Best Claude prompts for Data Scientists

Given a time series dataset with 10,000 observations and 20 features, including temperature, humidity, and wind speed, generate a comprehensive feature engineering strategy to improve the accuracy of a LSTM-based forecasting model. Consider techniques such as differencing, normalization, and spectral decomposition. Provide a Python code snippet to implement the proposed strategy using the Pandas and Scikit-learn libraries. Assume the data is stored in a CSV file named 'time_series_data.csv' and the target variable is 'energy_consumption'.

Develop a model-agnostic interpretability framework to analyze the feature importance and partial dependence plots of a classification model trained on a dataset with 100,000 samples and 50 features. The model is a random forest classifier with an accuracy of 0.85 on the test set. Use techniques such as SHAP values, LIME, and TreeExplainer to generate insights into the model's decision-making process. Provide a detailed report including visualizations and recommendations for model improvement. Assume the model is stored in a Python pickle file named 'classification_model.pkl' and the dataset is stored in a CSV file named 'classification_data.csv'.

Design a Bayesian optimization framework to tune the hyperparameters of a deep neural network trained on a dataset with 50,000 images and 10 classes. The network architecture consists of 5 convolutional layers and 2 fully connected layers. Consider hyperparameters such as learning rate, batch size, and dropout rate. Use a Gaussian process as the surrogate model and the expected improvement acquisition function to guide the search. Provide a Python code snippet to implement the proposed framework using the Optuna library. Assume the dataset is stored in a directory named 'image_data' and the network architecture is defined in a Python file named 'network.py'.

Develop a root cause analysis framework to detect and diagnose model drift in a production environment where a machine learning model is deployed to predict customer churn. The model is trained on a dataset with 100,000 samples and 20 features, and is updated quarterly using a sliding window approach. Consider metrics such as accuracy, precision, and recall to monitor the model's performance over time. Use techniques such as statistical process control and change point detection to identify potential drift points. Provide a detailed report including visualizations and recommendations for model retraining or updating. Assume the model is stored in a Python pickle file named 'churn_model.pkl' and the dataset is stored in a CSV file named 'churn_data.csv'.