File size: 2,895 Bytes
ad97490
 
 
 
fc3303a
8c0f157
fc3303a
8c0f157
fc3303a
 
 
8c0f157
fc3303a
8c0f157
fc3303a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8c0f157
 
 
 
fc3303a
 
 
 
 
 
8c0f157
fc3303a
 
 
 
8c0f157
fc3303a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ad97490
 
 
 
 
 
 
 
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
---
tags:
- ml-intern
---
# Item Weight Predictor

A machine learning model that predicts the weight (in grams) of items based on their text description. Designed for food delivery platforms to estimate weights for:

- **Menu items** (restaurant food portions)
- **Grocery items** (packaged food from supermarkets)
- **Non-food items** (household goods, electronics, toys, etc.)

## Model Architecture

The model uses **per-item-type Ridge regression** on TF-IDF features of text descriptions:

- Separate models trained for each item category (`menu_item`, `grocery`, `non_food`)
- Log-transformed target to handle wide weight range (grams to kilograms)
- TF-IDF with 1-3 gram features for robust text representation

## Performance

| Item Type  | MAE    | MAPE   | Training Samples | Validation Samples |
|------------|--------|--------|------------------|--------------------|
| Grocery    | 23.1g  | 4.4%   | 17,246           | 3,044              |
| Menu Item  | 63.1g  | 136.9% | 6,786            | 1,198              |
| Non-Food   | 479.9g | 113.6% | 7,152            | 1,263              |

Grocery items perform best because most are standardized to ~100g servings. Menu items and non-food have higher variance but still useful for rough estimates.

## Usage

```python
import joblib
from huggingface_hub import hf_hub_download

# Download and load model
model_path = hf_hub_download(repo_id="ZZandro/weight-predictor", filename="unified_predictor.pkl")
predictor = joblib.load(model_path)

# Predict weight
text = "[MENU_ITEM] grilled chicken breast | food | meal ingredient | portion"
weight_g = predictor.predict(text, item_type="menu_item")
print(f"Predicted weight: {weight_g:.1f}g")
```

### Supported Item Types

- `menu_item` - Restaurant menu items (e.g., "cheeseburger", "caesar salad")
- `grocery` - Packaged grocery products (e.g., "chocolate bar", "cereal box")
- `non_food` - General retail items (e.g., "laptop", "t-shirt", "toy car")

### Text Format

Include the item type tag at the start:
```
[MENU_ITEM] <item description>
[GROCERY] <item description>
[NON_FOOD] <item description>
```

## Data Sources

The training data combines:
- Amazon product data with shipping weights (non-food items)
- NutriBench meal descriptions with per-ingredient gram weights (menu items)
- USDA Foundation Food data with standard serving sizes (grocery items)

## Dataset

The processed training dataset is available at: [ZZandro/item-weight-dataset](https://huggingface.co/datasets/ZZandro/item-weight-dataset)

## License

Apache-2.0

<!-- ml-intern-provenance -->
## Generated by ML Intern

This dataset repository was generated by [ML Intern](https://github.com/huggingface/ml-intern), an agent for machine learning research and development on the Hugging Face Hub.

- Try ML Intern: https://smolagents-ml-intern.hf.space
- Source code: https://github.com/huggingface/ml-intern