For real?

.pre-commit-config.yaml

@@ -48,4 +48,4 @@ repos:
     language: system
     files: "paper/.*"
     pass_filenames: false
-    entry: sh -c "(cd paper && pipx run --spec shiny-mdc shinymdc -i main.md -o main.pdf -t stylish -m smalltabs=true)"
+    entry: sh -c "(cd paper && pipx run --spec git+https://github.com/jayanthkoushik/shiny-mdc shinymdc -i main.md -o main.pdf -t stylish -m smalltabs=true,nonidan=true)"

alphanet/plot.py

@@ -25,6 +25,7 @@
     patheffects as pe,
     pyplot as plt,
 )
+from matplotlib.container import BarContainer
 from matplotlib.ticker import (
     FixedFormatter,
     FixedLocator,
@@ -817,6 +818,7 @@ class PlotSplitAccVsExp(_BaseMultiExpPlotCmd, BasePlotCmd):
     xlabel: str = ""
     legend_loc: str = "upper right"
     legend_bbox_to_anchor: Optional[Tuple[float, float]] = None
+    show_titles: bool = True
 
     @staticmethod
     def _get_metric_desc(metric: Literal["euclidean", "cosine", "random"]) -> str:
@@ -891,8 +893,10 @@ def __call__(self):
                     fig=_fig, ax=_ax, left=True, right=True, bottom=False, top=True
                 )
 
-                if self.col is not None and len(cols) > 1:
+                if self.col is not None and len(cols) > 1 and self.show_titles:
                     _ax.set_title(_col)
+                else:
+                    _ax.set_title("")
 
                 _legend = _ax.legend()
                 if _i == 0:
@@ -1872,8 +1876,16 @@ def __call__(self):
             g.set_titles("")
         for facet_name, ax in g.axes_dict.items():
             ax_df = df[df["Dataset"] == facet_name]
-            ax.yaxis.set_major_formatter(PercentFormatter(xmax=len(ax_df)))
+            fmtr = PercentFormatter(xmax=len(ax_df))
+            ax.yaxis.set_major_formatter(fmtr)
             ax.set_yticks([int(len(ax_df) * _p / 100) for _p in [15, 30, 45]])
+            bar_containers = [
+                _con for _con in ax.containers if isinstance(_con, BarContainer)
+            ]
+            assert len(bar_containers) == 1
+            bar_container = bar_containers[0]
+            # pylint: disable=cell-var-from-loop,unnecessary-lambda
+            ax.bar_label(bar_container, fmt=lambda _n: fmtr(_n), label_type="center")
         g.set_xlabels("")
         g.set_ylabels("")
         g.despine(top=True, left=True, right=True, bottom=False, trim=False)

alphanet/train.py

@@ -285,6 +285,22 @@ def __call__(self) -> TrainResult:
             self.training.pred_scale,
         ).to(DEFAULT_DEVICE)
 
+        # # Initialize weights.
+        # _hact = self.alphanet.hact
+        # if _hact == "relu" or _hact == "leaky_relu":
+        #     _a = 0.01 if _hact == "leaky_relu" else 0
+
+        #     def _w_init_fn(_w):
+        #         torch.nn.init.kaiming_normal_(_w, a=_a, nonlinearity=_hact)
+
+        # else:
+        #     _w_init_fn = torch.nn.init.xavier_normal_
+        # _b_init_fn = torch.nn.init.zeros_
+
+        # for _param in self.alphanet.linear_layer__seq + [self.alphanet.conv_layer]:
+        #     _w_init_fn(_param.weight)
+        #     _b_init_fn(_param.bias)
+
         self.training.ptopt.set_weights(alphanet_classifier.parameters())
         loss_fn = torch.nn.CrossEntropyLoss()
         logging.info("setting up model...done")