更改requirements.txt文件的torch==1.12.0

app.py

@@ -3,8 +3,8 @@ from transformers import AutoTokenizer, AutoModelForSequenceClassification
 import torch
 
 # 加载模型和tokenizer
-tokenizer = AutoTokenizer.from_pretrained("Tokymin/Mood_Anxiety_Disorder_Classify")
-model = AutoModelForSequenceClassification.from_pretrained("Tokymin/Mood_Anxiety_Disorder_Classify", num_labels=8)
+tokenizer = AutoTokenizer.from_pretrained("Tokymin/Mood_Anxiety_Disorder_Classify/model_20240302-214915_lr1e-05_optAdamW_lossBCEWithLogitsLoss_batch16_epoch10.pt")
+model = AutoModelForSequenceClassification.from_pretrained("Tokymin/Mood_Anxiety_Disorder_Classify/model_20240302-214915_lr1e-05_optAdamW_lossBCEWithLogitsLoss_batch16_epoch10.pt", num_labels=8)
 model.eval()
 
 def predict(text):

saved_models/config.json

@@ -0,0 +1,46 @@
+{
+  "_name_or_path": "emilyalsentzer/Bio_ClinicalBERT",
+  "architectures": [
+    "BertForSequenceClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "id2label": {
+    "0": "Anxiety Normal",
+    "1": "Anxiety Mild",
+    "2": "Anxiety Moderate",
+    "3": "Anxiety Major",
+    "4": "Depression Normal",
+    "5": "Depression Mild",
+    "6": "Depression Moderate",
+    "7": "Depression Major"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "label2id": {
+    "Anxiety Normal": 0,
+    "Anxiety Mild": 1,
+    "Anxiety Moderate": 2,
+    "Anxiety Major": 3,
+    "Depression Normal": 4,
+    "Depression Mild": 5,
+    "Depression Moderate": 6,
+    "Depression Major": 7
+  },
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "problem_type": "single_label_classification",
+  "torch_dtype": "float32",
+  "transformers_version": "4.25.1",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 28996
+}

train.py

@@ -107,14 +107,35 @@ for epoch in range(epochs):  # 迭代多个epoch
         loss = loss_fn(logits, b_labels)
         total_eval_loss += loss.item()  # 累加批次损失到总损失
         # 使用sigmoid函数将logits转换为概率值
-        probs = torch.sigmoid(logits)
-        # 将概率高于0.5的预测为正类（1），低于0.5的预测为负类（0）
-        predictions = (probs > 0.5).int()
-        # 比较预测和真实标签
-        correct_predictions = (predictions == b_labels.int()).float()  # 确保标签也是整数类型
-        # 计算每个样本的正确预测的平均数，然后计算整个批次的平均值
-        accuracy_per_sample = correct_predictions.mean(dim=1)
-        accuracy = accuracy_per_sample.mean().item()
+        # probs = torch.sigmoid(logits)
+        # # 将概率高于0.5的预测为正类（1），低于0.5的预测为负类（0）
+        # predictions = (probs > 0.5).int()
+        # # 比较预测和真实标签
+        # correct_predictions = (predictions == b_labels.int()).float()  # 确保标签也是整数类型
+        # # 计算每个样本的正确预测的平均数，然后计算整个批次的平均值
+        # accuracy_per_sample = correct_predictions.mean(dim=1)
+        # accuracy = accuracy_per_sample.mean().item()
+
+        logits_sas = logits[:, :4]  # SAS_Class的4个输出
+        logits_sds = logits[:, 4:]  # SDS_Class的4个输出
+        # 应用softmax来获取概率分布
+        probs_sas = torch.softmax(logits_sas, dim=1)
+        probs_sds = torch.softmax(logits_sds, dim=1)
+
+        # 选择概率最高的类别作为预测结果
+        _, predictions_sas = torch.max(probs_sas, dim=1)
+        _, predictions_sds = torch.max(probs_sds, dim=1)
+
+        # 真实的标签
+        true_sas = b_labels[:, 0].long()  # 确保是长整型
+        true_sds = b_labels[:, 1].long()  # 确保是长整型
+
+        # 计算准确性
+        accuracy_sas = (predictions_sas == true_sas).float().mean()
+        accuracy_sds = (predictions_sds == true_sds).float().mean()
+
+        # 综合两个准确性得分
+        accuracy = (accuracy_sas + accuracy_sds) / 2
         total_eval_accuracy += accuracy
         # 更新进度条
         eval_progress_bar.set_postfix({'accuracy': f"{accuracy:.2f}"})