agh92 / courses-google-ml-crash-course

Code and exercises of the Machine Learning Crash Course

google-ml-crash-course

Code and exercises from Google´s Machine Learning Crash Course

Directories:

• data: Contains training and test data for the examples
• exercises: Contains all programming exercises of the course
• functions: Python package containing code that is reusable across the examples.

Steps to create a model

1. Load data as pandas DataFrame
2. Define the input features as pandas Series
3. Define the targets as pandas Series
4. Define an optimizer from tensorflow.train to reduce loss
5. Define the type of the model as tf.estimator
6. Define the input function that provides de estimator with a tuple of features and labels
7. Train the model
8. Make Predictions
9. Compute loss using tf.metrics or sklearn.metrics helper functions

Clipping Data

1. Analyse distribution to find outliers
2. Use Series.apply to replace or cut outliers out
3. Repeat steps 7 - 9 from above

Validation Strategy

1. Load a train data set
2. Load a validation data set
3. Train the model with the train set and check the loss against the validation set
4. Load a test data set
5. Check the loss against the test set and not validation set