TensorFlow Decision Forest Regression Model Log

General Discussion
, ,
#1

I have a question about the regression model in tutorials (model_7), Cree, entrene y evalúe modelos con TensorFlow Decision Forests

I want to log the accuracy, but the output is blank. I have tried:
model_7.compile(metrics=[“accuracy”])
or
model_7.compile(metrics=[“mae”, “acc”])

After evaluation, the accuracy is vary low:
evaluation = model_7.evaluate(test_ds, return_dict=True)
print()
loss: 0.0000e+00 - mae: 0.9490 - acc: 3.4258e-04

The log has no accuracy data:
logs = model_7.make_inspector().training_logs()

Do I miss something?

Thank you!

#2

The accuracy metric is used for classification (binary and multi-class) tasks. For regression, use one of the regression metrics.

model_7.make_inspector().training_logs() should contains the RMSE metric.

model_7.compile(metrics=[“acc”]) will evaluation a regression output (the scale depends on the label) as a probability. In other words, you will get garbage :).

2 Likes
#3

Thank you for the information, Mathieu!