seed everything

main.py

@@ -1,6 +1,9 @@
 import argparse
+import random
 
+import numpy as np
 import pytorch_lightning as pl
+import torch
 from pytorch_lightning.callbacks import EarlyStopping
 
 from callbacks import SaveImageCallback
@@ -29,6 +32,7 @@ def parse_args():
     parser.add_argument(
         "-L", "--log_every_n_steps", type=int, default=5, help="Logging frequency"
     )
+    parser.add_argument("--seed", default=21, type=int, help="Seed")
     parser.add_argument(
         "-w",
         "--num_workers",
@@ -42,8 +46,21 @@ def parse_args():
     return args
 
 
+def seed_everything(seed=42):
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    torch.cuda.manual_seed(seed)
+    torch.cuda.manual_seed_all(seed)  # if you are using multi-GPU.
+    torch.backends.cudnn.deterministic = True
+    torch.backends.cudnn.benchmark = False
+
+
 if __name__ == "__main__":
     args = parse_args()
+
+    seed_everything(args.seed)
+
     early_stop_callback = EarlyStopping(
         monitor="hp_metric",  # Metric to monitor
         min_delta=1e-5,  # Minimum change in the monitored quantity to qualify as an improvement
@@ -53,7 +70,7 @@ if __name__ == "__main__":
     )
 
     save_img_callback = SaveImageCallback(
-        save_interval=1,
+        save_interval=0,
         final_dir="out",
     )
 
@@ -77,6 +94,7 @@ if __name__ == "__main__":
 
     # Initialize trainer with parsed arguments
     trainer = pl.Trainer(
+        deterministic=True,
         callbacks=[early_stop_callback, save_img_callback],
         max_epochs=args.max_epochs,
         log_every_n_steps=args.log_every_n_steps,

makefile

@@ -13,4 +13,8 @@ animate:
 	ffmpeg -i lightning_logs/version_258/e%04d.png \
 		-c:v libx264 \
 		-vf "fps=12,format=yuv420p,pad=ceil(iw/2)*2:ceil(ih/2)*2" \
-		~/animated.mp4
+		~/animated.mp4
+
+clean:
+	rm -rf lightning_logs/*
+	rm out/*.png

search.py

@@ -18,19 +18,21 @@ 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, 6, 41), 4))
+alpha_values = list(np.round(np.linspace(2, 4, 21), 4))
 # learning_rate_values = list(np.round(np.logspace(-5, -3, 41), 5))
 learning_rate_values = [5e-4]
-batch_size_values = [64, 128]
+batch_size_values = [128]
 max_epochs_values = [500]
+seeds = list(range(21, 1992))
 
 # Generate all possible combinations of hyperparameters
 all_params = [
-    (alpha, lr, bs, me)
+    (alpha, lr, bs, me, s)
     for alpha in alpha_values
     for lr in learning_rate_values
     for bs in batch_size_values
     for me in max_epochs_values
+    for s in seeds
 ]
 
 
@@ -38,8 +40,8 @@ all_params = [
 search_params = sample(all_params, min(NUM_JOBS, len(all_params)))
 
 for idx, params in enumerate(search_params):
-    a, lr, bs, me = params
-    cmd = f"cd ~/colors && python main.py --alpha {a} --lr {lr} --bs {bs} --max_epochs {me}"
+    a, lr, bs, me, s = params
+    cmd = f"cd ~/colors && python main.py --alpha {a} --lr {lr} --bs {bs} --max_epochs {me} --seed {s}"
     # job_name = f"color2_{bs}_{a}_{lr:2.2e}"
     # job_plugin.run(cmd, machine=Machine.T4, name=job_name)
     print(f"Running {params}: {cmd}")