unique(np.linspace(0, len(labels) - 1, n).astype(int))
span = labels[-1] - labels[0]
if span <= pd.Timedelta(days=3):
fmt = "%Y-%m-%d %H:%M"
elif span <= pd.Timedelta(days=1200):
fmt = "%Y-%m-%d"
else:
fmt = "%Y-%m"
fig.update_xaxes(
tickmode="array",
tickvals=[positions[i] for i in idx],
ticktext=[labels[i].strftime(fmt) for i in idx],
)
def build_forecast_figure(
result: ForecastResult,
baseline_result: ForecastResult | None,
x,
*,
max_context_to_show: int,
ground_truth=None,
):
"""Assemble the forecast figure and return ``(fig, n_rows)``.
With covariates, the target is drawn as two stacked, directly comparable panels -
a univariate TiRex baseline and the covariate-informed forecast - followed by one
panel per covariate (mirrors ``tirex2.demo.plot_demo_forecast``). Without covariates
it draws a single target panel.
In every case the *full* context and covariate history is plotted; ``max_context_to_show``
only narrows the initial visible window by setting a shared x-axis range (zoom), so no
data is cut off - the viewer can pan/zoom out to reveal the entire history.
"""
quantile_levels = tuple(result.quantile_levels)
context = np.asarray(result.context[0], dtype=np.float32)
context_len = len(context)
positions, date_labels = _as_positions(x)
if baseline_result is None:
fig = make_subplots(rows=1, cols=1)
plot_forecast(
context=context,
forecasts=result.quantiles[0],
ground_truth=ground_truth,
x=positions,
quantile_levels=quantile_levels,
max_context_to_show=max_context_to_show,
engine="plotly",
fig=fig,
row=1,
col=1,
)
n_rows = 1
else:
cov_rows = list(_iter_covariate_rows(result, positions))
n_cov = len(cov_rows)
cov_heights = [0.32 / n_cov] * n_cov if n_cov else []
fig = make_subplots(
rows=2 + n_cov,
cols=1,
shared_xaxes=True,
vertical_spacing=0.06,
row_heights=[0.34, 0.34, *cov_heights],
row_titles=["Univariate", "Multivariate",
*(lbl for lbl, _, _ in cov_rows)],
)
for row, forecast in ((1, baseline_result.quantiles[0]), (2, result.quantiles[0])):
plot_forecast(
context=context,
forecasts=forecast,
ground_truth=ground_truth,
x=positions,
quantile_levels=quantile_levels,
max_context_to_show=max_context_to_show,
engine="plotly",
fig=fig,
row=row,
col=1,
)
for i, (label, xc, yc) in enumerate(cov_rows):
plot_covariate(
yc, x=xc, label=label, engine="plotly", fig=fig, row=i + 3, col=1,
color=COVARIATE_COLORS[i % len(COVARIATE_COLORS)],
)
n_rows = 2 + n_cov
# Enforce the zoom window as a shared axis range on *every* row (target and covariate
# panels alike) without dropping any data. This is what keeps context and covariates
# from being cut off: all points remain plotted, only the initial view is narrowed.
start = max(0, context_len - max_context_to_show) if max_context_to_show else 0
fig.update_xaxes(range=[positions[start], positions[-1]], autorange=False)
if date_labels is not None:
_apply_date_ticks(fig, positions, date_labels)
return fig, n_rows
def build_dataset_figure(x, y, *, label: str):
"""Plot a single selected series over time - a dataset preview with no forecast.
Shown as soon as a dataset/target is chosen, before (and regardless of) any run,
so users can eyeball the raw series. Uses the same datetime-axis handling and brand
chrome as the forecast figure for a consistent look.
"""
positions, date_labels = _as_positions(x)
y = np.asarray(y, dtype=np.float32)
positions = positions[: len(y)]
fig = go.Figure()
fig.add_trace(go.Scatter(
x=positions, y=y[: len(positions)], mode="lines", name=label,
line=dict(color=C_HISTORY, width=1.6),
))
if date_labels is not None:
_apply_date_ticks(fig, positions, date_labels)
fig.update_layout(
template=PLOTLY_TEMPLATE, title="", hovermode="x unified", font=FONT,
height=360, paper_bgcolor="rgba(0,0,0,0)", plot_bgcolor="rgba(0,0,0,0)",
margin=dict(t=48, b=40, l=54, r=20),
legend=dict(orientation="h", yanchor="bottom", y=1.03, xanchor="left", x=0),
)
_polish_axes(fig)
fig.update_xaxes(title_text="time")
return fig
def style_forecast_figure(fig, n_rows: int) -> None:
"""Apply the shared brand template, legend, and per-row axis chrome in place."""
fig.update_layout(
template=PLOTLY_TEMPLATE,
title="",
hovermode="x unified",
font=FONT,
height=300 + 150 * (n_rows - 1),
paper_bgcolor="rgba(0,0,0,0)",
plot_bgcolor="rgba(0,0,0,0)",
margin=dict(t=72, b=40, l=54, r=20),
legend=dict(orientation="h", yanchor="bottom", y=1.03, xanchor="left", x=0),
)
_polish_axes(fig)
for row in range(1, n_rows):
fig.update_xaxes(showticklabels=False, title_text="", row=row, col=1)
fig.update_xaxes(showticklabels=True, title_text="time", row=n_rows, col=1)
|