From 87f48dc8eee9f123fbe24406c1d365a410f589e4 Mon Sep 17 00:00:00 2001
From: unanmed <1319491857@qq.com>
Date: Mon, 24 Mar 2025 17:16:57 +0800
Subject: [PATCH] =?UTF-8?q?feat:=20=E5=90=88=E5=B9=B6=E5=8F=AF=E9=80=89?=
 =?UTF-8?q?=E6=98=AF=E5=90=A6=E5=85=81=E8=AE=B8=E9=87=8D=E5=A4=8D=E9=94=AE?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 data/src/merge.ts  | 2 +-
 data/src/review.ts | 4 ++--
 data/src/utils.ts  | 3 ++-
 ginka/train.py     | 8 ++++++++
 4 files changed, 13 insertions(+), 4 deletions(-)

diff --git a/data/src/merge.ts b/data/src/merge.ts
index b9f649c..73681d6 100644
--- a/data/src/merge.ts
+++ b/data/src/merge.ts
@@ -10,6 +10,6 @@ const [output, ...datasets] = process.argv.slice(2);
             return JSON.parse(file) as DatasetMergable<any>;
         })
     );
-    const merged = mergeDataset(...data);
+    const merged = mergeDataset(true, ...data);
     await writeFile(output, JSON.stringify(merged), 'utf-8');
 })();
diff --git a/data/src/review.ts b/data/src/review.ts
index 5ba0df6..989da4a 100644
--- a/data/src/review.ts
+++ b/data/src/review.ts
@@ -26,13 +26,13 @@ function getNum() {
     );
     const targetFile = await readFile(target, 'utf-8');
     const targetData = JSON.parse(targetFile) as DatasetMergable<any>;
-    const merged = mergeDataset(...datas);
+    const merged = mergeDataset(true, ...datas);
     const keys = Object.keys(merged.data);
     const toReview = chooseFrom(keys, n);
     const reviewData: DatasetMergable<any> = {
         datasetId: Math.floor(Math.random() * 1e12),
         data: Object.fromEntries(toReview.map(v => [v, merged.data[v]]))
     };
-    const reviewed = mergeDataset(targetData, reviewData);
+    const reviewed = mergeDataset(false, targetData, reviewData);
     await writeFile(target, JSON.stringify(reviewed), 'utf-8');
 })();
diff --git a/data/src/utils.ts b/data/src/utils.ts
index 0f0dde7..5040106 100644
--- a/data/src/utils.ts
+++ b/data/src/utils.ts
@@ -15,6 +15,7 @@ export interface FloorData {
 }
 
 export function mergeDataset<T>(
+    allowDuplicateKeys: boolean,
     ...datasets: DatasetMergable<T>[]
 ): DatasetMergable<T> {
     if (datasets.length === 1) {
@@ -24,7 +25,7 @@ export function mergeDataset<T>(
     const data: Record<string, T> = {};
     datasets.forEach(v => {
         for (const [key, value] of Object.entries(v.data)) {
-            if (usedKeys.has(key)) {
+            if (usedKeys.has(key) && allowDuplicateKeys) {
                 const dataKey = `${v.datasetId}/${key}`;
                 data[dataKey] = value;
                 usedKeys.add(dataKey);
diff --git a/ginka/train.py b/ginka/train.py
index 96d5d35..97e824e 100644
--- a/ginka/train.py
+++ b/ginka/train.py
@@ -61,12 +61,20 @@ def train():
         if args.load_optim:
             optimizer.load_state_dict(data["optimizer_state"])
         print("Train from loaded state.")
+        
+    else:
+        # 从头开始训练的话，初始时先把 minamo 损失值权重改为 0
+        criterion.weight[0] = 0.0
     
     # 开始训练
     for epoch in tqdm(range(args.epochs)):
         model.train()
         total_loss = 0
         
+        # 从头开始训练的，在第 10 个 epoch 将 minamo 损失值权重改回来
+        if not args.resume and epoch == 10:
+            criterion.weight[0] = 0.5
+        
         for batch in dataloader:
             # 数据迁移到设备
             target = batch["target"].to(device)