Fix some style issues

archetypes/datasets/make_archetypal_dataset.py

@@ -57,10 +57,17 @@ def make_archetypal_dataset(
     A = [np.zeros((s_i, a_i)) for s_i, a_i in zip(shape, n_archetypes)]
 
     for A_i, labels_i in zip(A, labels):
+        l_i_prev = -1
         for i, l_i in enumerate(labels_i):
-            alpha_i = [alpha] * A_i.shape[1]
-            alpha_i[l_i] = 1
-            A_i[i, :] = generator.dirichlet(alpha_i)
+            if l_i_prev != l_i:
+                alpha_i = [0] * A_i.shape[1]
+                alpha_i[l_i] = 1
+                A_i[i, :] = alpha_i
+                l_i_prev = l_i
+            else:
+                alpha_i = [alpha] * A_i.shape[1]
+                alpha_i[l_i] = 1
+                A_i[i, :] = generator.dirichlet(alpha_i)
 
     X = einsum(A, archetypes)
 

archetypes/datasets/permutations.py

@@ -88,9 +88,13 @@ def sort_by_archetype_similarity(data, alphas):
     data, info = permute_dataset(data, perms)
 
     labels = [np.argmax(a, axis=1) for a in alphas]
+    scores = [np.max(a, axis=1) for a in alphas]
     labels = [labels[i][perms[i]] for i in range(data.ndim)]
+    scores = [scores[i][perms[i]] for i in range(data.ndim)]
 
     info["labels"] = labels
+    info["scores"] = scores
+    info["n_archetypes"] = [ai.shape[1] for ai in alphas]
 
     return data, info
 

archetypes/visualization/bisimplex.py

@@ -2,19 +2,8 @@
 import matplotlib.pyplot as plt
 import numpy as np
 
-from archetypes.visualization import simplex
-
-
-def _create_palette(saturation, value, n_colors, int_colors=3):
-    hue = np.linspace(0, 1, n_colors, endpoint=False)
-    hue = np.hstack([hue[i::int_colors] for i in range(int_colors)])
-    saturation = np.full(n_colors, saturation)
-    value = np.full(n_colors, value)
-    # convert to RGB
-    c = mpl.colors.hsv_to_rgb(np.vstack([hue, saturation, value]).T)
-    # Create palette
-    palette = mpl.colors.ListedColormap(c)
-    return palette
+from .simplex import simplex
+from .utils import create_palette
 
 
 def bisimplex(alphas, archetypes, ax=None, **kwargs):
@@ -44,7 +33,7 @@ def bisimplex(alphas, archetypes, ax=None, **kwargs):
     n_archetypes = archetypes.shape
 
     # Get the colors for the vertices of the polytopes
-    palette = _create_palette(
+    palette = create_palette(
         saturation=0.35, value=0.9, n_colors=n_archetypes[0] + n_archetypes[1], int_colors=1
     )
 
@@ -85,7 +74,7 @@ def bisimplex(alphas, archetypes, ax=None, **kwargs):
 
     # Use grayscale palette from matplotlib between 0 and 1
     # archetypes[:] = .6
-    palette = mpl.colormaps["Grays"]
+    palette = mpl.colormaps["Greys"]
 
     for x_i, y_i, c_i, a_i in zip(
         xx.flatten(), yy.flatten(), archetypes.flatten(), archetypes_scaled.flatten()

archetypes/visualization/heatmap.py

@@ -1,17 +1,26 @@
 import matplotlib.pyplot as plt
 import numpy as np
+from matplotlib.colors import LinearSegmentedColormap, to_rgb
 from matplotlib.patches import Polygon
 
+from .utils import create_palette
 
-def heatmap(data, labels=None, ax=None, **kwargs):
+
+def heatmap(data, labels=None, n_archetypes=None, scores=None, ax=None, **kwargs):
     """Plot a heatmap of the data. If labels are provided, the heatmap is divided into cells.
 
     Parameters
     ----------
     data: np.ndarray
         The data to plot.
     labels: list of np.ndarray or None
-        The labels to use to divide the heatmap into cells. If None, no labels are used.
+        The labels values to use for the plot.
+        If None, the labels values are computed from the labels.
+    n_archetypes: list of int or None
+        The number of archetypes for each dimension.
+        If None, the number of archetypes is computed from the labels.
+    scores: list of np.ndarray or None
+        The scores values to use for the plot.
     ax: matplotlib.pyplot.axes or None
         The axes to plot on. If None, a new figure and axes is created.
     kwargs: dict
@@ -33,6 +42,8 @@ def heatmap(data, labels=None, ax=None, **kwargs):
     if "cmap" not in kwargs:
         kwargs["cmap"] = "Greys"
 
+    data_size = max(data.shape)
+
     # Plot line if labels[i] != labels[i+1]
     if labels is not None:
         # check labels is a list of 2 arrays
@@ -47,22 +58,87 @@ def heatmap(data, labels=None, ax=None, **kwargs):
                 f"labels must be a list of 2 arrays, got {type(labels[0])} and {type(labels[1])}"
             )
 
-        labels_h = np.concatenate([[-1], labels[1].flatten(), [-1]])
-        labels_v = np.concatenate([[-1], labels[0].flatten(), [-1]])
+        labels_h = np.concatenate([labels[1].flatten()])
+        labels_v = np.concatenate([labels[0].flatten()])
 
-        polygon_kwargs = {"color": "r", "lw": 1}
+        polygon_kwargs = {"color": "k", "lw": 1}
 
         for i in range(len(labels_h) - 1):
             if labels_h[i] != labels_h[i + 1]:
-                line = Polygon(np.array([[i, 0], [i, data.shape[0]]]) - 0.5, **polygon_kwargs)
+                line = Polygon(
+                    np.array([[i + 1, 0], [i + 1, data.shape[0]]]) - 0.5, **polygon_kwargs
+                )
                 ax.add_patch(line)
 
         for i in range(len(labels_v) - 1):
             if labels_v[i] != labels_v[i + 1]:
-                line = Polygon(np.array([[0, i], [data.shape[1], i]]) - 0.5, **polygon_kwargs)
+                line = Polygon(
+                    np.array([[0, i + 1], [data.shape[1], i + 1]]) - 0.5, **polygon_kwargs
+                )
                 ax.add_patch(line)
 
-    ax.matshow(data, rasterized=True, **kwargs)
+        if n_archetypes is None:
+            n_archetypes = [len(np.unique(labels[0])), len(np.unique(labels[1]))]
+
+        # Add a rectangle to frame the data
+        rect = Polygon(
+            np.array(
+                [[0, 0], [data.shape[1], 0], [data.shape[1], data.shape[0]], [0, data.shape[0]]]
+            )
+            - 0.5,
+            fill=False,
+            **polygon_kwargs,
+        )
+        ax.add_patch(rect)
+
+        palette = create_palette(
+            saturation=0.35, value=0.9, n_colors=n_archetypes[0] + n_archetypes[1], int_colors=1
+        )
+        colors = palette(np.arange(n_archetypes[0] + n_archetypes[1]))
+        colors_1 = colors[: n_archetypes[0]]
+        colors_2 = colors[n_archetypes[0] : n_archetypes[0] + n_archetypes[1]]
+
+        # Plot archetypes
+        counts = [np.count_nonzero(labels[0] == i) for i in range(n_archetypes[0])]
+        counts = np.cumsum(counts)
+        counts = np.concatenate([[0], counts]) - 0.5
+
+        arch_factor = 0.05 * data_size
+
+        if scores is None:
+            scores = [np.ones_like(labels[0]), np.ones_like(labels[1])]
+
+        for c, (i0, i1) in zip(colors_1, zip(counts, counts[1:])):
+            c1 = np.array(to_rgb(c))
+            c2 = np.array([1, 1, 1])
+
+            ax.imshow(
+                scores[0][int(i0 + 0.5) : int(i1 + 0.5)][::-1].reshape(-1, 1),
+                extent=[-arch_factor, -2 * arch_factor, i0, i1],
+                cmap=LinearSegmentedColormap.from_list("c", [c2, c1]),
+                interpolation="none",
+                vmax=1,
+                vmin=0,
+            )
+
+        counts = [np.count_nonzero(labels[1] == i) for i in range(n_archetypes[1])]
+        counts = np.cumsum(counts)
+        counts = np.concatenate([[0], counts]) - 0.5
+
+        for c, (i0, i1) in zip(colors_2, zip(counts, counts[1:])):
+            c1 = np.array(to_rgb(c))
+            c2 = np.array([1, 1, 1])
+
+            ax.imshow(
+                scores[1][int(i0 + 0.5) : int(i1 + 0.5)].reshape(1, -1),
+                extent=[i0, i1, -arch_factor, -2 * arch_factor],
+                cmap=LinearSegmentedColormap.from_list("c", [c2, c1]),
+                interpolation="none",
+                vmax=1,
+                vmin=0,
+            )
+
+    ax.matshow(data, interpolation="none", **kwargs)
 
     # set aspect ratio to equal
     ax.set_aspect("equal")
@@ -75,9 +151,8 @@ def heatmap(data, labels=None, ax=None, **kwargs):
     xlim = ax.get_xlim()
     ylim = ax.get_ylim()
 
-    exp_factor = 0.01 * max(np.abs(np.diff(xlim)), np.abs(np.diff(ylim)))
-
-    ax.set_xlim(xlim[0] - exp_factor, xlim[1] + exp_factor)
-    ax.set_ylim(ylim[0] + exp_factor, ylim[1] - exp_factor)
+    lim_factor = 0.01 * data_size
+    ax.set_xlim(xlim[0] - lim_factor, xlim[1] + lim_factor)
+    ax.set_ylim(ylim[0] + lim_factor, ylim[1] - lim_factor)
 
     return ax

archetypes/visualization/simplex.py

@@ -88,7 +88,7 @@ def simplex(
         for p1, p2 in edges:
             x1, y1 = p1
             x2, y2 = p2
-            ax.plot([x1, x2], [y1, y2], "-", linewidth=0.75, color="lightgray", zorder=0)
+            ax.plot([x1, x2], [y1, y2], "-", linewidth=1, color="lightgray", zorder=0)
 
     # ax.plot(vertices[:, 0], vertices[:, 1], "o", color="black", alpha=1)
     if show_vertices:

archetypes/visualization/utils.py

@@ -0,0 +1,14 @@
+import matplotlib as mpl
+import numpy as np
+
+
+def create_palette(saturation, value, n_colors, int_colors=3):
+    hue = np.linspace(0, 1, n_colors, endpoint=False)
+    hue = np.hstack([hue[i::int_colors] for i in range(int_colors)])
+    saturation = np.full(n_colors, saturation)
+    value = np.full(n_colors, value)
+    # convert to RGB
+    c = mpl.colors.hsv_to_rgb(np.vstack([hue, saturation, value]).T)
+    # Create palette
+    palette = mpl.colors.ListedColormap(c)
+    return palette