Upload training_metadata.json with huggingface_hub
Browse files- training_metadata.json +52 -0
training_metadata.json
ADDED
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"model_info": {
|
3 |
+
"name": "conservative_medical_phase2_20250724_024052",
|
4 |
+
"type": "conservative_few_shot_prototypical",
|
5 |
+
"phase": 2,
|
6 |
+
"base_model": "emilyalsentzer/Bio_ClinicalBERT"
|
7 |
+
},
|
8 |
+
"performance": {
|
9 |
+
"top_1_accuracy": 0.463,
|
10 |
+
"top_3_accuracy": "~0.52",
|
11 |
+
"top_5_accuracy": "~0.54",
|
12 |
+
"baseline_improvement": "+6.8pp",
|
13 |
+
"baseline_accuracy": 0.395
|
14 |
+
},
|
15 |
+
"architecture": {
|
16 |
+
"embedding_dim": 768,
|
17 |
+
"pooling_strategy": "cls_plus_mean",
|
18 |
+
"prototype_method": "ensemble_attention",
|
19 |
+
"loss_function": "simple_triplet"
|
20 |
+
},
|
21 |
+
"optimizations": {
|
22 |
+
"phase_1": "larger_embeddings_enhanced_pooling",
|
23 |
+
"phase_2": "ensemble_prototypes_attention_aggregation",
|
24 |
+
"conservative_approach": true,
|
25 |
+
"stable_training": true
|
26 |
+
},
|
27 |
+
"deployment": {
|
28 |
+
"ready": true,
|
29 |
+
"requires_dataset": false,
|
30 |
+
"self_contained": true,
|
31 |
+
"num_medical_codes": "~57000"
|
32 |
+
},
|
33 |
+
"usage": {
|
34 |
+
"input": "clinical_notes_text",
|
35 |
+
"output": "top_k_medical_codes",
|
36 |
+
"recommended_top_k": [
|
37 |
+
1,
|
38 |
+
3,
|
39 |
+
5,
|
40 |
+
10
|
41 |
+
],
|
42 |
+
"inference_method": "prototype_similarity"
|
43 |
+
},
|
44 |
+
"training": {
|
45 |
+
"epochs": 3,
|
46 |
+
"stable_loss": true,
|
47 |
+
"no_collapse": true,
|
48 |
+
"incremental_improvement": true
|
49 |
+
},
|
50 |
+
"timestamp": "20250724_024052",
|
51 |
+
"version": "2.0"
|
52 |
+
}
|