Upload fine-tuned food embeddings with improved score distribution

README.md

@@ -4,35 +4,35 @@ tags:
 - sentence-similarity
 - feature-extraction
 - generated_from_trainer
-- dataset_size:143558
+- dataset_size:210328
 - loss:ScaledCosineSimilarityLoss
 base_model: sentence-transformers/all-MiniLM-L6-v2
 widget:
-- source_sentence: Sapote, mamey, raw
+- source_sentence: Whale, bowhead oil
   sentences:
-  - Miracle Noodle Ready To Eat Spaghetti
-  - Beef ribeye, broiled lean
-  - Agave, raw
-- source_sentence: terebralia semistriata
+  - Cheese, American processed with vitamin D
+  - Cashews, dry roasted with salt
+  - Salmon, dried chum
+- source_sentence: acipenser naccarii bonaparte 1836
   sentences:
-  - Lamb, loin boneless lean cooked fast roasted
-  - Beef, top round roast, boneless, lean, choice, cooked
-  - mud whelk
-- source_sentence: cyttus novaezealandiae
+  - acipenser naccarii bonaparte, 1836
+  - butter clam
+  - Wild Rice, raw
+- source_sentence: Granola Bar, Nature Valley Chewy Trail Mix
   sentences:
-  - Yehuda Original Matzo-Style Squares
-  - Beef, outside skirt steak, choice grilled
-  - dory
-- source_sentence: Chocolate Cereal, prepared with water
+  - Sea lion meat, cooked (Alaska Native)
+  - Trail Mix, regular unsalted
+  - Soup, chunky vegetable, reduced sodium
+- source_sentence: Bear Meat, polar raw
   sentences:
-  - Soybeans, sprouted cooked steamed
-  - Granola Cereal, homemade
-  - Sausage, turkey pork and beef, low fat smoked
-- source_sentence: Lamb, Australian leg bottom raw
+  - Tea, tundra herb and Labrador blend
+  - Beef rib, small end, choice, cooked roasted
+  - sudan teak
+- source_sentence: Beef, tenderloin, raw
   sentences:
-  - Lamb, Australian leg whole lean and fat raw
-  - ipil ipil
-  - Whitefish, broad head eyes cheeks and soft bones
+  - Pork tenderloin, raw
+  - Fruit salad, tropical canned in heavy syrup
+  - Lamb leg, whole, raw
 pipeline_tag: sentence-similarity
 library_name: sentence-transformers
 metrics:
@@ -49,10 +49,10 @@ model-index:
       type: validation
     metrics:
     - type: pearson_cosine
-      value: 0.9130932672076298
+      value: 0.8074689102281711
       name: Pearson Cosine
     - type: spearman_cosine
-      value: 0.8292623565808726
+      value: 0.7665455013117164
       name: Spearman Cosine
 ---
 
@@ -106,9 +106,9 @@ from sentence_transformers import SentenceTransformer
 model = SentenceTransformer("jonny9f/food_embeddings4")
 # Run inference
 sentences = [
-    'Lamb, Australian leg bottom raw',
-    'Lamb, Australian leg whole lean and fat raw',
-    'ipil ipil',
+    'Beef, tenderloin, raw',
+    'Pork tenderloin, raw',
+    'Lamb leg, whole, raw',
 ]
 embeddings = model.encode(sentences)
 print(embeddings.shape)
@@ -155,8 +155,8 @@ You can finetune this model on your own dataset.
 
 | Metric              | Value      |
 |:--------------------|:-----------|
-| pearson_cosine      | 0.9131     |
-| **spearman_cosine** | **0.8293** |
+| pearson_cosine      | 0.8075     |
+| **spearman_cosine** | **0.7665** |
 
 <!--
 ## Bias, Risks and Limitations
@@ -177,19 +177,19 @@ You can finetune this model on your own dataset.
 #### Unnamed Dataset
 
 
-* Size: 143,558 training samples
+* Size: 210,328 training samples
 * Columns: <code>sentence_0</code>, <code>sentence_1</code>, and <code>label</code>
 * Approximate statistics based on the first 1000 samples:
   |         | sentence_0                                                                       | sentence_1                                                                       | label                                                           |
   |:--------|:---------------------------------------------------------------------------------|:---------------------------------------------------------------------------------|:----------------------------------------------------------------|
   | type    | string                                                                           | string                                                                           | float                                                           |
-  | details | <ul><li>min: 3 tokens</li><li>mean: 9.23 tokens</li><li>max: 22 tokens</li></ul> | <ul><li>min: 3 tokens</li><li>mean: 9.07 tokens</li><li>max: 24 tokens</li></ul> | <ul><li>min: 0.06</li><li>mean: 0.61</li><li>max: 1.0</li></ul> |
+  | details | <ul><li>min: 3 tokens</li><li>mean: 9.04 tokens</li><li>max: 24 tokens</li></ul> | <ul><li>min: 3 tokens</li><li>mean: 9.19 tokens</li><li>max: 24 tokens</li></ul> | <ul><li>min: 0.38</li><li>mean: 0.72</li><li>max: 1.0</li></ul> |
 * Samples:
-  | sentence_0                     | sentence_1                                | label                           |
-  |:-------------------------------|:------------------------------------------|:--------------------------------|
-  | <code>yerba mate</code>        | <code>ilex paraguariensis st.-hil.</code> | <code>0.6398421867965322</code> |
-  | <code>huauzontle</code>        | <code>lamb's quarters</code>              | <code>0.6517711020762244</code> |
-  | <code>Falafel, homemade</code> | <code>Sofrito sauce, homemade</code>      | <code>0.1312318314535366</code> |
+  | sentence_0                                      | sentence_1                                        | label                            |
+  |:------------------------------------------------|:--------------------------------------------------|:---------------------------------|
+  | <code>Tortilla, plain or mutton sandwich</code> | <code>Roast beef sandwich, plain</code>           | <code>0.42789756059646605</code> |
+  | <code>Lamb rib, cooked roasted</code>           | <code>Lamb, leg shank half, cooked roasted</code> | <code>0.7156221181154251</code>  |
+  | <code>red raspberry plant</code>                | <code>rubus idaeus var. idaeus l.</code>          | <code>0.8826086956521739</code>  |
 * Loss: <code>__main__.ScaledCosineSimilarityLoss</code>
 
 ### Training Hyperparameters
@@ -324,15 +324,20 @@ You can finetune this model on your own dataset.
 ### Training Logs
 | Epoch  | Step | Training Loss | validation_spearman_cosine |
 |:------:|:----:|:-------------:|:--------------------------:|
-| 0.1114 | 500  | 0.0253        | -                          |
-| 0.2229 | 1000 | 0.017         | -                          |
-| 0.3343 | 1500 | 0.0159        | -                          |
-| 0.4457 | 2000 | 0.0152        | -                          |
-| 0.5572 | 2500 | 0.0139        | -                          |
-| 0.6686 | 3000 | 0.0135        | -                          |
-| 0.7800 | 3500 | 0.0134        | -                          |
-| 0.8915 | 4000 | 0.0126        | -                          |
-| 1.0    | 4487 | -             | 0.8293                     |
+| 0.0761 | 500  | 0.0179        | -                          |
+| 0.1521 | 1000 | 0.0067        | -                          |
+| 0.2282 | 1500 | 0.0059        | -                          |
+| 0.3043 | 2000 | 0.0051        | -                          |
+| 0.3803 | 2500 | 0.0048        | -                          |
+| 0.4564 | 3000 | 0.0046        | -                          |
+| 0.5325 | 3500 | 0.0043        | -                          |
+| 0.6086 | 4000 | 0.004         | -                          |
+| 0.6846 | 4500 | 0.0038        | -                          |
+| 0.7607 | 5000 | 0.0037        | -                          |
+| 0.8368 | 5500 | 0.0037        | -                          |
+| 0.9128 | 6000 | 0.0034        | -                          |
+| 0.9889 | 6500 | 0.0033        | -                          |
+| 1.0    | 6573 | -             | 0.7665                     |
 
 
 ### Framework Versions

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1288de054125174b0b50ec2dfea722751b6e1a7be787c03e2771b69dc73c8e73
+oid sha256:31e5917015fa832c3eb1819a92e73c59337a0d7a52b6c19765354e2fbbde837d
 size 90864192