Spaces:
Paused
Paused
File size: 6,568 Bytes
d8328bf |
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 |
"""Tools for visualising embedding spaces using UMAP."""
from __future__ import annotations
import argparse
from pathlib import Path
from typing import List
import matplotlib.pyplot as plt
import numpy as np
from matplotlib.lines import Line2D
from mpl_toolkits.mplot3d import Axes3D # noqa: F401 - needed for 3D projections
from umap import UMAP
from pipeline.storage import load_corpus, load_embeddings
DEFAULT_INDEX_DIR = Path("index")
DEFAULT_CORPUS_PATH = DEFAULT_INDEX_DIR / "corpus.json"
DEFAULT_EMBEDDINGS_PATH = DEFAULT_INDEX_DIR / "embeddings.npy"
def parse_args(argv: List[str] | None = None) -> argparse.Namespace:
"""Parse command-line options for the visualiser.
Parameters
----------
argv: List[str] | None, default None
Optional argument list override for testing.
Returns
-------
argparse.Namespace
Parsed CLI arguments.
"""
parser = argparse.ArgumentParser(description="Visualise SPECTER2 embeddings in 2D or 3D.")
parser.add_argument(
"--embeddings",
type=Path,
default=DEFAULT_EMBEDDINGS_PATH,
help="Path to embeddings.npy (default: index/embeddings.npy)",
)
parser.add_argument(
"--corpus",
type=Path,
default=DEFAULT_CORPUS_PATH,
help="Path to corpus.json metadata (default: index/corpus.json)",
)
parser.add_argument(
"--dims",
type=int,
choices=(2, 3),
default=2,
help="Number of UMAP dimensions (2 or 3, default: 2)",
)
parser.add_argument(
"--output",
type=Path,
default=None,
help="Optional output path for the generated plot (default: derived from embeddings path)",
)
parser.add_argument(
"--show",
action="store_true",
help="Display the plot interactively after saving",
)
return parser.parse_args(argv)
def plot_embeddings(
embeddings_path: Path,
corpus_path: Path,
dims: int = 2,
output_path: Path | None = None,
show: bool = False,
) -> Path:
"""Create a UMAP projection and save the resulting plot.
Parameters
----------
embeddings_path: Path
Location of the `embeddings.npy` file.
corpus_path: Path
Location of the `corpus.json` file.
dims: int, default 2
Number of UMAP dimensions (2 or 3).
output_path: Path | None, default None
Optional destination for the saved figure. Uses a default if not provided.
show: bool, default False
Whether to display the plot after saving.
Returns
-------
Path
The path to the saved figure.
"""
embeddings = load_embeddings(embeddings_path)
corpus = load_corpus(corpus_path)
if embeddings.shape[0] != len(corpus):
raise ValueError(
"Embeddings and corpus lengths do not match. Ensure the inputs originate from the same build run."
)
reducer = UMAP(n_components=dims, n_neighbors=15, min_dist=0.1, random_state=42)
coordinates = reducer.fit_transform(embeddings)
categories = [metadata.get("categories", []) for metadata in corpus]
primary_labels = [category[0] if category else "unknown" for category in categories]
label_to_index = {label: idx for idx, label in enumerate(sorted(set(primary_labels)))}
colour_indices = np.array([label_to_index[label] for label in primary_labels])
fig = _create_figure(coordinates, colour_indices, primary_labels, label_to_index, dims)
derived_output = embeddings_path.with_name(f"embedding_plot_{dims}d.png")
output = output_path or derived_output
output.parent.mkdir(parents=True, exist_ok=True)
fig.savefig(output, dpi=200, bbox_inches="tight")
if show:
plt.show()
else:
plt.close(fig)
print(f"Saved {dims}D embedding visualisation to {output}")
return output
def _create_figure(
coordinates: np.ndarray,
colour_indices: np.ndarray,
labels: List[str],
label_to_index: dict[str, int],
dims: int,
) -> plt.Figure:
"""Create a matplotlib figure for the requested dimensionality.
Parameters
----------
coordinates: np.ndarray
UMAP-reduced coordinates of shape (n_samples, dims).
colour_indices: np.ndarray
Integer indices representing colour assignments per sample.
labels: List[str]
Primary category labels aligned with the coordinates.
label_to_index: dict[str, int]
Mapping from label names to integer colour indices.
dims: int
Dimensionality of the embedding visualisation (2 or 3).
Returns
-------
plt.Figure
The generated matplotlib figure.
"""
plt.rcdefaults()
fig = plt.figure(figsize=(10, 8))
if dims == 2:
ax = fig.add_subplot(111)
scatter = ax.scatter(
coordinates[:, 0],
coordinates[:, 1],
c=colour_indices,
cmap="tab20",
s=20,
alpha=0.85,
)
ax.set_xlabel("UMAP 1")
ax.set_ylabel("UMAP 2")
else:
ax = fig.add_subplot(111, projection="3d")
scatter = ax.scatter(
coordinates[:, 0],
coordinates[:, 1],
coordinates[:, 2],
c=colour_indices,
cmap="tab20",
s=20,
alpha=0.85,
)
ax.set_xlabel("UMAP 1")
ax.set_ylabel("UMAP 2")
ax.set_zlabel("UMAP 3")
ax.set_title(f"SPECTER2 Embeddings ({dims}D UMAP)")
# Build a small legend using the primary labels.
unique_labels = sorted(set(labels))
handles = []
for label in unique_labels:
colour_value = label_to_index[label]
rgba = scatter.cmap(scatter.norm(colour_value))
handle = Line2D([0], [0], marker="o", color="w", label=label, markerfacecolor=rgba, markersize=8)
handles.append(handle)
if len(handles) <= 12:
ax.legend(handles=handles, title="Primary Category", bbox_to_anchor=(1.05, 1), loc="upper left")
return fig
def main(argv: List[str] | None = None) -> None:
"""Entry point for the visualisation CLI.
Parameters
----------
argv: List[str] | None, default None
Optional argument override when invoking programmatically.
"""
args = parse_args(argv)
plot_embeddings(
embeddings_path=args.embeddings,
corpus_path=args.corpus,
dims=args.dims,
output_path=args.output,
show=args.show,
)
if __name__ == "__main__": # pragma: no cover - CLI entry point
main()
|