Skip to content

Commit 6dbc8eb

Browse files
author
M.Notter
committed
Adds 2023 ML intro blog posts
1 parent a5cc27e commit 6dbc8eb

5 files changed

+51
-43
lines changed

‎_posts/2023-12-15-01_scikit_simple.md ‎_posts/2023-10-23-01_scikit_simple.md

+3-3
Original file line numberDiff line numberDiff line change
@@ -14,7 +14,7 @@ The series consists of four parts:
1414
3. **[Advanced Machine Learning with Scikit-learn]({{ site.baseurl }}/blog/2023/03_scikit_advanced)** (Part 3)<br>Exploring complex regression problems and model optimization
1515
4. **[Advanced Neural Networks with TensorFlow]({{ site.baseurl }}/blog/2023/04_tensorflow_advanced)** (Part 4)<br>Implementing sophisticated neural network architectures
1616

17-
## Why These Tools?
17+
### Why These Tools?
1818

1919
[Scikit-learn](https://scikit-learn.org/stable/) is Python's most popular machine learning library for a reason. It provides:
2020
- A consistent interface across different algorithms
@@ -408,7 +408,7 @@ We've seen how Scikit-learn's consistent API makes it easy to experiment with di
408408

409409
In the next post, we'll tackle the same MNIST classification problem using TensorFlow, introducing neural networks and deep learning concepts. This will help you understand the differences between classical machine learning approaches and deep learning, and when to use each.
410410

411-
Key takeaways:
411+
**Key takeaways:**
412412
1. Even simple models can achieve good performance on well-structured problems
413413
2. Start with simple models and gradually increase complexity
414414
3. Cross-validation is crucial for reliable performance estimation
@@ -418,4 +418,4 @@ Key takeaways:
418418

419419
In Part 2, we'll explore how neural networks approach the same problem using TensorFlow, introducing deep learning concepts and comparing the two approaches.
420420

421-
[Continue to Part 2: Basic Neural Networks with TensorFlow →]({{ site.baseurl }}/blog/2023/02_tensorflow_simple)
421+
[Continue to Part 2 →]({{ site.baseurl }}/blog/2023/02_tensorflow_simple)

‎_posts/2023-12-15-02_tensorflow_simple.md ‎_posts/2023-10-23-02_tensorflow_simple.md

+6-5
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@ description: Building your first neural network for image classification
88

99
In this second part of our machine learning series, we'll implement the same MNIST classification task using [TensorFlow](https://www.tensorflow.org/). While Scikit-learn excels at classical machine learning, TensorFlow shines when building neural networks. We'll see how deep learning approaches differ from traditional methods and learn the basic concepts of neural network architecture.
1010

11-
## Why Neural Networks?
11+
### Why Neural Networks?
1212

1313
While our Scikit-learn models performed well in Part 1, neural networks offer several key advantages for image classification:
1414
- **Automatic feature learning**: No need to manually engineer features
@@ -291,7 +291,9 @@ plt.title("Confusion matrix")
291291
plt.show()
292292
```
293293

294-
<img class="img-fluid rounded z-depth-1" src="{{ site.baseurl }}/assets/ex_plots/ex_03_tensorflow_simple_output_22_0.png" data-zoomable width=800px style="padding-top: 20px; padding-right: 20px; padding-bottom: 20px; padding-left: 20px">
294+
<div style="text-align: center">
295+
<img class="img-fluid rounded z-depth-1" src="{{ site.baseurl }}/assets/ex_plots/ex_03_tensorflow_simple_output_22_0.png" data-zoomable width=500px style="padding-top: 20px; padding-right: 20px; padding-bottom: 20px; padding-left: 20px">
296+
</div><br>
295297

296298
## 5. Model parameters
297299

@@ -329,7 +331,7 @@ In this tutorial, we've introduced neural networks using TensorFlow:
329331

330332
Our neural network achieved comparable accuracy to our Scikit-learn models (~99%), but this time on images with a higher resoltuion with the potential for even better performance through further optimization.
331333

332-
Key takeaways:
334+
**Key takeaways:**
333335
1. Neural networks can work directly with structured data like images
334336
2. Architecture design is crucial for good performance
335337
3. Training requires careful parameter selection
@@ -338,5 +340,4 @@ Key takeaways:
338340

339341
In Part 3, we'll explore more advanced machine learning concepts using Scikit-learn, focusing on regression problems and complex preprocessing pipelines.
340342

341-
[← Back to Part 1: Getting Started with Scikit-learn]({{ site.baseurl }}/blog/2023/01_scikit_simple)
342-
[Continue to Part 3: Advanced Machine Learning with Scikit-learn →]({{ site.baseurl }}/blog/2023/03_scikit_advanced)
343+
[← Back to Part 1]({{ site.baseurl }}/blog/2023/01_scikit_simple) or [Continue to Part 3 →]({{ site.baseurl }}/blog/2023/03_scikit_advanced)

‎_posts/2023-12-15-03_scikit_advanced.md ‎_posts/2023-10-23-03_scikit_advanced.md

+18-16
Original file line numberDiff line numberDiff line change
@@ -10,7 +10,7 @@ In this third part of our series, we'll explore more sophisticated machine learn
1010

1111
**Note**: The purpose of this post is to highlight the flexibility and capabilities of scikit-learn's advanced features. Therefore, this tutorial focuses on introducing you to those advanced routines rather than creating the optimal regression model.
1212

13-
## Why Advanced Preprocessing?
13+
### Why Advanced Preprocessing?
1414

1515
Real-world data rarely comes in a clean, ready-to-use format. Data scientists often spend more time preparing data than training models. Common challenges include:
1616
- Missing values that need imputation
@@ -429,18 +429,21 @@ df_res = df_res.sort_values('mean_test_score', ascending=False)
429429
df_res.head(10)
430430
```
431431

432-
| | ridge__alpha | numerical__scaler | numerical__polytrans__interaction_only | numerical__polytrans__degree | numerical__imputer_numeric__strategy | numerical__imputer_numeric__add_indicator | numerical__dim_reduction__pca | numerical__dim_reduction__feat_selecter__score_func | numerical__dim_reduction__feat_selecter__k | categorical__imputer_cat__strategy | categorical__imputer_cat__add_indicator | mean_test_score | std_test_score | mean_train_score | std_train_score |
433-
|----:|---------------:|:--------------------|:-----------------------------------------|-------------------------------:|:---------------------------------------|:--------------------------------------------|:--------------------------------|:------------------------------------------------------|---------------------------------------------:|:-------------------------------------|:------------------------------------------|------------------:|-----------------:|-------------------:|------------------:|
434-
| 4 | 10 | PowerTransformer() | True | 1 | constant | False | PCA(n_components=0.99) | <function mutual_info_regression at 0x177e80280> | 5 | most_frequent | True | -0.0910812 | 0.00726406 | -0.0690628 | 0.00763033 |
435-
| 212 | 10 | RobustScaler() | True | 2 | mean | False | drop | <function mutual_info_regression at 0x177e80280> | 100 | most_frequent | False | -0.0914982 | 0.0101425 | -0.0658095 | 0.00502802 |
436-
| 88 | 10 | PowerTransformer() | False | 1 | most_frequent | False | PCA(n_components=0.99) | <function f_regression at 0x177dbac10> | 25 | most_frequent | True | -0.0920796 | 0.00689614 | -0.0685709 | 0.008111 |
437-
| 44 | 10 | PowerTransformer() | True | 1 | median | False | PCA(n_components=0.9) | <function mutual_info_regression at 0x177e80280> | 25 | constant | False | -0.0922996 | 0.00720934 | -0.0685542 | 0.0080486 |
438-
| 28 | 10 | RobustScaler() | False | 2 | constant | True | PCA(n_components=0.99) | <function mutual_info_regression at 0x177e80280> | 100 | most_frequent | True | -0.0942421 | 0.0122098 | -0.0646248 | 0.00574666 |
439-
| 34 | 1 | RobustScaler() | True | 2 | most_frequent | False | drop | <function mutual_info_regression at 0x177e80280> | 100 | most_frequent | False | -0.0947194 | 0.00546636 | -0.0560936 | 0.00559328 |
440-
| 214 | 10 | StandardScaler() | False | 2 | most_frequent | True | PCA(n_components=0.9) | <function mutual_info_regression at 0x177e80280> | 100 | constant | False | -0.0961898 | 0.0107717 | -0.0603564 | 0.00414943 |
441-
| 108 | 1 | PowerTransformer() | False | 1 | constant | True | PCA(n_components=0.9) | <function f_regression at 0x177dbac10> | 25 | constant | True | -0.0968886 | 0.0075747 | -0.0598735 | 0.00662794 |
442-
| 148 | 1 | PowerTransformer() | False | 1 | most_frequent | True | PCA(n_components=0.99) | <function mutual_info_regression at 0x177e80280> | 25 | constant | False | -0.0971127 | 0.00728891 | -0.0597802 | 0.00659613 |
443-
| 136 | 1 | PowerTransformer() | True | 1 | median | True | PCA(n_components=0.99) | <function f_regression at 0x177dbac10> | 5 | most_frequent | False | -0.0971796 | 0.00667626 | -0.0604481 | 0.00637354 |
432+
| :-------------: | :------------------: | :-----------------------------------------: | :-----------------------------: | :-------------------------------------: | :-----------------------------------------: | :--------------------------------: | :-------------------------------------------: | :-------------------------------------: | :----------------------------------------: | :----------------: | :---------------: | :-----------------: | :----------------: |
433+
| ridge alpha | numerical scaler | numerical polytrans interaction_only | numerical polytrans degree | numerical imputer_numeric strategy | numerical imputer_numeric add_indicator | numerical dim_reduction pca | numerical dim_reduction feat_selecter k | categorical imputer_cat strategy | categorical imputer_cat add_indicator | mean_test_score | std_test_score | mean_train_score | std_train_score |
434+
| :-------------: | :------------------: | :-----------------------------------------: | :-----------------------------: | :-------------------------------------: | :-----------------------------------------: | :--------------------------------: | :-------------------------------------------: | :-------------------------------------: | :----------------------------------------: | :----------------: | :---------------: | :-----------------: | :----------------: |
435+
| 10 | PowerTransformer() | True | 1 | constant | False | PCA(n_components=0.99) | 5 | most_frequent | True | -0.0910812 | 0.00726406 | -0.0690628 | 0.00763033 |
436+
| 10 | RobustScaler() | True | 2 | mean | False | drop | 100 | most_frequent | False | -0.0914982 | 0.0101425 | -0.0658095 | 0.00502802 |
437+
| 10 | PowerTransformer() | False | 1 | most_frequent | False | PCA(n_components=0.99) | 25 | most_frequent | True | -0.0920796 | 0.00689614 | -0.0685709 | 0.008111 |
438+
| 10 | PowerTransformer() | True | 1 | median | False | PCA(n_components=0.9) | 25 | constant | False | -0.0922996 | 0.00720934 | -0.0685542 | 0.0080486 |
439+
| 10 | RobustScaler() | False | 2 | constant | True | PCA(n_components=0.99) | 100 | most_frequent | True | -0.0942421 | 0.0122098 | -0.0646248 | 0.00574666 |
440+
| 1 | RobustScaler() | True | 2 | most_frequent | False | drop | 100 | most_frequent | False | -0.0947194 | 0.00546636 | -0.0560936 | 0.00559328 |
441+
| 10 | StandardScaler() | False | 2 | most_frequent | True | PCA(n_components=0.9) | 100 | constant | False | -0.0961898 | 0.0107717 | -0.0603564 | 0.00414943 |
442+
| 1 | PowerTransformer() | False | 1 | constant | True | PCA(n_components=0.9) | 25 | constant | True | -0.0968886 | 0.0075747 | -0.0598735 | 0.00662794 |
443+
| 1 | PowerTransformer() | False | 1 | most_frequent | True | PCA(n_components=0.99) | 25 | constant | False | -0.0971127 | 0.00728891 | -0.0597802 | 0.00659613 |
444+
| 1 | PowerTransformer() | True | 1 | median | True | PCA(n_components=0.99) | 5 | most_frequent | False | -0.0971796 | 0.00667626 | -0.0604481 | 0.00637354 |
445+
446+
<br>
444447

445448
If you explore this table a bit you can better judge which parameter variations in your grid search are
446449
actually useful and which ones aren't. In this example we will not focus on this and directly continue with
@@ -607,7 +610,7 @@ In this tutorial, we've covered advanced scikit-learn concepts:
607610
- Model comparison and evaluation
608611
- Analyzing feature importance
609612

610-
Key takeaways:
613+
**Key takeaways:**
611614
1. Preprocessing pipelines make complex workflows manageable
612615
2. Grid search helps find optimal parameters systematically
613616
3. Feature selection can improve model performance
@@ -616,5 +619,4 @@ Key takeaways:
616619

617620
In Part 4, we'll explore advanced neural network architectures with TensorFlow, building on both the neural network concepts from Part 2 and the preprocessing techniques we've learned here.
618621

619-
[← Back to Part 2: Basic Neural Networks with TensorFlow]({{ site.baseurl }}/blog/2023/02_tensorflow_simple)
620-
[Continue to Part 4: Advanced Neural Networks with TensorFlow →]({{ site.baseurl }}/blog/2023/04_tensorflow_advanced)
622+
[← Back to Part 2]({{ site.baseurl }}/blog/2023/02_tensorflow_simple) or [Continue to Part 4 →]({{ site.baseurl }}/blog/2023/04_tensorflow_advanced)

0 commit comments

Comments
 (0)