Migrating to YDF¶

YDF is Google's new library to train Decision Forests and the successor of TensorFlow Decision Forests.

Both libraries rely on the same high-performance C++ implementation called YDF C++ which have been developed for production since 2018. However, YDF is significantly more feature-rich, efficient, and easier to use than TF-DF. Migrating to YDF will reduce your development and maintenance cost while potentially improving the quality of your model.

Most functions in TF-DF have their equivalent in YDF. The following table shows the mapping.

Note: Many features / functions in YDF do not exist in TF-DF. Therefore, reading the YDF's Getting started guide is likely a good investment of your time.

Note: The migration message in TF-DF can be removed by setting environment variable TFDF_DISABLE_WELCOME_MESSAGE.

Action	TF-DF	YDF
Train a model	tf_ds = tfdf.keras.pd_dataframe_to_tf_dataset(ds, label="l") model = tfdf.keras.RandomForestModel() model.fit(tf_ds)	model = ydf.GradientBoostedTreesLearner(label="l").train(ds)
Look at a model	model.summary()	model.describe()
Evaluatea a model	model.compile(["accuracy", tf.keras.metrics.AUC()]) model.evaluate(test_ds)	model.evaluate(ds)
Save a model	model.save("project/model")	model.save("project/model")
Load a model	model = tf_keras.models.load_model("project/model")	model = ydf.load_model("project/model")
Export model as TF SavedModel	model.save("project/model")	model.to_tensorflow_saved_model("project/model", mode="tf")

Following is a 1:1 equivalent TF-DF and YDF code.

TF-DF

YDF

!pip install tensorflow tensorflow_decision_forests # Install TF-DF import tensorflow_decision_forests as tfdf import tensorflow as tf import pandas as pd # Load a dataset with Pandas train_df = pd.read_csv("train.csv") test_df = pd.read_csv("test.csv") # Convert the dataset to a TensorFlow Dataset. train_ds = tfdf.keras.pd_dataframe_to_tf_dataset(train_df, label="my_label") test_ds = tfdf.keras.pd_dataframe_to_tf_dataset(test_df, label="my_label") # Train a model model = tfdf.keras.RandomForestModel(num_trees=500) model.fit(train_ds) # Evaluate model. model.compile([tf.keras.metrics.SparseCategoricalAccuracy(),tf.keras.metrics.AUC()]) model.evaluate(test_ds) # Saved model model.save("/tmp/my_model")

pip install ydf # Install YDF import ydf import pandas as pd # Load a dataset with Pandas train_ds = pd.read_csv("train.csv") test_ds = pd.read_csv("test.csv") # Train a model model = ydf.RandomForestLearner(label="my_label", num_trees=500).train(train_ds) # Evaluate model. model.evaluate(test_ds) # Save the model model.save("/tmp/my_model")