copy script

.gitignore

@@ -1,3 +1,5 @@
+lightning_logs*/
+lightning_logs_dev/
 lightning_logs/
 __pycache__/
 out/

scrape.py → archive/scrape.py

@@ -2,7 +2,7 @@ import glob
 import shutil
 from pathlib import Path
 
-from check import make_image
+from check import create_rectangle
 
 
 def get_exps(pattern: str, splitter: str = "_", dry_run: bool = True):
@@ -33,8 +33,8 @@ def get_exps(pattern: str, splitter: str = "_", dry_run: bool = True):
             print(latest_checkpoint)
         if not dry_run:
             shutil.copy(source_path, dir_path)
-            make_image(latest_checkpoint, f"out/version_{i}")
+            create_rectangle(latest_checkpoint, f"out/version_{i}")
-            # make_image(latest_checkpoint, f"out/version_{i}b", color=False)
+            # create_rectangle(latest_checkpoint, f"out/version_{i}b", color=False)
         else:
             print("Would copy", source_path, dir_path)
     return H

baseline.py

@@ -0,0 +1,170 @@
+import argparse
+from pathlib import Path
+
+import matplotlib.colors as mcolors
+import matplotlib.pyplot as plt
+import numpy as np
+from hilbertcurve.hilbertcurve import HilbertCurve
+
+from check import plot_preds
+
+# Extract XKCD colors
+colors = list(mcolors.XKCD_COLORS.keys())
+rgb_values = [mcolors.to_rgb(mcolors.XKCD_COLORS[color]) for color in colors]
+
+# Parse command-line arguments
+parser = argparse.ArgumentParser()
+parser.add_argument("-s", "--sort-by", type=str, default="hsv", help="kind of sorting")
+parser.add_argument("--seed", type=int, default=21, help="seed for UMAP")
+parser.add_argument("--dpi", type=int, default=300, help="dpi for saving")
+parser.add_argument("--size", type=float, default=6.0, help="size of figure")
+parser.add_argument(
+    "--fontsize",
+    type=float,
+    default=0,
+    help="fontsize of annotation (default: 0 = None)",
+)
+parser.add_argument(
+    "--radius", type=float, default=1 / 2, help="inner radius of circle"
+)
+args = parser.parse_args()
+KIND = args.sort_by
+SEED = args.seed
+DPI = args.dpi
+SIZE = args.size
+FONTSIZE = args.fontsize
+INNER_RADIUS = args.radius
+DIR = "/teamspace/studios/this_studio/colors/colors-umap"
+
+
+prefix = ""
+if KIND == "umap":
+    prefix = f"{SEED:04d}"
+    FDIR = DIR
+else:
+    FDIR = f"{DIR}/{KIND}"
+Path(FDIR).mkdir(exist_ok=True, parents=True)
+fname = f"{FDIR}/v{prefix}.png"
+
+
+if KIND in ("lex", "alpha", "abc"):
+    preds = np.array(colors)
+
+elif KIND == "umap":
+    # from umap import UMAP
+    from cuml import UMAP
+
+    # Use UMAP to create a 1D representation
+    reducer = UMAP(
+        n_components=1,
+        n_neighbors=250,
+        min_dist=1e-2,
+        metric="euclidean",
+        random_state=SEED,
+        negative_sample_rate=2,
+    )
+    embedding = reducer.fit_transform(np.array(rgb_values))
+
+    # Sort colors by the 1D representation
+    preds = embedding[:, 0]
+    del reducer, embedding
+
+elif KIND in ("cielab", "lab", "ciede2000"):
+    from skimage.color import deltaE_ciede2000, rgb2lab
+
+    # CIELAB
+    # Convert RGB values to CIELAB
+    lab_values = rgb2lab([rgb_values])
+
+    # Reference color for sorting (can be the first color or any other reference point)
+    reference_color = lab_values[0]
+
+    # Compute CIEDE2000 distances of all colors to the reference color
+    distances = [deltaE_ciede2000(reference_color, color) for color in lab_values]
+
+    # Sort colors by their CIEDE2000 distance to the reference color
+    # preds = distances).flatten()  # awful
+    lab_values_flat = lab_values.reshape(-1, 3)
+    # Sort colors based on the L* value in the CIELAB space
+    # 0 corresponds to the L* channel
+    preds = lab_values_flat[:, 0]
+
+elif KIND == "hsv":
+    from matplotlib.colors import rgb_to_hsv
+
+    # Convert RGB values to HSV
+    hsv_values = np.array([rgb_to_hsv(np.array(rgb)) for rgb in rgb_values])
+
+    # Sort colors based on the hue value
+    # 0 corresponds to the hue component
+    preds = hsv_values[:, 0]
+else:
+    raise ValueError(f"Unknown kind: {KIND}")
+
+sorted_indices = np.argsort(preds)
+
+# Save the sorted indices to disk
+# if (KIND == "umap") or (KIND != "umap"):
+PDIR = f"scripts/{KIND}"
+Path(PDIR).mkdir(parents=True, exist_ok=True)
+file_path = f"{PDIR}/{SEED:06d}.npy"
+np.save(file_path, preds.ravel())
+print(f"Predictions saved to {file_path}")
+
+# Sort colors by the 1D representation
+sorted_colors = [colors[i] for i in sorted_indices]
+
+plot_preds(
+    preds,
+    np.array(rgb_values),
+    fname,
+    roll=False,
+    dpi=DPI,
+    radius=INNER_RADIUS,
+    figsize=(SIZE, SIZE),
+    fsize=FONTSIZE,
+    label=f"{KIND.upper()}",
+)
+print(f"saved {fname}")
+
+HILBERT = False
+
+if HILBERT:
+    # Create Hilbert curve
+    # We'll set the order such that the number of positions is greater than or equal to the number of colors
+    hilbert_order = int(np.ceil(0.5 * np.log2(len(sorted_colors))))
+    hilbert_curve = HilbertCurve(hilbert_order, 2)
+
+    # Create an image for visualization
+    image_size = 2**hilbert_order
+    image = np.ones((image_size, image_size, 3))
+
+    for i, color in enumerate(sorted_colors):
+        # Convert linear index to Hilbert coordinates
+        coords = hilbert_curve.point_from_distance(i)
+        image[coords[1], coords[0]] = mcolors.to_rgb(color)
+
+    # annotation in upper right
+    # Display the image
+    fig, ax = plt.subplots(1, 1, figsize=(SIZE, SIZE))
+    ax.imshow(image)
+    ax.annotate(
+        f"{KIND.upper()}",
+        (1.0, 1.0),
+        ha="right",
+        va="top",
+        size=FONTSIZE,
+        xycoords="axes fraction",
+    )
+    ax.axis("off")
+    ax.set_aspect("equal")
+    fig.tight_layout()
+    fname = f"{DIR}/{prefix}{KIND}_sorted_colors_hilbert.png"
+    fig.savefig(
+        fname,
+        dpi=DPI,
+        transparent=True,
+        # bbox_inches="tight",
+        # pad_inches=0
+    )
+    print(f"Saved {fname}")

callbacks.py

@@ -1,14 +1,27 @@
+# import subprocess
+import os
 from pathlib import Path
 
 from lightning import Callback
 
-from check import create_circle
+from check import create_circle_nonblocking as create_circle
 
 
 class SaveImageCallback(Callback):
-    def __init__(self, save_interval=1, final_dir: str = None):
+    def __init__(
+        self,
+        save_interval=1,
+        final_dir: str = None,
+        radius: float = 0.5,
+        dpi: int = 300,
+        figsize=(15, 15),  # 4k is at 14.4
+    ):
         self.save_interval = save_interval
         self.final_dir = final_dir
+        self.radius = radius
+        self.dpi = dpi
+        self.figsize = figsize
+        self.processes = []
 
     def on_train_epoch_end(self, trainer, pl_module):
         epoch = trainer.current_epoch
@@ -20,12 +33,15 @@ class SaveImageCallback(Callback):
             pl_module.eval()
 
             # Save the image
-            # if pl_module.trainer.logger:
-            #
-            # else:
-            #     version = 0
             fname = Path(pl_module.trainer.logger.log_dir) / Path(f"e{epoch:04d}")
-            create_circle(pl_module, fname=fname, dpi=300, figsize=(6, 6))
+            p = create_circle(
+                pl_module,
+                fname=fname,
+                dpi=self.dpi,
+                figsize=self.figsize,
+                radius=self.radius,
+            )
+            self.processes.append(p)
 
             # Make sure to set it back to train mode
             pl_module.train()
@@ -35,14 +51,24 @@ class SaveImageCallback(Callback):
             version = pl_module.trainer.logger.version
             fname = Path(f"{self.final_dir}") / Path(f"v{version}")
             pl_module.eval()
-            create_circle(pl_module, fname=fname, dpi=300, figsize=(6, 6))
+            p = create_circle(
-            if self.save_interval > 0:
+                pl_module,
-                import os
+                fname=fname,
+                dpi=self.dpi,
+                figsize=self.figsize,
+                radius=self.radius,
+            )
+            self.processes.append(p)
+
+            # Wait for all subprocesses to finish
+            for p in self.processes:
+                p.join()
 
+            if self.save_interval > 0:
                 log_dir = str(Path(pl_module.trainer.logger.log_dir))
                 fps = 12
-                _cmd = f'ffmpeg -r {fps} -f image2 -i {log_dir}/e%04d.png -vcodec libx264 -crf 25 -pix_fmt yuv420p -vf "scale=1920:1080:force_original_aspect_ratio=decrease,pad=1920:1080:(ow-iw)/2:(oh-ih)/2:color=white" {log_dir}/a{version}.mp4'
+                # w, h = self.figsize[0] * self.dpi, self.figsize[1] * self.dpi
+                w, h = 7680, 4320
+                _cmd = f'ffmpeg -r {fps} -f image2 -i {log_dir}/e%04d.png -vcodec libx264 -crf 25 -pix_fmt yuv420p -vf "scale={w}:{h}:force_original_aspect_ratio=decrease,pad={w}:{h}:(ow-iw)/2:(oh-ih)/2:color=white" {log_dir}/a{version}.mp4'
+                # _ = subprocess.Popen(_cmd, shell=True)
                 os.system(_cmd)
-                # os.system(
-                #     f'ffmpeg -i {log_dir}/e%04d.png -c:v libx264 -vf "fps={fps},format=yuv420p,pad=ceil(iw/2)*2:ceil(ih/2)*2" {log_dir}/a{version}.mp4'
-                # )

check.py

@@ -1,6 +1,8 @@
 # import matplotlib.patches as patches
+import pickle
+from multiprocessing import Process
 from pathlib import Path
-from typing import Union
+from typing import Tuple, Union
 
 import matplotlib.patches as patches
 import matplotlib.pyplot as plt
@@ -13,21 +15,21 @@ from model import ColorTransformerModel
 # import matplotlib.colors as mcolors
 
 
-def make_image(ckpt: str, fname: str, color=True, **kwargs):
+def create_rectangle(ckpt: str, fname: str, color: bool = True, **kwargs):
     M = ColorTransformerModel.load_from_checkpoint(ckpt)
 
     # preds = M(rgb_tensor)
-    if not color:
+    if color is False:  # black and white ordering...
         N = 949
         linear_space = torch.linspace(0, 1, N)
         rgb_tensor = linear_space.unsqueeze(1).repeat(1, 3)
     else:
         rgb_tensor, names = extract_colors()
 
-    rgb_values = rgb_tensor.detach().numpy()
+    rgb_tensor = preprocess_data(rgb_tensor).to(M.device)
-    rgb_tensor = preprocess_data(rgb_tensor)
+    preds = M(rgb_tensor).detach().cpu().numpy()
-    preds = M(rgb_tensor)
+    rgb_values = rgb_tensor.detach().cpu().numpy()
-    sorted_inds = np.argsort(preds.detach().numpy().ravel())
+    sorted_inds = np.argsort(preds.ravel())
 
     fig, ax = plt.subplots()
     for i in range(len(sorted_inds)):
@@ -42,11 +44,9 @@ def make_image(ckpt: str, fname: str, color=True, **kwargs):
     plt.savefig(f"{fname}.png", **kwargs)
 
 
-def create_circle(
+def do_inference(ckpt: Union[str, ColorTransformerModel]):
-    ckpt: Union[str, ColorTransformerModel], fname: str, skip: bool = True, **kwargs
-):
-    if isinstance(ckpt, str):
 
+    if isinstance(ckpt, str):
         M = ColorTransformerModel.load_from_checkpoint(
             ckpt, map_location=lambda storage, loc: storage
         )
@@ -55,16 +55,49 @@ def create_circle(
 
     xkcd_colors, _ = extract_colors()
     xkcd_colors = preprocess_data(xkcd_colors).to(M.device)
-    preds = M(xkcd_colors)
+    preds = M(xkcd_colors).detach().cpu().numpy()
     rgb_array = xkcd_colors.detach().cpu().numpy()
+    return preds, rgb_array
+
+
+def create_circle(ckpt: Union[str, ColorTransformerModel], fname: str, **kwargs):
+    preds, rgb_array = do_inference(ckpt)
     plot_preds(preds, rgb_array, fname=fname, **kwargs)
 
 
+def _plot_preds_serialized(serialized_data, fname, **kwargs):
+    # Deserialize the data
+    preds, rgb_array = pickle.loads(serialized_data)
+    plot_preds(preds, rgb_array, fname=fname, **kwargs)
+
+
+def create_circle_nonblocking(
+    ckpt: Union[str, ColorTransformerModel], fname: str, **kwargs
+):
+    preds, rgb_array = do_inference(ckpt)
+
+    # Serialize the data
+    serialized_data = pickle.dumps((preds, rgb_array))
+
+    # Run _plot_preds_serialized function in a separate process
+    p = Process(
+        target=_plot_preds_serialized, args=(serialized_data, fname), kwargs=kwargs
+    )
+    p.start()
+    return p
+
+
 def plot_preds(
-    preds, rgb_values, fname: str, roll: bool = False, dpi: int = 300, figsize=(6, 6)
+    preds: np.ndarray,
+    rgb_values: np.ndarray,
+    fname: str,
+    roll: bool = False,
+    radius: float = 1 / 2,
+    dpi: int = 300,
+    figsize: Tuple[float] = (6, 6),
+    fsize: int = 0,
+    label: str = "",
 ):
-    if isinstance(preds, torch.Tensor):
-        preds = preds.detach().cpu().numpy()
     sorted_inds = np.argsort(preds.ravel())
     colors = rgb_values[sorted_inds, :3]
     if roll:
@@ -107,56 +140,68 @@ def plot_preds(
     ax.set_yticks([])
     ax.set_aspect("equal")
     ax.axis("off")
-    radius = 1
+    ax.set_ylim(0, 1)  # implicit outer radius of 1
-    ax.set_ylim(-radius, radius)
 
     # Overlay white circle
-    inner_radius = 1 / 3
     circle = patches.Circle(
-        (0, 0), inner_radius, transform=ax.transData._b, color="white", zorder=2
+        (0, 0), radius, transform=ax.transData._b, color="white", zorder=2
     )
     ax.add_patch(circle)
 
+    if fsize > 0.0:
+        center = (0, 0)
+        ax.annotate(
+            label,
+            center,
+            ha="center",
+            va="center",
+            size=fsize,
+            color="black",
+        )
+
     fig.tight_layout(pad=0)
 
-    plt.savefig(
+    plt.savefig(fname, dpi=dpi, transparent=True, pad_inches=0, bbox_inches="tight")
-        f"{fname}.png", dpi=dpi, transparent=False, pad_inches=0, bbox_inches="tight"
-    )
     plt.close()
 
 
 if __name__ == "__main__":
-    # name = "color_128_0.3_1.00e-06"
     import argparse
     import glob
 
     parser = argparse.ArgumentParser()
-    # make the following accept a list of arguments
     parser.add_argument("-v", "--version", type=int, nargs="+", default=[0])
     parser.add_argument(
         "--dpi", type=int, default=300, help="Resolution for saved image."
     )
+    parser.add_argument(
+        "--studio",
+        type=str,
+        default=["this_studio"],
+        nargs="+",
+        help="Checkpoint studio name.",
+    )
     parser.add_argument("--figsize", type=int, default=6, help="Figure size")
     args = parser.parse_args()
     versions = args.version
-    for v in versions:
+    for studio in args.studio:
-        # name = f"out/v{v}"
-        studio = "colors-refactor-supervised"
-        # studio = "colors-refactor-unsupervised"
-        # studio = "colors-refactor-unsupervised-anchors"
-        # studio = "this_studio"
         Path(studio).mkdir(exist_ok=True, parents=True)
-        name = f"{studio}/v{v}"
+        for v in versions:
-        # ckpt = f"/teamspace/jobs/{name}/work/colors/lightning_logs/version_2/checkpoints/epoch=999-step=8000.ckpt"
+            name = f"{studio}/v{v}"
-        # ckpt_path = f"/teamspace/studios/this_studio/colors/lightning_logs/version_{v}/checkpoints/*.ckpt"
+            # ckpt = f"/teamspace/jobs/{name}/work/colors/lightning_logs/version_2/checkpoints/epoch=999-step=8000.ckpt"
-        ckpt_path = f"/teamspace/studios/{studio}/colors/lightning_logs/version_{v}/checkpoints/*.ckpt"
+            # ckpt_path = f"/teamspace/studios/this_studio/colors/lightning_logs/version_{v}/checkpoints/*.ckpt"
-        ckpt = glob.glob(ckpt_path)
+            ckpt_path = f"/teamspace/studios/{studio}/colors/lightning_logs/version_{v}/checkpoints/*.ckpt"
-        if len(ckpt) > 0:
+            ckpt = glob.glob(ckpt_path)
-            ckpt = ckpt[-1]
+            if len(ckpt) > 0:  # get latest checkpoint
-            print(f"Generating image for checkpoint: {ckpt}")
+                ckpt = ckpt[-1]  # TODO: allow specification via CLI
-            create_circle(
+                print(f"Generating image for checkpoint: {ckpt}")
-                ckpt, fname=name, dpi=args.dpi, figsize=[args.figsize] * 2, roll=False
+                create_circle(
-            )
+                    ckpt,
-        else:
+                    fname=name,
-            print(f"No checkpoint found for version {v}")
+                    dpi=args.dpi,
-        # make_image(ckpt, fname=name + "b", color=False, dpi=args.dpi,)
+                    figsize=[args.figsize] * 2,
+                    roll=False,
+                )
+            else:
+                print(f"No checkpoint found for version {v}")
+            # make_image(ckpt, fname=name + "b", color=False, dpi=args.dpi,)

datamodule.py

@@ -35,14 +35,18 @@ class ColorDataModule(L.LightningDataModule):
         return [(c, cls.get_hue(c)) for c in train_rgb]
 
     @classmethod
-    def get_xkcd_colors(cls):
+    def get_xkcd_colors(cls, label="hues"):
         rgb_tensor, xkcd_color_names = extract_colors()
         rgb_tensor = preprocess_data(rgb_tensor, skip=True)
-        # return [
+        if label == "names":
-        #     (rgb_tensor[i], xkcd_color_names[i].replace("xkcd:", ""))
+            return [
-        #     for i in range(len(rgb_tensor))
+                (rgb_tensor[i], xkcd_color_names[i].replace("xkcd:", ""))
-        # ]
+                for i in range(len(rgb_tensor))
-        return [(c, cls.get_hue(c)) for c in rgb_tensor]
+            ]
+        if label == "hues":
+            return [(c, cls.get_hue(c)) for c in rgb_tensor]
+        else:
+            raise ValueError("Please specify `label` as one of ['hues', 'names'].")
 
     def setup(self, stage: str):
         # Assign train/val datasets for use in dataloaders

makefile

@@ -56,13 +56,15 @@ umap:
 	done
 
 sort_umap:
-	python scripts/sortcolor.py -s umap --dpi 300 --seed 21
+	python baseline.py -s umap --dpi 300 --seed 21
 
 parallel_umap:
-	parallel -j 4 python scripts/sortcolor.py -s umap --dpi 300 --seed ::: $$(seq 1 100)
+	parallel -j 4 python baseline.py -s umap --dpi 300 --seed ::: $$(seq 1 100)
 
 parallel_check:
-	parallel -j 4 python check.py -v ::: $$(seq 0 99)
+	parallel -j 3 python check.py \
+		--studio colors-refactor-unsupervised colors-refactor-supervised colors-refactor-unsupervised-anchors \
+		-v ::: $$(seq 0 99)
 
 sort_lex:
 	python scripts/sortcolor.py -s lex --dpi 300
@@ -70,6 +72,9 @@ sort_lex:
 sort_hsv:
 	python scripts/sortcolor.py -s hsv --dpi 300
 
+poster: sort_lex sort_hsv
+	python scripts/color_poster.py -k hsv lex --rows 119
+
 clean:
 	rm -rf lightning_logs
 	rm -rf .lr_find_*.ckpt

newsearch.py

@@ -1,4 +1,4 @@
-import subprocess
+import subprocess  # noqa: F401
 import sys
 from random import sample, seed
 
@@ -8,17 +8,7 @@ from lightning_sdk import Machine, Studio  # noqa: F401
 # consistency of randomly sampled experiments.
 seed(19920921)
 
-NUM_JOBS = 50
+NUM_JOBS = 20
-
-# reference to the current studio
-# if you run outside of Lightning, you can pass the Studio name
-# studio = Studio()
-
-# use the jobs plugin
-# studio.install_plugin("jobs")
-# job_plugin = studio.installed_plugins["jobs"]
-
-# do a sweep over learning rates
 
 # Define the ranges or sets of values for each hyperparameter
 # alpha_values = list(np.round(np.linspace(2, 4, 21), 4))
@@ -29,16 +19,17 @@ learning_rate_values = [1e-3]
 # alpha_values = [0, .25, 0.5, 0.75, 1]  # alpha = 0 is unsupervised. alpha = 1 is supervised.
 alpha_values = [0, 0.1]
 widths = [2**k for k in range(4, 13)]
-depths = [1, 2, 4, 8, 16]
+depths = [1, 2, 4, 8]
+dropouts = [0, 0.5]
 # widths, depths = [512], [4]
 
-batch_size_values = [256]
+batch_size_values = [64, 256]
-max_epochs_values = [420]  # at 12 fps, around 35s
+max_epochs_values = [720 * 3]  # at 12fps 720 frames = 60s
 seeds = list(range(21, 1992))
 optimizers = [
     # "Adagrad",
     "Adam",
-    # "SGD",
+    "SGD",
     # "AdamW",
     # "LBFGS",
     # "RAdam",
@@ -48,7 +39,7 @@ optimizers = [
 
 # Generate all possible combinations of hyperparameters
 all_params = [
-    (alpha, lr, bs, me, s, w, d, opt)
+    (alpha, lr, bs, me, s, w, d, opt, dr)
     for alpha in alpha_values
     for lr in learning_rate_values
     for bs in batch_size_values
@@ -57,6 +48,7 @@ all_params = [
     for w in widths
     for d in depths
     for opt in optimizers
+    for dr in dropouts
 ]
 
 
@@ -67,10 +59,10 @@ search_params = sample(all_params, min(NUM_JOBS, len(all_params)))
 # --trainer.callbacks.init_args.monitor hp_metric \
 
 for idx, params in enumerate(search_params):
-    a, lr, bs, me, s, w, d, opt = params
+    a, lr, bs, me, s, w, d, opt, dr = params
     # cmd = f"cd ~/colors && python main.py --alpha {a} --lr {lr} --bs {bs} --max_epochs {me} --seed {s} --width {w}"
     cmd = f"""
-python newmain.py fit \
+cd ~/colors && python newmain.py fit \
 --seed_everything {s} \
 --data.batch_size {bs} \
 --data.train_size 0 \
@@ -81,6 +73,7 @@ python newmain.py fit \
 --model.bias true \
 --model.loop true \
 --model.transform tanh \
+--model.dropout {dr} \
 --trainer.min_epochs 10 \
 --trainer.max_epochs {me} \
 --trainer.log_every_n_steps 3 \
@@ -92,20 +85,23 @@ python newmain.py fit \
 --optimizer torch.optim.{opt} \
 --optimizer.init_args.lr {lr} \
 --trainer.callbacks+ lightning.pytorch.callbacks.LearningRateFinder
-# --lr_scheduler lightning.pytorch.cli.ReduceLROnPlateau \
-# --lr_scheduler.init_args.monitor hp_metric \
-# --lr_scheduler.init_args.factor 0.05 \
-# --lr_scheduler.init_args.patience 5 \
-# --lr_scheduler.init_args.cooldown 10 \
-# --lr_scheduler.init_args.verbose true
 """
-    # job_name = f"color2_{bs}_{a}_{lr:2.2e}"
+    # --lr_scheduler lightning.pytorch.cli.ReduceLROnPlateau \
-    # job_plugin.run(cmd, machine=Machine.T4, name=job_name)
+    # --lr_scheduler.init_args.monitor hp_metric \
+    # --lr_scheduler.init_args.factor 0.05 \
+    # --lr_scheduler.init_args.patience 5 \
+    # --lr_scheduler.init_args.cooldown 10 \
+    # --lr_scheduler.init_args.verbose true
     print(f"Running {params}: {cmd}")
     try:
+        studio = Studio("colors-animate-jobs")
+        studio.install_plugin("jobs")
+        job_plugin = studio.installed_plugins["jobs"]
+        job_name = f"hour-colors-20240303-{idx+1}"
+        job_plugin.run(cmd, machine=Machine.T4, name=job_name)
+
         # Run the command and wait for it to complete
-        # subprocess.run(test_cmd, shell=True, check=True)
+        # subprocess.run(cmd, shell=True, check=True)
-        subprocess.run(cmd, shell=True, check=True)
     except KeyboardInterrupt:
         print("Interrupted by user")
         sys.exit(1)

color_poster.py → scripts/color_poster.py

@@ -3,9 +3,14 @@ from typing import List
 import matplotlib.colors as mcolors
 import matplotlib.pyplot as plt
 import numpy as np
+from matplotlib.font_manager import FontProperties
+
+pt_mono_font = FontProperties(
+    fname="/usr/share/fonts/truetype/PT_Mono/PTMono-Regular.ttf"
+)
 
 # set font by default to courier new
-plt.rcParams["font.family"] = "PT Mono"
+# plt.rcParams["font.family"] = "PT Mono"
 # # sort by the proximity to the colors in viridis
 # # Importing necessary functions
 
@@ -63,6 +68,7 @@ def create_color_calibration_image(
             color.replace("xkcd:", ""),
             va="center",
             fontsize=fontsize,
+            fontproperties=pt_mono_font,
             # bbox=dict(facecolor='gray', alpha=0.21),
         )
         ax.text(
@@ -71,6 +77,7 @@ def create_color_calibration_image(
             f"{hex_code}\n({rgb_triple})",
             va="center",
             fontsize=6,
+            fontproperties=pt_mono_font,
             # bbox=dict(facecolor='gray', alpha=0.33),
         )
         # ax.text(
@@ -79,6 +86,7 @@ def create_color_calibration_image(
         #     f"{hex_code}",
         #     va="center",
         #     fontsize=fontsize * 0.6,
+        #     fontproperties=pt_mono_font,
         # )
 
         # Draw color square
@@ -160,14 +168,14 @@ if __name__ == "__main__":
         "--rows", type=int, default=73, help="Number of entries per column"
     )
     parser.add_argument(
-        "--dir", type=str, default="/Volumes/TMP/tests", help="Directory to save images"
+        "--dir", type=str, default="~/out_sortcolors", help="Directory to save images"
     )
     parser.add_argument(
         "-k",
         "--kind",
         type=str,
         nargs="+",
-        default=["hsv", "lex", "lab", "umap"],
+        default=["hsv", "lex"],
         help="Kinds of sorting",
     )
     parser.add_argument(
@@ -201,7 +209,7 @@ if __name__ == "__main__":
     # KIND = "hsv"  # choose from umap, hsv
     for KIND in KINDS:
         colors = list(mcolors.XKCD_COLORS.keys())
-        sorted_indices = np.load(f"scripts/{KIND}_sorted_indices.npy")
+        sorted_indices = np.load(f"/teamspace/studios/this_studio/out_sortcolors/{KIND}/sorted_indices.npy")
         sorted_colors = [colors[idx] for idx in sorted_indices]
 
         colors = sorted_colors

scripts/grab.sh

@@ -0,0 +1,23 @@
+#!/bin/bash
+
+# Define the function to run the scp command for a specific value of j
+run_scp() {
+    # studio id
+    id=$1
+    # version number
+    j=$2
+
+    # job prefix
+    job=hour-colors-20240303-${j}
+    # name of file (locally)
+    out=hour-a${j}.mp4
+    # name of file (remotely)
+    input=colors/lightning_logs/version_0/a0.mp4
+    scp ${id}@ssh.lightning.ai:/teamspace/jobs/${job}/work/${input} ${out}
+}
+
+# Export the function so that it can be used by parallel
+export -f run_scp
+
+# Run scp command in parallel with 2 threads for j in 1 and 2
+parallel -j 2 run_scp ::: {1..3}

scripts/install_ptmono.sh

@@ -0,0 +1,44 @@
+#!/bin/bash
+
+# Source: https://blog.programster.org/ubuntu-install-pt-mono-font
+# Installs PT Mono font onto Ubuntu 12.04 or 14.04 Systems
+
+# Check to make sure the user has the unzip package installed
+export NO_UNZIP=$(apt-cache policy unzip | grep "Installed: (none)" | wc -l)
+
+# Result will be 1 if it is NOT installed
+if [ "$NO_UNZIP" = "0" ]; then
+
+    export TEMP_DIR='temp-technostu-script'
+    cd ~
+    mkdir $TEMP_DIR
+    cd $TEMP_DIR
+
+    # Download PT mono from google fonts
+    export FONT_URL="http://www.google.com/fonts/download"
+    export FONT_URL="$FONT_URL?kit=7qsh9BNBJbZ6khIbS3ZpfKCWcynf_cDxXwCLxiixG1c"
+    wget --content-disposition "$FONT_URL"
+
+    # Create a PT_Mono directory which we will copy across into the fonts directory.
+    mkdir PT_Mono
+    mv PT_Mono.zip PT_Mono/.
+    cd PT_Mono
+    unzip PT_Mono.zip
+    rm PT_Mono.zip
+    cd ..
+    sudo mv PT_Mono /usr/share/fonts/truetype/.
+
+    # Re-cache the fonts
+    echo 'Re-caching fonts...'
+    sudo fc-cache -fv
+
+    # cleanup
+    cd ~
+    sudo rm -rf $TEMP_DIR
+
+    echo 'done!'
+else
+    # User doesnt have unzip installed, tell them how to install it
+    echo 'You need to install unzip for this to work: try '
+    echo '"sudo apt-get install unzip"'
+fi

scripts/sortcolor.py

@@ -82,26 +82,38 @@ def peano_curve(n):
 
 
 if KIND in ("lex", "alpha", "abc"):
+    KIND = "lex"
     preds = np.array(colors)
 
 elif KIND == "umap":
-    # from umap import UMAP
+    PDIR = f"/teamspace/studios/this_studio/out_sortcolors/{KIND}"
-    from cuml import UMAP
+    Path(PDIR).mkdir(parents=True, exist_ok=True)
-
+    file_path = f"{PDIR}/{SEED:06d}.npy"
-    # Use UMAP to create a 1D representation
+    if Path(file_path).exists():
-    reducer = UMAP(
+        print(f"Loading {file_path}")
-        n_components=1,
+        preds = np.load(file_path)
-        n_neighbors=250,
+
-        min_dist=1e-2,
+    else:
-        metric="euclidean",
+        # from umap import UMAP
-        random_state=SEED,
+        from cuml import UMAP  # not fully deterministic.
-        negative_sample_rate=2,
+
-    )
+        # Use UMAP to create a 1D representation
-    embedding = reducer.fit_transform(np.array(rgb_values))
+        reducer = UMAP(
+            n_components=1,
+            n_neighbors=250,
+            min_dist=1e-2,
+            metric="euclidean",
+            random_state=SEED,
+            negative_sample_rate=2,
+        )
+        embedding = reducer.fit_transform(np.array(rgb_values))
 
-    # Sort colors by the 1D representation
+        # Sort colors by the 1D representation
-    preds = embedding[:, 0]
+        preds = embedding[:, 0]
-    del reducer, embedding
+        del reducer, embedding
+    # Save the preds to disk
+    print(f"Saving {file_path}")
+    np.save(file_path, preds.ravel())
 
 elif KIND in ("cielab", "lab", "ciede2000"):
     from skimage.color import deltaE_ciede2000, rgb2lab
@@ -135,18 +147,15 @@ elif KIND == "hsv":
 else:
     raise ValueError(f"Unknown kind: {KIND}")
 
-sorted_indices = np.argsort(preds)
 
-# Save the sorted indices to disk
+PDIR = f"/teamspace/studios/this_studio/out_sortcolors/{KIND}"
-# if (KIND == "umap") or (KIND != "umap"):
-PDIR = f"scripts/{KIND}"
 Path(PDIR).mkdir(parents=True, exist_ok=True)
-file_path = f"{PDIR}/{SEED:06d}.npy"
+file_path = f"{PDIR}/sorted_indices.npy"
-np.save(file_path, preds.ravel())
-print(f"Predictions saved to {file_path}")
-
 # Sort colors by the 1D representation
+sorted_indices = np.argsort(preds)
 sorted_colors = [colors[i] for i in sorted_indices]
+print(f"Saving {file_path}")
+np.save(file_path, sorted_indices.ravel())
 
 # # Display the sorted colors around the ring of a circle
 # # Create a new figure for the circle visualization
@@ -215,9 +224,9 @@ def plot_preds(
     rgb_values,
     fname: str,
     roll: bool = False,
-    dpi: int = 150,
+    radius: float = 1 / 3,
-    inner_radius: float = 1 / 3,
+    dpi: int = 300,
-    figsize=(3, 3),
+    figsize=(6, 6),
 ):
     sorted_inds = np.argsort(preds.ravel())
     colors = rgb_values[sorted_inds, :3]
@@ -261,12 +270,11 @@ def plot_preds(
     ax.set_yticks([])
     ax.set_aspect("equal")
     ax.axis("off")
-    radius = 1
+    ax.set_ylim(0, 1)
-    ax.set_ylim(0, radius)
 
     # Overlay white circle
     circle = patches.Circle(
-        (0, 0), inner_radius, transform=ax.transData._b, color="white", zorder=2
+        (0, 0), radius, transform=ax.transData._b, color="white", zorder=2
     )
     ax.add_patch(circle)
 
@@ -294,7 +302,7 @@ plot_preds(
     fname,
     roll=False,
     dpi=DPI,
-    inner_radius=INNER_RADIUS,
+    radius=INNER_RADIUS,
     figsize=(SIZE, SIZE),
 )
 print(f"saved {fname}")