Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Commit 2c2a447

Browse files
piotrlaczkowskipiotrlaczkowski
committed
fix(KDP): Fix linter errors: unused variable and rearranged imports
1 parent 3f91e19 commit 2c2a447

23 files changed

+2770
-951
lines changed

‎docs/advanced/custom-preprocessing.md‎

Lines changed: 689 additions & 0 deletions
Large diffs are not rendered by default.

‎docs/advanced/distribution-aware-encoding.md‎

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -111,7 +111,6 @@ The encoder automatically detects and handles these distribution types:
111111
| `embedding_dim` | int | None | Output dimension for feature projection |
112112
| `add_distribution_embedding` | bool | False | Add learned distribution type embedding |
113113
| `epsilon` | float | 1e-6 | Small value to prevent numerical issues |
114-
| `auto_detect` | bool | True | Automatically detect distribution type |
115114
| `transform_type` | str | "auto" | Type of transformation to apply |
116115

117116
## 🎯 Best Practices

‎docs/advanced/distribution-aware-testing.md‎

Lines changed: 0 additions & 132 deletions
This file was deleted.

‎docs/advanced/imgs/custom_preprocessing_flow.txt‎

Lines changed: 37 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,37 @@
1+
┌────────────┐ ┌───────────────────────────────────────────────────┐
2+
│ │ │ │
3+
│ Raw Data │──────▶│ Feature Definition │
4+
│ │ │ │
5+
└────────────┘ └─────────────────────┬─────────────────────────┘
6+
│
7+
▼
8+
┌───────────────────────────────────────────────────┐
9+
│ │
10+
│ Custom Preprocessing Pipeline │
11+
│ │
12+
└────────────────┬──────────────────────────────────┘
13+
│
14+
▼
15+
┌──────────────────┬─────────────────────────────┬─────────────────────────┐
16+
│ │ │ │
17+
▼ ▼ ▼ ▼
18+
┌─────────┐ ┌──────────────┐ ┌─────────────────┐ ┌──────────────┐
19+
│ │ │ │ │ │ │ │
20+
│ Layer 1 │────▶│ Layer 2 │──────────▶│ Layer 3 │─────────▶│ Layer N │
21+
│ │ │ │ │ │ │ │
22+
└─────────┘ └──────────────┘ └─────────────────┘ └──────────────┘
23+
24+
┌───────┐ ┌───────┐ ┌───────┐ ┌───────┐
25+
│ │ │ │ │ │ │ │
26+
│ Param │ │ Param │ │ Param │ ..... │ Param │
27+
│ 1 │ │ 2 │ │ 3 │ │ N │
28+
│ │ │ │ │ │ │ │
29+
└───────┘ └───────┘ └───────┘ └───────┘
30+
31+
│
32+
▼
33+
┌───────────────────────────────────────────────────┐
34+
│ │
35+
│ Processed Feature │
36+
│ │
37+
└───────────────────────────────────────────────────┘

‎docs/features/categorical-features.md‎

Lines changed: 15 additions & 49 deletions
Original file line numberDiff line numberDiff line change
@@ -49,24 +49,14 @@ features = {
4949
name="occupation",
5050
feature_type=FeatureType.STRING_CATEGORICAL,
5151
category_encoding=CategoryEncodingOptions.EMBEDDING, # Use embeddings
52-
embedding_dim=16, # Custom embedding dimension
53-
vocabulary_size=1000 # Limit vocabulary size
54-
),
55-
56-
# High-cardinality feature with hashing
57-
"product_id": CategoricalFeature(
58-
name="product_id",
59-
feature_type=FeatureType.STRING_CATEGORICAL,
60-
category_encoding=CategoryEncodingOptions.HASHING, # Use hashing for high cardinality
61-
num_hash_bins=10000, # Number of hash buckets
62-
embedding_dim=32 # Embedding dimension after hashing
52+
embedding_size=16 # Custom embedding size
6353
),
6454

6555
# One-hot encoding for low-cardinality feature
6656
"day_of_week": CategoricalFeature(
6757
name="day_of_week",
6858
feature_type=FeatureType.STRING_CATEGORICAL,
69-
category_encoding=CategoryEncodingOptions.ONE_HOT, # One-hot encoding
59+
category_encoding=CategoryEncodingOptions.ONE_HOT_ENCODING, # One-hot encoding
7060
vocabulary=["Mon", "Tue", "Wed", "Thu", "Fri", "Sat", "Sun"] # Pre-defined vocabulary
7161
)
7262
}
@@ -77,16 +67,13 @@ features = {
7767
| Parameter | Description | Default | Suggested Range |
7868
|-----------|-------------|---------|----------------|
7969
| `feature_type` | Base feature type | Based on data | `STRING_CATEGORICAL`, `INTEGER_CATEGORICAL` |
80-
| `category_encoding` | Encoding method | `EMBEDDING` | `EMBEDDING`, `ONE_HOT`, `HASHING` |
81-
| `embedding_dim` | Dimensionality of embedding | Auto-scaled | 8-128 |
82-
| `vocabulary_size` | Maximum vocabulary size | 10,000 | 100-1,000,000 |
70+
| `category_encoding` | Encoding method | `EMBEDDING` | `EMBEDDING`, `ONE_HOT_ENCODING` |
71+
| `embedding_size` | Dimensionality of embedding | Auto-scaled | 8-128 |
8372
| `vocabulary` | Pre-defined vocabulary | `None` | List of categories |
84-
| `num_hash_bins` | Number of hash buckets | 10,000 | 1,000-100,000 |
85-
| `hash_key` | Hash seed for deterministic hashing | `None` | Integer seed |
8673

8774
## 🔥 Power Features
8875

89-
### Automatic Vocabulary Sizing
76+
### Automatic Embedding Sizing
9077

9178
KDP automatically determines optimal embedding sizes based on cardinality:
9279

@@ -100,41 +87,22 @@ preprocessor = PreprocessingModel(
10087

10188
### Handling High-Cardinality Features
10289

103-
When dealing with millions of categories:
90+
When dealing with many categories, KDP automatically adjusts the embedding size:
10491

10592
```python
106-
# For features with huge numbers of categories
93+
# For features with many categories
10794
preprocessor = PreprocessingModel(
10895
features_specs={
10996
"user_id": CategoricalFeature(
11097
name="user_id",
11198
feature_type=FeatureType.STRING_CATEGORICAL,
112-
category_encoding=CategoryEncodingOptions.HASHING,
113-
num_hash_bins=100000, # Large number of buckets
114-
embedding_dim=64 # Rich representation
99+
category_encoding=CategoryEncodingOptions.EMBEDDING,
100+
# embedding size will be automatically determined based on cardinality
115101
)
116102
}
117103
)
118104
```
119105

120-
### Cross-Category Features
121-
122-
Capture interactions between categorical features:
123-
124-
```python
125-
# Create interactions between categories
126-
preprocessor = PreprocessingModel(
127-
features_specs={
128-
"product_category": FeatureType.STRING_CATEGORICAL,
129-
"user_country": FeatureType.STRING_CATEGORICAL
130-
},
131-
# Define cross features
132-
feature_crosses=[
133-
("product_category", "user_country", 32) # Names and embedding dimension
134-
]
135-
)
136-
```
137-
138106
## 📊 Model Architecture
139107

140108
Below are visualizations of categorical feature processing in KDP:
@@ -156,18 +124,16 @@ For more control, you can use the `CategoricalFeature` class:
156124
## 💡 Pro Tips
157125

158126
1. **Choose the Right Encoding**
159-
- Use `ONE_HOT` for very low cardinality (< 10 categories)
160-
- Use `EMBEDDING` for medium cardinality (10-10,000 categories)
161-
- Use `HASHING` for high cardinality (> 10,000 categories)
127+
- Use `ONE_HOT_ENCODING` for very low cardinality (< 10 categories)
128+
- Use `EMBEDDING` for medium to high cardinality (≥ 10 categories)
162129

163130
2. **Embedding Dimension Rules of Thumb**
164-
- A good starting point: `embedding_dim = 1.6 * num_categories^0.56`
165-
- For very important features, increase this by 50%
166-
- Cap around 512 dimensions even for extremely high cardinality
131+
- KDP automatically calculates optimal embedding size using the rule: `min(500, 1.6 * num_categories^0.56)`
132+
- For very important features, you can override with custom `embedding_size`
167133

168134
3. **Vocabulary Management**
169-
- Limit vocabulary size for memory efficiency
170-
- Consider the "minimum_frequency" parameter to drop rare categories
135+
- For low-cardinality features, consider providing a pre-defined vocabulary
136+
- This ensures consistent encoding across different datasets
171137

172138
4. **Cross Features for Interactions**
173139
- Use cross features when combinations have special meaning

0 commit comments

Comments
 (0)