ppak10's picture
Updates figures
8692359
Raw
History Blame Contribute Delete
9.68 kB
"""Shared helpers for scripts/plots/*.py.
House style is ported from the AdditiveLLM2-OA figures
(ppak10/AdditiveLLM2-OA, figures/*/*.py): DM Sans typeface, a curated
saturated palette anchored on #2D6A9F / #2AAA8A / #D44000 / #8B5CF6 with
#F97415 orange reserved for the reference/highlight series, a *framed*
(not despined) look with heavy spines and inward ticks, a light dashed
grid, and dual PNG@1200 + PDF export. Call `apply_house_style()` once at
import (done here), `style_axes(ax)` per Axes, and `save_figure(fig, stem)`
to write both formats.
"""
import json
from pathlib import Path
import matplotlib
import matplotlib.pyplot as plt
import matplotlib.colors as mcolors
import matplotlib.font_manager as fm
matplotlib.use("Agg")
ROOT = Path(__file__).parent.parent.parent
DATA_DIR = ROOT / "data"
OUT_DIR = ROOT / "assets"
FONT_DIR = Path(__file__).parent / "fonts"
# ── House style ───────────────────────────────────────────────────────────────
# Publication export DPI, matching the AdditiveLLM2 figures. save_figure()
# emits both a PNG at this DPI and a vector PDF.
EXPORT_DPI = 1200
# Reference categorical palette (AdditiveLLM2 figures/*/*.py). Used verbatim on
# the low-series-count figures (PLA/PETG controls, Type IV, Nylon 12 White).
REF_BLUE = "#2D6A9F"
REF_TEAL = "#2AAA8A"
REF_REDORANGE = "#D44000"
REF_PURPLE = "#8B5CF6"
# The signature accent — reserved for the reference / highlighted series
# (here the FormLabs PA12GF benchtop control the SLS batches are compared
# against), exactly as #F97415 flags the "Overall"/"best" series in the
# AdditiveLLM2 charts. Never assigned to an SLS batch.
ACCENT = "#F97415"
# Neutral fallback for any control material without a dedicated color.
CONTROL_COLOR = "#6B7280"
def apply_house_style() -> None:
"""Register DM Sans and set the AdditiveLLM2 rcParams. Idempotent."""
for ttf in sorted(FONT_DIR.glob("*.ttf")):
fm.fontManager.addfont(str(ttf))
plt.rcParams.update({
"font.family": "DM Sans",
"axes.linewidth": 1.4, # heavy framed spines
"axes.titlesize": 13,
"axes.titleweight": "bold",
"axes.labelsize": 12,
"xtick.labelsize": 10,
"ytick.labelsize": 10,
"xtick.direction": "in", # inward ticks
"ytick.direction": "in",
"xtick.major.size": 4,
"ytick.major.size": 4,
"xtick.major.width": 1.2,
"ytick.major.width": 1.2,
"legend.fontsize": 10,
"legend.frameon": True,
"legend.framealpha": 0.95,
"legend.edgecolor": "#D1D5DB",
"grid.linestyle": "--",
"grid.linewidth": 1.0,
"grid.alpha": 0.4,
"grid.color": "#B0B0B0",
"savefig.dpi": EXPORT_DPI,
})
def style_axes(ax) -> None:
"""Apply the framed look to one Axes: light dashed grid behind the data,
origin anchored at zero. Spines/ticks come from rcParams."""
ax.grid(True, zorder=0)
ax.set_axisbelow(True)
ax.set_xlim(left=0)
ax.set_ylim(bottom=0)
def save_figure(fig, out_stem: Path) -> Path:
"""Write `out_stem.png` (dpi=EXPORT_DPI) and `out_stem.pdf`, matching the
AdditiveLLM2 dual-format export. Returns the PNG path."""
out_stem.parent.mkdir(parents=True, exist_ok=True)
png = out_stem.with_suffix(".png")
fig.savefig(png, dpi=EXPORT_DPI, bbox_inches="tight", pad_inches=0.15)
fig.savefig(out_stem.with_suffix(".pdf"), bbox_inches="tight", pad_inches=0.15)
return png
apply_house_style()
# ── Batch color ramp ──────────────────────────────────────────────────────────
# Ordered list of every batch label, in print chronology. J_MB (media-blasted
# variant of print J) sits right after J so the two share a neighborhood on the
# ramp — encoding their shared print origin — while staying distinct.
ORDERED_BATCHES = ["A", "B", "C", "D", "E", "F", "G", "H", "I",
"J", "J_MB", "K", "L", "M", "N"]
# The 15 batches are chronological, so their color is an *ordered ramp* rather
# than an arbitrary categorical cycle: a warm gold→orange→brown sweep built
# around the signature #F97415 orange (ACCENT) — the whole batch palette is
# "based off that shade of orange" per user instruction. Adjacent batches read
# as neighbors — intended, since batch order is time order — and the legend +
# curve position disambiguate within a figure. The FormLabs reference series is
# deliberately *not* orange (see FORMLABS_COLOR) so it stands apart from the
# batches it's benchmarked against.
_RAMP = mcolors.LinearSegmentedColormap.from_list(
"batch_ramp", ["#F7C948", "#F9931E", ACCENT, "#C7430C", "#6E2206"])
def _build_batch_colors() -> dict[str, str]:
n = len(ORDERED_BATCHES)
return {batch: mcolors.to_hex(_RAMP(i / (n - 1)))
for i, batch in enumerate(ORDERED_BATCHES)}
# Kept consistent across figures so e.g. batch C is the same color everywhere.
BATCH_COLORS = _build_batch_colors()
# The FormLabs PA12GF benchtop reference is drawn in a contrasting blue rather
# than the orange batch family, so on the cluster / batch-average figures it
# reads clearly as the external benchmark the SLS batches are compared against.
FORMLABS_COLOR = REF_BLUE
# Non-SLS materials get explicit reference-palette colors (rather than a shared
# gray) so they read as intentional on their own figures.
MATERIAL_COLORS = {
"PA12GF_FL": FORMLABS_COLOR, # FormLabs PA12 GF reference → contrasting blue
"NYLON12_WHITE_FL": REF_PURPLE, # FormLabs Nylon 12 White reference
"PLA": REF_BLUE,
"PETG": REF_TEAL,
}
MATERIAL_STYLES = {"SLS": "-", "PLA": "--", "PETG": ":"}
# Non-default ASTM specimen types get their own linestyle so e.g. Batch M's
# Type IV (narrow-section) tensile specimens are visually tagged apart from
# its Type I specimens without needing a separate batch letter or color.
TYPE_LINESTYLES = {"Type IV": "--"}
FILAMENT_CONTROLS = {"PLA", "PETG"}
# FormLabs SLS reference-material controls that get their own dedicated
# figures instead of joining the main SLS composite/batch-averages plots —
# see scripts/plots/01_composite.py's module docstring.
NYLON_CONTROLS = {"NYLON12_WHITE_FL"}
# D638 (tensile) materials whose raw curve continues past the stress peak as
# a near-straight diagonal decline back toward zero — the crosshead keeps
# extending after the specimen separates while load reads ~0, and with few
# points sampled through the break itself this draws as a misleading
# diagonal rather than the near-vertical drop a real break shows (as seen in
# the other SLS batches, whose analyzed curves have many points through the
# break). Per user instruction, these curves are cut at their stress peak
# and given a synthetic vertical drop to zero at that same strain, matching
# the other batches' visual convention. D790 rows aren't affected: their
# break isn't a full separation the same way, and their curves don't show
# this artifact.
VERTICAL_BREAK_MATERIALS = {"NYLON12_WHITE_FL"}
def vertical_break_at_peak(strain: list[float], stress_mpa: list[float]) -> tuple[list[float], list[float]]:
"""Cut the curve at its stress peak and append a point at zero stress,
same strain, so it plots as a vertical drop — used for the individual
raw-curve figure. See VERTICAL_BREAK_MATERIALS."""
if not stress_mpa:
return strain, stress_mpa
peak_i = max(range(len(stress_mpa)), key=lambda i: stress_mpa[i])
return strain[:peak_i + 1] + [strain[peak_i]], stress_mpa[:peak_i + 1] + [0.0]
def truncate_at_peak(strain: list[float], stress_mpa: list[float]) -> tuple[list[float], list[float]]:
"""Cut the curve at its stress peak with no added point — used for the
mean +/- SD average figure, where a synthetic vertical segment would
distort the shared strain grid / averaging. See VERTICAL_BREAK_MATERIALS."""
if not stress_mpa:
return strain, stress_mpa
peak_i = max(range(len(stress_mpa)), key=lambda i: stress_mpa[i])
return strain[:peak_i + 1], stress_mpa[:peak_i + 1]
def load_specimen(path: Path) -> dict | None:
with path.open() as f:
row = json.loads(f.readline())
pairs = [
(s, t)
for s, t in zip(row["curves"]["strain"], row["curves"]["stress_pa"])
if s is not None and t is not None
]
if not pairs:
return None
strain, stress_pa = zip(*pairs)
return {
"row": row,
"strain": list(strain),
"stress_mpa": [t / 1e6 for t in stress_pa],
}
def load_standard(standard: str) -> list[dict]:
"""Load every specimen with a non-empty curve for a config. Callers that
need the VERTICAL_BREAK_MATERIALS peak-cut apply it themselves (see
vertical_break_at_peak / truncate_at_peak) — the two figures that need it
want different treatments (synthetic vertical drop vs. plain cut), so it
isn't baked into this loader."""
paths = sorted((DATA_DIR / standard).glob("*.jsonl"))
return [s for p in paths if (s := load_specimen(p))]
def style_for(row: dict) -> tuple[str, str]:
material = row["material_class"]
batch = row["batch_label"]
if material == "SLS":
color = BATCH_COLORS.get(batch, CONTROL_COLOR)
linestyle = TYPE_LINESTYLES.get(row["astm"].get("type"), "-")
else:
color = MATERIAL_COLORS.get(material, CONTROL_COLOR)
linestyle = MATERIAL_STYLES.get(material, "-")
return color, linestyle