Ревізії Save and load hyperparameters

1 file changed, 9 insertions, 2 deletions

hparams_save-load.py

			@@ -24,8 +24,14 @@ class MyModel:
24	24		hparams = yaml.safe_load(f)
25	25		return cls(**hparams)
26	26
	27	+	def load_checkpoint(self, checkpoint: Path \| str):
	28	+	print(f"Loading checkpoint from {checkpoint}...")
	29	+
27	30		def train(self):
28	31		print("Training...")
	32	+	print("Saving checkpoint...")
	33	+	with open(Path(self.outdir) / "checkpoint.pt", "w") as f:
	34	+	f.write("Hello, world!")
29	35
30	36		def evaluate(self):
31	37		print("Evaluating...")
			@@ -54,5 +60,6 @@ with open(outdir / "hparams.yaml", "w") as f:
54	60		yaml.dump(hparams, f)
55	61
56	62		# Evaluate the model
57		-	model = MyModel.from_config(outdir / "hparams.yaml")
58		-	model.evaluate()
	63	+	new_model = MyModel.from_config(outdir / "hparams.yaml")
	64	+	new_model.load_checkpoint(outdir / "checkpoint.pt")
	65	+	new_model.evaluate()

1 file changed, 58 insertions

hparams_save-load.py(файл створено)

		@@ -0,0 +1,58 @@
1	+	from datetime import datetime
2	+	from pathlib import Path
3	+
4	+	import yaml
5	+
6	+
7	+	# Define the model
8	+	class MyModel:
9	+	def __init__(
10	+	self,
11	+	n_layers: int,
12	+	n_heads: int,
13	+	dropout: float,
14	+	outdir: str \| Path,
15	+	):
16	+	self.n_layers = n_layers
17	+	self.n_heads = n_heads
18	+	self.dropout = dropout
19	+	self.outdir = outdir
20	+
21	+	@classmethod
22	+	def from_config(cls, config: Path \| str):
23	+	with open(config, "r") as f:
24	+	hparams = yaml.safe_load(f)
25	+	return cls(**hparams)
26	+
27	+	def train(self):
28	+	print("Training...")
29	+
30	+	def evaluate(self):
31	+	print("Evaluating...")
32	+
33	+
34	+	# Create the output directory
35	+	now = datetime.now().strftime("%Y%m%d_%H%M%S")
36	+	outdir = Path(f"{now}_runs")
37	+	outdir.mkdir(parents=True, exist_ok=True)
38	+
39	+	# Instantiate the model
40	+	hparams = {
41	+	"n_layers": 3,
42	+	"n_heads": 8,
43	+	"dropout": 0.1,
44	+	"outdir": str(outdir),
45	+	}
46	+
47	+	model = MyModel(**hparams)
48	+
49	+	# Train the model
50	+	model.train()
51	+
52	+	# Save the hyperparameters in the same directory
53	+	with open(outdir / "hparams.yaml", "w") as f:
54	+	yaml.dump(hparams, f)
55	+
56	+	# Evaluate the model
57	+	model = MyModel.from_config(outdir / "hparams.yaml")
58	+	model.evaluate()

Новіше Пізніше