Mean Absolute Error

We need to check that the model that we have actuallu works on new data

We get a metric to describe this called Mean Absolute Error (MAE) which just the actual value - predicted.

This is a module in sklearn

from sklearn.metrics import mean_absolute_error

predicted_home_prices = model.predict(X)
mean_absolute_error(y, predicted_home_prices)

However This method above is completely wrong as it uses the training data in the prediction.

This is known as a 'in-sample' score]

This is when you use a sample for training and evaluating

We can better evaluate our data by splitting it between test and validation data.

We do this by excluding data from the model build

from sklearn.model_selection import train_test_split

train_X, val_X, train_y, val_y = train_test_split(X, y, random_state=0)

model= DecisionTreeRegressor()

model.fit(train_X, train_y)

val_predict = model.predict(val_X)

print(mean_absolute_error(val_y, val_predict))

At this point the model would be fully ready for the data.

final_model = DecisionTreeRegressor(max_leaf_nodes = best_tree_size, random_state = 0)

final_model.fit(X, y)