Remove ordered list indicator

Author: M.Notter

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

@@ -462,7 +462,7 @@ plt.close()
 
 Before wrapping up, let's discuss some important pitfalls to avoid when working on classification tasks:
 
-1. **Data Leakage**: Always split your data before any preprocessing or feature engineering
+**Data Leakage**: Always split your data before any preprocessing or feature engineering
 
 ```python
 # Wrong: Preprocessing before split
@@ -475,7 +475,7 @@ X_tr_scaled = preprocessing.scale(X_tr)
 X_te_scaled = preprocessing.scale(X_te)
 ```
 
-2. **Class Imbalance**: Always check your class distribution
+**Class Imbalance**: Always check your class distribution
 
 ```python
 # Using pandas for better visualization
@@ -493,7 +493,7 @@ plt.xlabel('Class')
 plt.ylabel('Frequency (%)')
 ```
 
-3. **Overfitting**: Monitor these warning signs
+**Overfitting**: Monitor these warning signs
     - Large gap between training and validation scores
     - Perfect training accuracy (like we saw with RandomForest)
     - Poor generalization to new data
@@ -507,7 +507,7 @@ print(f"CV Scores: {scores}")
 print(f"Mean: {scores.mean():.3f} (±{scores.std()*2:.3f})")
 ```
 
-4. **Memory Management**: For large datasets, consider these approaches
+**Memory Management**: For large datasets, consider these approaches
 
 ```python
 # Use n_jobs parameter for parallel processing
@@ -517,7 +517,7 @@ rf = RandomForestClassifier(n_jobs=-1)  # Use all available cores
 rf = RandomForestClassifier(max_samples=0.8)  # Use 80% of samples per tree
 ```
 
-5. **Feature Scaling**: Different algorithms have different scaling requirements
+**Feature Scaling**: Different algorithms have different scaling requirements
 
 ```python
 # SVM requires scaling, Random Forests don't
@@ -532,7 +532,7 @@ X_te_scaled = scaler.transform(X_te)
 rf.fit(X_tr, y_tr)  # No scaling needed
 ```
 
-6. **Model Selection Bias**: Don't use test set for model selection
+**Model Selection Bias**: Don't use test set for model selection
 
 ```python
 # Wrong: Using test set for parameter tuning
@@ -546,7 +546,7 @@ grid.fit(X_tr, y_tr)
 # Only use test set for final evaluation
 ```
 
-7. **Model Troubleshooting Tips**
+**Model Troubleshooting Tips**
 
 ```python
 # Check for data issues first
@@ -564,7 +564,7 @@ if np.any(y_prob > 1.0) or np.any(y_prob < 0.0):
     print("Warning: Invalid probability predictions!")
 ```
 
-8. **Common Error Messages and Solutions**
+**Common Error Messages and Solutions**
    - `ValueError: Input contains NaN`: Clean your data before training
    - `MemoryError`: Reduce batch size or use data generators
 

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

@@ -365,7 +365,7 @@ plt.show()
 ### Common Deep Learning Pitfalls
 When starting with TensorFlow and neural networks, watch out for these common issues:
 
-1. **Data Preparation**
+**Data Preparation**
    - (Almost) always scale input data (like we did with `/255.0`)
    - Check for missing or invalid values
    - Ensure consistent data types
@@ -376,7 +376,7 @@ x_train = x_train.astype('float32') / 255.0
 x_test = x_test.astype('float32') / 255.0
 ```
 
-2. **Model Architecture**
+**Model Architecture**
    - Start simple, add complexity only if needed
    - Match output layer to your task (softmax for classification)
    - Use appropriate layer sizes
@@ -391,7 +391,7 @@ model = keras.Sequential([
 ])
 ```
 
-3. **Training Issues**
+**Training Issues**
    - Monitor training metrics (loss not decreasing)
    - Watch for overfitting (validation loss increasing)
    - Use appropriate batch sizes
@@ -406,7 +406,7 @@ history = model.fit(
 )
 ```
 
-4. **Memory Management**
+**Memory Management**
    - Clear unnecessary variables
    - Use appropriate data types
    - Watch batch sizes on limited hardware

_posts/2023-10-23-04_tensorflow_advanced.md

@@ -677,7 +677,7 @@ plot_history(history_file=history_file, title="Training overview of best model")
 
 When working with complex neural networks and regression tasks, be aware of these advanced challenges:
 
-1. **Gradient Issues**
+**Gradient Issues**
    - Vanishing/exploding gradients in deep networks
    - Unstable training with certain architectures
 
@@ -694,7 +694,7 @@ x = layers.BatchNormalization()(x)
 x = layers.ReLU()(x)
 ```
 
-2. **Learning Rate Dynamics**
+**Learning Rate Dynamics**
    - Static learning rates often suboptimal
    - Different layers may need different rates
 
@@ -720,7 +720,7 @@ def warmup_cosine_decay(step):
     return tf.where(step < warmup_steps, warmup_rate, cosine_rate)
 ```
 
-3. **Complex Loss Functions**
+**Complex Loss Functions**
    - Multiple objectives need careful weighting
    - Custom losses require gradient consideration
    - Handle edge cases and numerical stability
@@ -739,7 +739,7 @@ class WeightedMSE(keras.losses.Loss):
         return tf.reduce_mean(weighted_errors, axis=-1)
 ```
 
-4. **Data Pipeline Bottlenecks**
+**Data Pipeline Bottlenecks**
    - I/O can become training bottleneck
    - Memory constraints with large datasets
 
@@ -757,7 +757,7 @@ def data_generator():
             yield x_tr[i:i+batch_size], y_tr[i:i+batch_size]
 ```
 
-5. **Model Architecture Complexity**
+**Model Architecture Complexity**
    - Deeper isn't always better
    - Skip connections can help with gradient flow
 
@@ -774,7 +774,7 @@ def residual_block(x, filters):
     return layers.ReLU()(x)
 ```
 
-6. **Regularization Strategy**
+**Regularization Strategy**
 - Different layers may need different regularization
 - Combine multiple regularization techniques
 
@@ -789,7 +789,7 @@ x = layers.BatchNormalization()(x)
 x = layers.Dropout(0.5)(x)
 ```
 
-7. **Model Debugging**
+**Model Debugging**
    - Add metrics to monitor internal states
    - Use callbacks for detailed inspection
    - Clear unused variables and models