make chroma refinements for v1.3.0

examples/class.py

@@ -0,0 +1,68 @@
+class WLBPosteriorEstimator(PosteriorEstimatorTrainer):
+    """
+    Weighted likelihood bootstrap (WLB) estimator.
+
+    Trains models to approximate draws from the *weight posterior* (under
+    Jeffrey's prior).
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        assert not self.use_non_atomic_loss
+
+    def get_dataloaders(
+        self,
+        starting_round: int = 0,
+        training_batch_size: int = 200,
+        validation_fraction: float = 0.1,
+        resume_training: bool = False,
+        dataloader_kwargs: dict | None = None,
+    ) -> tuple[data.DataLoader, data.DataLoader]:
+        """
+        Add logic for generating session-specific WLB weights.
+
+        This is probably the easiest place to stick some fixed weights on a
+        point-wise basis for a given training run, and we load them later where
+        we need them in the ``train()`` loop.
+        """
+        theta, x, prior_masks = self.get_simulations(starting_round)
+
+        # generate session specific WLB weights to attach point-wise
+        N = theta.shape[0]
+        wlb_z = Exponential(1.0).sample((N,))
+        wlb_w = (wlb_z / wlb_z.sum()) * N
+
+        dataset = data.TensorDataset(theta, x, prior_masks, wlb_w)
+
+        num_examples = theta.size(0)
+        num_training_examples = int((1 - validation_fraction) * num_examples)
+        num_validation_examples = num_examples - num_training_examples
+
+        if not resume_training:
+            permuted_indices = torch.randperm(num_examples)
+            self.train_indices, self.val_indices = (
+                permuted_indices[:num_training_examples],
+                permuted_indices[num_training_examples:],
+            )
+
+        train_loader_kwargs = {
+            "batch_size": min(training_batch_size, num_training_examples),
+            "drop_last": True,
+            "sampler": SubsetRandomSampler(self.train_indices.tolist()),
+        }
+        val_loader_kwargs = {
+            "batch_size": min(training_batch_size, num_validation_examples),
+            "shuffle": False,
+            "drop_last": True,
+            "sampler": SubsetRandomSampler(self.val_indices.tolist()),
+        }
+        if dataloader_kwargs is not None:
+            train_loader_kwargs = dict(train_loader_kwargs, **dataloader_kwargs)
+            val_loader_kwargs = dict(val_loader_kwargs, **dataloader_kwargs)
+
+        train_loader = data.DataLoader(dataset, **train_loader_kwargs)
+        val_loader = data.DataLoader(dataset, **val_loader_kwargs)
+
+        return train_loader, val_loader
+

scripts/render.sh

@@ -4,12 +4,14 @@
 
 set -euo pipefail
 w=${1:?width}; h=${2:?height}
-out="kitty-$(date +%Y%m%d-%H%M%S).png"
-shift 2 || true
+out=${3:?file};
+#out="kitty-$(date +%Y%m%d-%H%M%S).png"
+shift 3 || true
 
 # spawn a kitty window w/ a mark
 title="kitty-$(date +%s%N)"
-kitty --title "$title" "$@" &
+sock="$XDG_RUNTIME_DIR/kitty-$title.sock"
+kitty -o allow_remote_control=yes --listen-on "unix:$sock" --title "$title" "$@" &
 
 # create a targeted rule for the marked window and resize
 sleep 2
@@ -24,12 +26,15 @@ read gx gy _ < <(awk -F'[ ,x]' '{print $1,$2}' <<<"$geom")
 read wx wy ww wh < <(awk -F'[ ,x]' '{print $1,$2,$3,$4}' <<<"$wgeom")
 inner_geom="$((gx+wx)),$((gy+wy)) ${ww}x${wh}"
 
-echo title=$title
-echo geom=$geom
-echo out=$out
+echo "+ title=$title"
+echo "+ geom=$geom"
+echo "+ out=$out"
 
 # take a screenshot
 mkdir -p "$(dirname "$out")"
 grim -g "$inner_geom" "$out"
 echo "saved: $out"
 
+# close the kitty window
+kitty @ --to "unix:$sock" close-window
+

scripts/screens.sh

@@ -1,18 +1,33 @@
 #!/usr/bin/env bash
+# usage: screens.sh prefix
+
 set -euo pipefail
+prefix=${1:-}
 
 script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 render_script="$script_dir/render.sh"
 
-biomes=(alpine badlands chaparral savanna grassland tundra)
-modes=(light dark)
+#biomes=(alpine)
+#modes=(light dark)
+biomes=(alpine)
+modes=(dark)
 
 for biome in "${biomes[@]}"; do
     for mode in "${modes[@]}"; do
-        symconf config -m "$mode" -s "default-$biome-monobiome"
-        sleep 5
-        "$render_script" 800 600 nvim \
+        echo "Applying $biome-$mode theme"
+        symconf config \
+            -a kitty,nvim \
+            -m "$mode" \
+            -s "default-$biome-monobiome" \
+            -T font=Berkeley
+        sleep 2
+
+        echo "Taking screenshot..."
+        "$render_script" 800 600 "images/render/$prefix-$biome-$mode.png" nvim \
             +':highlight Cursor blend=100' \
             +':set guicursor=n:block-Cursor' \
             +':silent! setlocal nonumber nocursorline signcolumn=no foldcolumn=no' \
-            examples/runner.py
+            +':lua vim.diagnostic.config({virtual_text=false,signs=false,underline=false})' \
+            examples/class.py
+    done
+done