Creare nuove feature

La feature engineering include anche la creazione vera e propria di nuove feature. Crearle è importante perché il modello si affida a queste feature per ottenere previsioni accurate. In questo esercizio, controllerai le proprietà di tre colonne che appaiono come interi nei dati ma rappresentano valori categorici. Queste tre colonne sono: search_engine_type, product_type e advertiser_type. Creerai feature di conteggio per queste 3 colonne, oltre che per device_id e site_id. Queste feature di conteggio rappresentano il numero di clic per ciascuna di queste colonne e verranno utilizzate più avanti per la previsione.

Il modulo pandas è disponibile come pd nel tuo workspace e il DataFrame di esempio è caricato come df.

Questo esercizio fa parte del corso

Prevedere il CTR con il Machine Learning in Python

Visualizza il corso

Istruzioni dell'esercizio

Stampa il numero totale di valori e il numero di valori unici per ciascuna feature nella lista feature_list.
Crea nuove feature a partire dalle feature in new_feature_list contando il numero di clic per ciascuna feature usando .transform().

Esercizio pratico interattivo

Prova a risolvere questo esercizio completando il codice di esempio.

# Get counts of total and unique values for given features
feature_list = ["search_engine_type", "product_type", "advertiser_type"]
for feature in feature_list:
	print(df[feature].____)
	print(df[feature].____)

# Define new features as counts
new_feature_list = ['device_id', 'site_id'] + feature_list
for new_feature in new_feature_list:
  df[new_feature + '_count'] = df.____(
    new_feature)['click'].____("count")
print(df.head(5))

Modifica ed esegui il codice

Questo esercizio fa parte del corso

Prevedere il CTR con il Machine Learning in Python

IntermediárioNível de habilidade

4.9+

Inizia il corso gratis

Chances are you’re on this page because you clicked a link. In this chapter, you’ll learn why click-through-rates (CTR) are integral to targeted advertising, how to perform basic DataFrame manipulation, and how you can use machine learning models to predict CTR.

Exercise 1: Introduction to click-through rates Exercise 2: Beginning steps Exercise 3: Feature exploration Exercise 4: First evaluation of data Exercise 5: Overview of machine learning models Exercise 6: Logistic regression for breast cancer Exercise 7: Logistic regression for images Exercise 8: A second toy model Exercise 9: CTR prediction using decision trees Exercise 10: Model implementation Exercise 11: A first CTR model Exercise 12: Beyond only accuracy

This chapter provides the foundations for exploratory data analysis (EDA). Using sample data you’ll use the pandas library to look at columns and data types, explore missing data, and use hashing to perform feature engineering on categorical features. All of which are important when exploring features for more accurate CTR prediction.

Exercise 1: Analisi esplorativa dei dati Exercise 2: Un primo sguardo Exercise 3: Verificare i valori mancanti Exercise 4: Distribuzioni per CTR Exercise 5: Feature engineering Exercise 6: Analizzare le colonne datetime Exercise 7: Conversione delle variabili categoriche Exercise 8: Creare nuove feature

Esercizio in corso

Exercise 9: Standardizzare le feature Exercise 10: Normalizzazione logaritmica Exercise 11: Capire la standardizzazione Exercise 12: Standard scaling

It’s time to dive deeper. Find out how you can use measures of model performance including precision and recall to answer real-world questions, such as evaluating ROI on ad spend. You’ll also learn ways to improve upon those evaluation metrics, such as ensemble methods and hyperparameter tuning.

Exercise 1: Applications of metric evaluation Exercise 2: Four categories of outcomes Exercise 3: Evaluating four categories Exercise 4: ROI on ad spend Exercise 5: Model evaluation Exercise 6: Precision and recall Exercise 7: Baseline Exercise 8: Classifier comparison Exercise 9: Tuning models Exercise 10: Regularization Exercise 11: Cross validation Exercise 12: Model selection Exercise 13: Ensembles and hyperparameter tuning Exercise 14: Understanding hyperparameter tuning Exercise 15: Random forests Exercise 16: Grid search

Profits can be heavily impacted by your campaign’s CTR. In this chapter, you’ll learn how deep learning can be used to reduce that risk. You’ll focus on multi-layer perceptron (MLP) and neural network models, and learn how these can be used to capture the complex relationship between variables to more accurately predict CTR. Lastly, you’ll explore how to apply the basics of hyperparameter tuning and regularization to classification models.

Exercise 1: Introduction to deep learning Exercise 2: Understanding MLPs Exercise 3: Beginning model Exercise 4: MLPs for CTR Exercise 5: Hyperparameter tuning in deep learning Exercise 6: Hyperparameter tuning in MLPs Exercise 7: Varying hyperparameters Exercise 8: MLP Grid Search Exercise 9: Model evaluation Exercise 10: F-beta score Exercise 11: Low precision and high AUC Exercise 12: Precision, ROI, and AUC Exercise 13: Model review and comparison Exercise 14: Model comparison warmup Exercise 15: Evaluating precision and ROI Exercise 16: Total scoring Exercise 17: Wrap-up video