Release/0.2.8

examples/fashion-mnist-parallel-coords-6d.py

@@ -0,0 +1,245 @@
+# /// script
+# requires-python = ">=3.12"
+# dependencies = [
+#     "marimo",
+#     "polars==1.38.1",
+#     "numpy==2.4.2",
+#     "scikit-learn==1.8.0",
+#     "wigglystuff==0.2.37",
+#     "matplotlib==3.10.8",
+#     "pandas==3.0.1",
+#     "anywidget-vector==0.2.7",
+# ]
+# ///
+
+import marimo
+
+__generated_with = "0.20.2"
+app = marimo.App(width="full")
+
+
+@app.cell
+def _():
+    import marimo as mo
+    import matplotlib.pyplot as plt
+    import numpy as np
+    import polars as pl
+    from sklearn.datasets import fetch_openml
+    from sklearn.decomposition import PCA
+    from wigglystuff import ParallelCoordinates
+
+    from anywidget_vector import VectorSpace
+
+    return PCA, ParallelCoordinates, VectorSpace, fetch_openml, mo, np, pl, plt
+
+
+@app.cell(hide_code=True)
+def _(mo):
+    mo.md(r"""
+    # Fashion MNIST: Parallel Coordinates + 3D Vector View
+
+    Brush the parallel coordinates axes to filter. The 3D scatter updates in real time.
+    Use the dropdowns to map PCA dimensions to visual channels.
+    """)
+    return
+
+
+@app.cell
+def _(fetch_openml, np):
+    mnist = fetch_openml("Fashion-MNIST", version=1, as_frame=False, parser="auto")
+    images = mnist.data.astype(np.float32)
+    labels = mnist.target.astype(int)
+
+    label_names = {
+        0: "T-shirt/top",
+        1: "Trouser",
+        2: "Pullover",
+        3: "Dress",
+        4: "Coat",
+        5: "Sandal",
+        6: "Shirt",
+        7: "Sneaker",
+        8: "Bag",
+        9: "Ankle boot",
+    }
+    LABEL_COLORS = {
+        "T-shirt/top": "#6366f1",
+        "Trouser": "#f59e0b",
+        "Pullover": "#10b981",
+        "Dress": "#ef4444",
+        "Coat": "#8b5cf6",
+        "Sandal": "#06b6d4",
+        "Shirt": "#f97316",
+        "Sneaker": "#84cc16",
+        "Bag": "#ec4899",
+        "Ankle boot": "#14b8a6",
+    }
+    SHAPE_NAMES = ["sphere", "cube", "cone", "octahedron", "cylinder", "tetrahedron"]
+    return LABEL_COLORS, SHAPE_NAMES, images, label_names, labels
+
+
+@app.cell
+def _(mo):
+    n_samples_slider = mo.ui.slider(start=500, stop=5000, step=500, value=2000, label="Samples")
+    n_components_slider = mo.ui.slider(start=3, stop=15, step=1, value=8, label="PCA dims")
+    mo.hstack([n_samples_slider, n_components_slider])
+    return n_components_slider, n_samples_slider
+
+
+@app.cell
+def _(
+    PCA,
+    images,
+    label_names,
+    labels,
+    n_components_slider,
+    n_samples_slider,
+    np,
+    pl,
+):
+    _rng = np.random.default_rng(42)
+    idx = _rng.choice(len(images), size=n_samples_slider.value, replace=False)
+
+    _pca = PCA(n_components=n_components_slider.value)
+    components = _pca.fit_transform(images[idx])
+
+    df = pl.DataFrame({f"PC{i + 1}": components[:, i] for i in range(n_components_slider.value)}).with_columns(
+        pl.Series("label", [label_names[labels[i]] for i in idx])
+    )
+    return components, df, idx
+
+
+@app.cell
+def _(ParallelCoordinates, df, mo):
+    widget = mo.ui.anywidget(ParallelCoordinates(df, color_by="label"))
+    widget
+    return (widget,)
+
+
+@app.cell
+def _(mo, n_components_slider):
+    pcs = [f"PC{i + 1}" for i in range(n_components_slider.value)]
+    x_dim = mo.ui.dropdown(options=pcs, value="PC1", label="X")
+    y_dim = mo.ui.dropdown(options=pcs, value="PC2", label="Y")
+    z_dim = mo.ui.dropdown(options=pcs, value="PC3", label="Z")
+    size_dim = mo.ui.dropdown(options=["none", *pcs], value="none", label="Size")
+    color_dim = mo.ui.dropdown(options=["label", *pcs], value="label", label="Color")
+    shape_dim = mo.ui.dropdown(options=["label", "none"], value="none", label="Shape")
+    mo.hstack([x_dim, y_dim, z_dim, size_dim, color_dim, shape_dim], gap=0.5)
+    return color_dim, shape_dim, size_dim, x_dim, y_dim, z_dim
+
+
+@app.cell
+def _(
+    LABEL_COLORS,
+    SHAPE_NAMES,
+    VectorSpace,
+    color_dim,
+    components,
+    idx,
+    label_names,
+    labels,
+    mo,
+    shape_dim,
+    size_dim,
+    x_dim,
+    y_dim,
+    z_dim,
+):
+    def _pc(name):
+        return int(name[2:]) - 1
+
+    _xi, _yi, _zi = _pc(x_dim.value), _pc(y_dim.value), _pc(z_dim.value)
+    _color_pc = None if color_dim.value == "label" else _pc(color_dim.value)
+    _size_pc = None if size_dim.value == "none" else _pc(size_dim.value)
+    _unique_labels = sorted(set(label_names.values()))
+    _shape_map = {_n: SHAPE_NAMES[_j % len(SHAPE_NAMES)] for _j, _n in enumerate(_unique_labels)}
+
+    vs_points = []
+    for _i in range(len(components)):
+        _name = label_names[labels[idx[_i]]]
+        _p = {
+            "id": f"p_{_i}",
+            "x": float(components[_i, _xi]),
+            "y": float(components[_i, _yi]),
+            "z": float(components[_i, _zi]),
+            "label": _name,
+        }
+        if _color_pc is None:
+            _p["color"] = LABEL_COLORS[_name]
+        else:
+            _p["color_val"] = float(components[_i, _color_pc])
+        if _size_pc is not None:
+            _p["size_val"] = float(components[_i, _size_pc])
+        if shape_dim.value == "label":
+            _p["shape_cat"] = _name
+        vs_points.append(_p)
+
+    _vs_kwargs = {
+        "axis_labels": {"x": x_dim.value, "y": y_dim.value, "z": z_dim.value},
+    }
+    if _color_pc is not None:
+        _vs_kwargs["color_field"] = "color_val"
+        _vs_kwargs["color_scale"] = "viridis"
+    if _size_pc is not None:
+        _vs_kwargs["size_field"] = "size_val"
+        _vs_kwargs["size_range"] = [0.01, 0.06]
+    if shape_dim.value == "label":
+        _vs_kwargs["shape_field"] = "shape_cat"
+        _vs_kwargs["shape_map"] = _shape_map
+
+    vs_widget = VectorSpace(
+        points=vs_points,
+        width=1200,
+        height=500,
+        dark_mode=False,
+        background="#fafafa",
+        show_toolbar=True,
+        show_settings=True,
+        show_properties=False,
+        **_vs_kwargs,
+    )
+    vs = mo.ui.anywidget(vs_widget)
+    vs
+    return vs, vs_points, vs_widget
+
+
+@app.cell
+def _(LABEL_COLORS, color_dim, mo, vs_points, vs_widget, widget):
+    _filtered = set(widget.widget.filtered_indices)
+    _dim_color = "#d1d5db"
+    _updated = []
+    for _i, _p in enumerate(vs_points):
+        _new_p = dict(_p)
+        if _i in _filtered:
+            if color_dim.value == "label":
+                _new_p["color"] = LABEL_COLORS[_p["label"]]
+            else:
+                _new_p.pop("color", None)
+        else:
+            _new_p["color"] = _dim_color
+        _updated.append(_new_p)
+    vs_widget.points = _updated
+    mo.md(f"**{len(_filtered)}** / {len(vs_points)} points selected")
+    return
+
+
+@app.cell
+def _(idx, images, label_names, labels, np, plt, widget):
+    _filtered = widget.widget.filtered_indices
+    _sample_idx = np.array(_filtered[:10]) if len(_filtered) >= 10 else np.array(_filtered)
+
+    _fig, _axes = plt.subplots(1, len(_sample_idx), figsize=(2 * len(_sample_idx), 2))
+    if len(_sample_idx) == 1:
+        _axes = [_axes]
+    for _ax, _si in zip(_axes, _sample_idx, strict=False):
+        _ax.imshow(images[idx[_si]].reshape(28, 28), cmap="gray")
+        _ax.set_title(label_names[labels[idx[_si]]], fontsize=9)
+        _ax.axis("off")
+    plt.tight_layout()
+    _fig
+    return
+
+
+if __name__ == "__main__":
+    app.run()

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "anywidget-vector"
-version = "0.2.7"
+version = "0.2.8"
 description = "Interactive 3D vector visualization with query UI for vector databases"
 readme = "README.md"
 license = "Apache-2.0"

src/anywidget_vector/ui/canvas.js

@@ -461,6 +461,9 @@ export function createCanvas(model, container, callbacks) {
     const distance = size / (2 * Math.tan(Math.PI * camera.fov / 360));
     controls.target.copy(center);
     camera.position.copy(center.clone().add(new THREE.Vector3(0, 0, distance * 1.2)));
+    camera.near = Math.max(0.01, distance * 0.001);
+    camera.far = Math.max(1000, distance * 10);
+    camera.updateProjectionMatrix();
     controls.update();
   }