Spaces:
Sleeping
Sleeping
File size: 7,663 Bytes
a6067aa | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 | from pathlib import Path
import json
import pickle
import numpy as np
import torch
import torch.nn as nn
from torch.utils.data import Dataset, DataLoader
RANDOM_STATE = 42
BATCH_SIZE = 64
EPOCHS = 30
LEARNING_RATE = 1e-3
EARLY_STOPPING_PATIENCE = 3
HIDDEN_DIM = 768
DROPOUT = 0.1
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
torch.manual_seed(RANDOM_STATE)
np.random.seed(RANDOM_STATE)
torch.set_num_threads(1)
class FlatEmbedDataset(Dataset):
def __init__(self, X, y):
self.X = torch.tensor(X, dtype=torch.float32)
self.y = torch.tensor(y, dtype=torch.long)
def __len__(self):
return self.X.shape[0]
def __getitem__(self, idx):
return self.X[idx], self.y[idx]
class FlatEmbedMLP(nn.Module):
def __init__(self, input_dim, n_classes, hidden_dim=HIDDEN_DIM, dropout=DROPOUT):
super().__init__()
self.net = nn.Sequential(
nn.Linear(input_dim, hidden_dim),
nn.ReLU(),
nn.Dropout(dropout),
nn.Linear(hidden_dim, hidden_dim),
nn.ReLU(),
nn.Dropout(dropout),
nn.Linear(hidden_dim, n_classes),
)
def forward(self, x):
return self.net(x)
def topk_accuracy(logits, y, k=5):
k = min(k, logits.shape[1])
topk = torch.topk(logits, k=k, dim=1).indices
hits = topk.eq(y.unsqueeze(1)).any(dim=1).float()
return hits.mean().item()
def evaluate(model, loader, criterion):
model.eval()
total_loss = 0.0
total_n = 0
correct = 0
top5 = 0
with torch.no_grad():
for x, y in loader:
x = x.to(DEVICE)
y = y.to(DEVICE)
logits = model(x)
loss = criterion(logits, y)
batch_n = x.size(0)
total_loss += loss.item() * batch_n
total_n += batch_n
correct += (torch.argmax(logits, dim=1) == y).sum().item()
k = min(5, logits.shape[1])
topk = torch.topk(logits, k=k, dim=1).indices
top5 += topk.eq(y.unsqueeze(1)).any(dim=1).sum().item()
return {
"loss": total_loss / total_n,
"acc_y6": correct / total_n,
"top5_y6": top5 / total_n,
}
def main():
print("entered main", flush=True)
project_dir = Path(__file__).resolve().parents[2]
processed_dir = project_dir / "data" / "processed"
artifacts_dir = project_dir / "training" / "artifacts"
label_maps_dir = artifacts_dir / "label_maps"
embedder_dir = artifacts_dir / "embedder"
models_dir = artifacts_dir / "models"
models_dir.mkdir(parents=True, exist_ok=True)
X_train = np.load(processed_dir / "X_train_embed.npy")
X_valid = np.load(processed_dir / "X_valid_embed.npy")
X_test = np.load(processed_dir / "X_test_embed.npy")
print("loaded X arrays", X_train.shape, X_valid.shape, X_test.shape, flush=True)
y_train_obj = np.load(processed_dir / "y_train_embed.npz")
y_valid_obj = np.load(processed_dir / "y_valid_embed.npz")
y_test_obj = np.load(processed_dir / "y_test_embed.npz")
y_train = y_train_obj["y6"]
y_valid = y_valid_obj["y6"]
y_test = y_test_obj["y6"]
print("loaded y6 arrays", flush=True)
with open(label_maps_dir / "label_maps_embed.pkl", "rb") as f:
label_maps = pickle.load(f)
with open(embedder_dir / "embed_metadata.pkl", "rb") as f:
embed_metadata = pickle.load(f)
input_dim = int(X_train.shape[1])
n_classes = len(label_maps["y6"]["classes"])
train_ds = FlatEmbedDataset(X_train, y_train)
valid_ds = FlatEmbedDataset(X_valid, y_valid)
test_ds = FlatEmbedDataset(X_test, y_test)
train_loader = DataLoader(train_ds, batch_size=BATCH_SIZE, shuffle=True)
valid_loader = DataLoader(valid_ds, batch_size=BATCH_SIZE, shuffle=False)
test_loader = DataLoader(test_ds, batch_size=BATCH_SIZE, shuffle=False)
model = FlatEmbedMLP(
input_dim=input_dim,
n_classes=n_classes,
).to(DEVICE)
optimizer = torch.optim.Adam(
[p for p in model.parameters() if p.requires_grad],
lr=LEARNING_RATE,
)
criterion = nn.CrossEntropyLoss()
best_valid_acc = -1.0
best_epoch = None
epochs_without_improvement = 0
history = []
print("starting training loop", flush=True)
for epoch in range(1, EPOCHS + 1):
print(f"starting epoch {epoch}", flush=True)
model.train()
running_loss = 0.0
total_n = 0
for batch_idx, (x, y) in enumerate(train_loader):
x = x.to(DEVICE)
y = y.to(DEVICE)
optimizer.zero_grad()
logits = model(x)
loss = criterion(logits, y)
loss.backward()
optimizer.step()
batch_n = x.size(0)
running_loss += loss.item() * batch_n
total_n += batch_n
if batch_idx % 50 == 0:
print(f"epoch {epoch} batch {batch_idx} loss {loss.item():.4f}", flush=True)
train_loss = running_loss / total_n
valid_metrics = evaluate(model, valid_loader, criterion)
row = {
"epoch": epoch,
"train_loss": train_loss,
"valid_loss": valid_metrics["loss"],
"valid_acc_y6": valid_metrics["acc_y6"],
"valid_top5_y6": valid_metrics["top5_y6"],
}
history.append(row)
print(
f"Epoch {epoch:02d} | "
f"train_loss={train_loss:.4f} | "
f"valid_loss={valid_metrics['loss']:.4f} | "
f"valid_acc_y6={valid_metrics['acc_y6']:.4f} | "
f"valid_top5_y6={valid_metrics['top5_y6']:.4f}",
flush=True,
)
if valid_metrics["acc_y6"] > best_valid_acc:
best_valid_acc = valid_metrics["acc_y6"]
best_epoch = epoch
epochs_without_improvement = 0
torch.save(model.state_dict(), models_dir / "flat_embed_best.pt")
print("saved new best model", flush=True)
else:
epochs_without_improvement += 1
print(f"no improvement for {epochs_without_improvement} epoch(s)", flush=True)
if epochs_without_improvement >= EARLY_STOPPING_PATIENCE:
print(
f"early stopping triggered after {EARLY_STOPPING_PATIENCE} epochs without improvement",
flush=True,
)
break
print(f"best epoch: {best_epoch}", flush=True)
print(f"best valid_acc_y6: {best_valid_acc:.4f}", flush=True)
model.load_state_dict(torch.load(models_dir / "flat_embed_best.pt", map_location=DEVICE))
print("evaluating test set", flush=True)
test_metrics = evaluate(model, test_loader, criterion)
with open(models_dir / "flat_embed_history.json", "w") as f:
json.dump(history, f, indent=2)
with open(models_dir / "flat_embed_test_metrics.json", "w") as f:
json.dump(test_metrics, f, indent=2)
config = {
"batch_size": BATCH_SIZE,
"epochs": EPOCHS,
"learning_rate": LEARNING_RATE,
"early_stopping_patience": EARLY_STOPPING_PATIENCE,
"hidden_dim": HIDDEN_DIM,
"dropout": DROPOUT,
"device": DEVICE,
"embedder_model_name": embed_metadata["model_name"],
"embedding_dim": embed_metadata["embedding_dim"],
}
with open(models_dir / "flat_embed_config.json", "w") as f:
json.dump(config, f, indent=2)
print("done", flush=True)
print("test metrics:", flush=True)
for k, v in test_metrics.items():
print(f"{k}: {v:.4f}", flush=True)
if __name__ == "__main__":
print("script started", flush=True)
main() |