NTT-learning/ntt_learning/visuals.py

329 lines
11 KiB
Python

"""Blunt visual helpers for the NTT notebooks."""
from __future__ import annotations
from typing import Sequence
import ipywidgets as widgets
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
from IPython.display import clear_output, display
from .toy_ntt import TransformStage, TransformTrace, pairwise_product_grid, wraparound_contributions
def _value_colors(values: Sequence[int]) -> list[str]:
colors = []
for value in values:
if value < 0:
colors.append("#f08a5d")
elif value == 0:
colors.append("#d9d9d9")
else:
colors.append("#7ad3a8")
return colors
def _draw_value_row(ax, values: Sequence[int], y: float, prefix: str) -> None:
colors = _value_colors(values)
for index, (value, color) in enumerate(zip(values, colors)):
ax.text(
index,
y,
f"{prefix}{index}\n{value}",
ha="center",
va="center",
fontsize=10,
family="monospace",
bbox={
"boxstyle": "round,pad=0.35",
"facecolor": color,
"edgecolor": "#222222",
"linewidth": 1.2,
},
)
def plot_convolution_grid(
left: Sequence[int], right: Sequence[int], title: str = "Schoolbook Product Grid"
):
"""Plot the full schoolbook multiplication table and the diagonal sums."""
grid = pairwise_product_grid(left, right)
diagonal_sums = []
for diagonal in range(len(left) + len(right) - 1):
total = 0
for row in range(len(left)):
column = diagonal - row
if 0 <= column < len(right):
total += grid[row][column]
diagonal_sums.append(total)
fig, axes = plt.subplots(2, 1, figsize=(max(7, len(right) * 1.2), 6), height_ratios=[3, 1])
heatmap_ax, sum_ax = axes
heatmap_ax.imshow(grid, cmap="YlGnBu", aspect="auto")
heatmap_ax.set_title(title, fontsize=14, fontweight="bold")
heatmap_ax.set_xlabel("right coefficient index")
heatmap_ax.set_ylabel("left coefficient index")
heatmap_ax.set_xticks(range(len(right)))
heatmap_ax.set_yticks(range(len(left)))
for row, row_values in enumerate(grid):
for column, value in enumerate(row_values):
heatmap_ax.text(column, row, str(value), ha="center", va="center", color="#101010", fontsize=10)
sum_ax.axis("off")
sum_ax.set_title("Diagonal Sums = Convolution Coefficients", fontsize=12, fontweight="bold", pad=8)
for index, value in enumerate(diagonal_sums):
sum_ax.text(
index,
0,
f"y{index}\n{value}",
ha="center",
va="center",
fontsize=10,
family="monospace",
bbox={
"boxstyle": "round,pad=0.35",
"facecolor": "#f4f1de",
"edgecolor": "#222222",
"linewidth": 1.0,
},
)
sum_ax.set_xlim(-0.5, len(diagonal_sums) - 0.5)
sum_ax.set_ylim(-1, 1)
fig.tight_layout()
return fig
def plot_wraparound(
coefficients: Sequence[int],
n: int,
*,
negacyclic: bool = True,
title: str | None = None,
):
"""Plot how the tail wraps back into degree < n."""
rows = wraparound_contributions(coefficients, n=n, negacyclic=negacyclic)
if title is None:
title = "Negacyclic Folding" if negacyclic else "Cyclic Folding"
fig, ax = plt.subplots(figsize=(max(8, len(coefficients) * 1.1), 5.5))
ax.set_title(title, fontsize=14, fontweight="bold")
ax.axis("off")
top_y = 2.4
bottom_y = 0.4
_draw_value_row(ax, coefficients, top_y, "x^")
reduced_values = [row["total"] for row in rows]
_draw_value_row(ax, reduced_values, bottom_y, "slot ")
for slot, row in enumerate(rows):
for contribution in row["contributions"]:
source_index = contribution["source_index"]
color = "#d1495b" if contribution["sign"] < 0 else "#2a9d8f"
label = "-" if contribution["sign"] < 0 else "+"
ax.annotate(
"",
xy=(slot, bottom_y + 0.3),
xytext=(source_index, top_y - 0.25),
arrowprops={"arrowstyle": "->", "color": color, "linewidth": 2.0},
)
mid_x = (slot + source_index) / 2
mid_y = (top_y + bottom_y) / 2 + 0.25
ax.text(
mid_x,
mid_y,
f"{label} wrap {contribution['wraps']}",
ha="center",
va="center",
fontsize=9,
color=color,
family="monospace",
)
ax.set_xlim(-0.8, max(len(coefficients), n) - 0.2)
ax.set_ylim(-0.4, 3.2)
fig.tight_layout()
return fig
def plot_bit_reversal_mapping(length: int, title: str = "Normal Order To Bit-Reversed Order"):
"""Plot the bit-reversal permutation as explicit wires."""
if length <= 0 or length & (length - 1):
raise ValueError("plot_bit_reversal_mapping requires a power-of-two length")
from .toy_ntt import bit_reversed_indices
permutation = bit_reversed_indices(length)
width = length.bit_length() - 1
fig, ax = plt.subplots(figsize=(8, max(4, length * 0.65)))
ax.set_title(title, fontsize=14, fontweight="bold")
ax.axis("off")
for index, target in enumerate(permutation):
ax.text(
0,
-index,
f"{index:>2} | {index:0{width}b}",
ha="center",
va="center",
family="monospace",
bbox={"boxstyle": "round,pad=0.25", "facecolor": "#edf6f9", "edgecolor": "#264653"},
)
ax.text(
4,
-target,
f"{target:>2} | {target:0{width}b}",
ha="center",
va="center",
family="monospace",
bbox={"boxstyle": "round,pad=0.25", "facecolor": "#fff3b0", "edgecolor": "#9c6644"},
)
ax.plot([0.6, 3.4], [-index, -target], color="#7f5539", linewidth=2.2, alpha=0.9)
ax.text(0, 1, "NO", ha="center", va="center", fontsize=12, fontweight="bold")
ax.text(4, 1, "BO", ha="center", va="center", fontsize=12, fontweight="bold")
ax.set_xlim(-1.2, 5.2)
ax.set_ylim(-length + 0.2, 1.8)
fig.tight_layout()
return fig
def plot_stage(stage: TransformStage, title: str | None = None):
"""Plot one explicit butterfly stage with input and output rows."""
if title is None:
title = f"{stage.algorithm.upper()} Stage {stage.stage_index}"
fig, ax = plt.subplots(figsize=(max(8, len(stage.input_values) * 1.35), 5.8))
ax.set_title(title, fontsize=14, fontweight="bold")
ax.axis("off")
input_y = 2.6
output_y = 0.5
_draw_value_row(ax, stage.input_values, input_y, "i")
_draw_value_row(ax, stage.output_values, output_y, "o")
colors = ["#264653", "#2a9d8f", "#e76f51", "#8d99ae", "#c1121f", "#3a86ff"]
for pair_index, ((left, right), zeta) in enumerate(zip(stage.pairings, stage.zetas)):
color = colors[pair_index % len(colors)]
center_x = (left + right) / 2
ax.plot([left, right], [input_y - 0.45, input_y - 0.45], color=color, linewidth=2.5)
ax.plot([left, left], [input_y - 0.45, output_y + 0.55], color=color, linewidth=1.5, alpha=0.85)
ax.plot([right, right], [input_y - 0.45, output_y + 0.55], color=color, linewidth=1.5, alpha=0.85)
ax.text(
center_x,
1.55,
f"pair {left}-{right}\nzeta={zeta}",
ha="center",
va="center",
fontsize=10,
family="monospace",
bbox={
"boxstyle": "round,pad=0.35",
"facecolor": "#ffffff",
"edgecolor": color,
"linewidth": 1.4,
},
)
ax.text(
len(stage.input_values) / 2 - 0.5,
-0.05,
stage.note,
ha="center",
va="center",
fontsize=10,
color="#333333",
)
ax.set_xlim(-0.8, len(stage.input_values) - 0.2)
ax.set_ylim(-0.5, 3.3)
fig.tight_layout()
return fig
def plot_trace_overview(trace: TransformTrace, title: str | None = None):
"""Plot every stage output as a column of values."""
if title is None:
title = f"{trace.algorithm.upper()} Trace Overview"
columns = [trace.input_values] + [stage.output_values for stage in trace.stages]
fig, ax = plt.subplots(figsize=(max(9, len(columns) * 2.0), max(4.5, len(trace.input_values) * 0.7)))
ax.set_title(title, fontsize=14, fontweight="bold")
ax.axis("off")
for column_index, values in enumerate(columns):
x = column_index * 2.0
for row_index, value in enumerate(values):
ax.text(
x,
-row_index,
str(value),
ha="center",
va="center",
fontsize=10,
family="monospace",
bbox={
"boxstyle": "round,pad=0.25",
"facecolor": _value_colors([value])[0],
"edgecolor": "#222222",
"linewidth": 1.0,
},
)
if column_index == 0:
label = "input"
else:
label = f"stage {column_index}"
ax.text(x, 1, label, ha="center", va="center", fontsize=11, fontweight="bold")
ax.set_xlim(-1.0, (len(columns) - 1) * 2.0 + 1.0)
ax.set_ylim(-len(trace.input_values) + 0.2, 1.8)
fig.tight_layout()
return fig
def interactive_trace(trace: TransformTrace, title: str | None = None):
"""Return a slider-based stage explorer for a transform trace."""
if title is None:
title = f"{trace.algorithm.upper()} Stage Explorer"
slider = widgets.IntSlider(
value=1,
min=1,
max=max(1, len(trace.stages)),
step=1,
description="Stage",
continuous_update=False,
)
output = widgets.Output()
def render(stage_index: int) -> None:
with output:
clear_output(wait=True)
stage = trace.stages[stage_index - 1]
fig = plot_stage(stage, title=f"{title} | Stage {stage_index}")
display(fig)
plt.close(fig)
rows = []
for pair, zeta in zip(stage.pairings, stage.zetas):
left, right = pair
rows.append(
f"pair {pair}: inputs=({stage.input_values[left]}, {stage.input_values[right]}) "
f"-> outputs=({stage.output_values[left]}, {stage.output_values[right]}) | zeta={zeta}"
)
print("\n".join(rows))
slider.observe(lambda change: render(change["new"]), names="value")
render(slider.value)
widget = widgets.VBox([widgets.HTML(f"<h4>{title}</h4>"), slider, output])
return widget
def show_trace(trace: TransformTrace, title: str | None = None):
"""Display the interactive trace widget immediately."""
widget = interactive_trace(trace, title=title)
display(widget)
return widget