Composing Decision Forest and Neural Network models

Introduction

Welcome to the model composition tutorial for TensorFlow Decision Forests (TF-DF). This notebook shows you how to compose multiple decision forest and neural network models together using a common preprocessing layer and the Keras functional API.

You might want to compose models together to improve predictive performance (ensembling), to get the best of different modeling technologies (heterogeneous model ensembling), to train different part of the model on different datasets (e.g. pre-training), or to create a stacked model (e.g. a model operates on the predictions of another model).

This tutorial covers an advanced use case of model composition using the Functional API. You can find examples for simpler scenarios of model composition in the "feature preprocessing" section of this tutorial and in the "using a pretrained text embedding" section of this tutorial.

Here is the structure of the model you'll build:

Toggle code

!pip install graphviz -U --quiet

from graphviz import Source

Source("""
digraph G {
  raw_data [label="Input features"];
  preprocess_data [label="Learnable NN pre-processing", shape=rect];

  raw_data -> preprocess_data

  subgraph cluster_0 {
    color=grey;
    a1[label="NN layer", shape=rect];
    b1[label="NN layer", shape=rect];
    a1 -> b1;
    label = "Model #1";
  }

   subgraph cluster_1 {
    color=grey;
    a2[label="NN layer", shape=rect];
    b2[label="NN layer", shape=rect];
    a2 -> b2;
    label = "Model #2";
  }

  subgraph cluster_2 {
    color=grey;
    a3[label="Decision Forest", shape=rect];
    label = "Model #3";
  }

  subgraph cluster_3 {
    color=grey;
    a4[label="Decision Forest", shape=rect];
    label = "Model #4";
  }

  preprocess_data -> a1;
  preprocess_data -> a2;
  preprocess_data -> a3;
  preprocess_data -> a4;

  b1  -> aggr;
  b2  -> aggr;
  a3 -> aggr;
  a4 -> aggr;

  aggr [label="Aggregation (mean)", shape=rect]
  aggr -> predictions
}
""")

Your composed model has three stages:

1. The first stage is a preprocessing layer composed of a neural network and common to all the models in the next stage. In practice, such a preprocessing layer could either be a pre-trained embedding to fine-tune, or a randomly initialized neural network.
2. The second stage is an ensemble of two decision forest and two neural network models.
3. The last stage averages the predictions of the models in the second stage. It does not contain any learnable weights.

The neural networks are trained using the backpropagation algorithm and gradient descent. This algorithm has two important properties: (1) The layer of neural network can be trained if its receives a loss gradient (more precicely, the gradient of the loss according to the layer's output), and (2) the algorithm "transmits" the loss gradient from the layer's output to the layer's input (this is the "chain rule"). For these two reasons, Backpropagation can train together multiple layers of neural networks stacked on top of each other.

In this example, the decision forests are trained with the Random Forest (RF) algorithm. Unlike Backpropagation, the training of RF does not "transmit" the loss gradient to from its output to its input. For this reasons, the classical RF algorithm cannot be used to train or fine-tune a neural network underneath. In other words, the "decision forest" stages cannot be used to train the "Learnable NN pre-processing block".

1. Train the preprocessing and neural networks stage.
2. Train the decision forest stages.

Install TensorFlow Decision Forests

Install TF-DF by running the following cell.

pip install tensorflow_decision_forests -U --quiet

Wurlitzer is needed to display the detailed training logs in Colabs (when using verbose=2 in the model constructor).

pip install wurlitzer -U --quiet

Import libraries

import tensorflow_decision_forests as tfdf

import os
import numpy as np
import pandas as pd
import tensorflow as tf
import math
import matplotlib.pyplot as plt

WARNING:root:TF Parameter Server distributed training not available (this is expected for the pre-build release).

Dataset

You will use a simple synthetic dataset in this tutorial to make it easier to interpret the final model.

def make_dataset(num_examples, num_features, seed=1234):
  np.random.seed(seed)
  features = np.random.uniform(-1, 1, size=(num_examples, num_features))
  noise = np.random.uniform(size=(num_examples))

  left_side = np.sqrt(
      np.sum(np.multiply(np.square(features[:, 0:2]), [1, 2]), axis=1))
  right_side = features[:, 2] * 0.7 + np.sin(
      features[:, 3] * 10) * 0.5 + noise * 0.0 + 0.5

  labels = left_side <= right_side
  return features, labels.astype(int)

Generate some examples:

make_dataset(num_examples=5, num_features=4)

(array([[-0.6169611 ,  0.24421754, -0.12454452,  0.57071717],
        [ 0.55995162, -0.45481479, -0.44707149,  0.60374436],
        [ 0.91627871,  0.75186527, -0.28436546,  0.00199025],
        [ 0.36692587,  0.42540405, -0.25949849,  0.12239237],
        [ 0.00616633, -0.9724631 ,  0.54565324,  0.76528238]]),
 array([0, 0, 0, 1, 0]))

You can also plot them to get an idea of the synthetic pattern:

plot_features, plot_label = make_dataset(num_examples=50000, num_features=4)

plt.rcParams["figure.figsize"] = [8, 8]
common_args = dict(c=plot_label, s=1.0, alpha=0.5)

plt.subplot(2, 2, 1)
plt.scatter(plot_features[:, 0], plot_features[:, 1], **common_args)

plt.subplot(2, 2, 2)
plt.scatter(plot_features[:, 1], plot_features[:, 2], **common_args)

plt.subplot(2, 2, 3)
plt.scatter(plot_features[:, 0], plot_features[:, 2], **common_args)

plt.subplot(2, 2, 4)
plt.scatter(plot_features[:, 0], plot_features[:, 3], **common_args)

<matplotlib.collections.PathCollection at 0x7f675755add0>

Note that this pattern is smooth and not axis aligned. This will advantage the neural network models. This is because it is easier for a neural network than for a decision tree to have round and non aligned decision boundaries.

On the other hand, we will train the model on a small datasets with 2500 examples. This will advantage the decision forest models. This is because decision forests are much more efficient, using all the available information from the examples (decision forests are "sample efficient").

Our ensemble of neural networks and decision forests will use the best of both worlds.

Let's create a train and test tf.data.Dataset:

def make_tf_dataset(batch_size=64, **args):
  features, labels = make_dataset(**args)
  return tf.data.Dataset.from_tensor_slices(
      (features, labels)).batch(batch_size)


num_features = 10

train_dataset = make_tf_dataset(
    num_examples=2500, num_features=num_features, batch_size=100, seed=1234)
test_dataset = make_tf_dataset(
    num_examples=10000, num_features=num_features, batch_size=100, seed=5678)

Model structure

Define the model structure as follows:

# Input features.
raw_features = tf.keras.layers.Input(shape=(num_features,))

# Stage 1
# =======

# Common learnable pre-processing
preprocessor = tf.keras.layers.Dense(10, activation=tf.nn.relu6)
preprocess_features = preprocessor(raw_features)

# Stage 2
# =======

# Model #1: NN
m1_z1 = tf.keras.layers.Dense(5, activation=tf.nn.relu6)(preprocess_features)
m1_pred = tf.keras.layers.Dense(1, activation=tf.nn.sigmoid)(m1_z1)

# Model #2: NN
m2_z1 = tf.keras.layers.Dense(5, activation=tf.nn.relu6)(preprocess_features)
m2_pred = tf.keras.layers.Dense(1, activation=tf.nn.sigmoid)(m2_z1)


# Model #3: DF
model_3 = tfdf.keras.RandomForestModel(num_trees=1000, random_seed=1234)
m3_pred = model_3(preprocess_features)

# Model #4: DF
model_4 = tfdf.keras.RandomForestModel(
    num_trees=1000,
    #split_axis="SPARSE_OBLIQUE", # Uncomment this line to increase the quality of this model
    random_seed=4567)
m4_pred = model_4(preprocess_features)

# Since TF-DF uses deterministic learning algorithms, you should set the model's
# training seed to different values otherwise both
# `tfdf.keras.RandomForestModel` will be exactly the same.

# Stage 3
# =======

mean_nn_only = tf.reduce_mean(tf.stack([m1_pred, m2_pred], axis=0), axis=0)
mean_nn_and_df = tf.reduce_mean(
    tf.stack([m1_pred, m2_pred, m3_pred, m4_pred], axis=0), axis=0)

# Keras Models
# ============

ensemble_nn_only = tf.keras.models.Model(raw_features, mean_nn_only)
ensemble_nn_and_df = tf.keras.models.Model(raw_features, mean_nn_and_df)

Use /tmp/tmpe2xrrxob as temporary training directory
Warning:  The model was called directly (i.e. using `model(data)` instead of using `model.predict(data)`) before being trained. The model will only return zeros until trained. The output shape might change after training Tensor("inputs:0", shape=(None, 10), dtype=float32)
Use /tmp/tmpgy1mhjkn as temporary training directory
Warning:  The model was called directly (i.e. using `model(data)` instead of using `model.predict(data)`) before being trained. The model will only return zeros until trained. The output shape might change after training Tensor("inputs:0", shape=(None, 10), dtype=float32)

Before you train the model, you can plot it to check if it is similar to the initial diagram.

from keras.utils.vis_utils import plot_model

plot_model(ensemble_nn_and_df, to_file="/tmp/model.png", show_shapes=True)

Model training

First train the preprocessing and two neural network layers using the backpropagation algorithm.

%%time
ensemble_nn_only.compile(
        optimizer=tf.keras.optimizers.Adam(),
        loss=tf.keras.losses.BinaryCrossentropy(),
        metrics=["accuracy"])

ensemble_nn_only.fit(train_dataset, epochs=20, validation_data=test_dataset)

Epoch 1/20
25/25 [==============================] - 1s 17ms/step - loss: 0.6534 - accuracy: 0.6620 - val_loss: 0.6348 - val_accuracy: 0.6962
Epoch 2/20
25/25 [==============================] - 0s 11ms/step - loss: 0.6118 - accuracy: 0.7216 - val_loss: 0.5994 - val_accuracy: 0.7278
Epoch 3/20
25/25 [==============================] - 0s 11ms/step - loss: 0.5790 - accuracy: 0.7420 - val_loss: 0.5724 - val_accuracy: 0.7372
Epoch 4/20
25/25 [==============================] - 0s 11ms/step - loss: 0.5542 - accuracy: 0.7496 - val_loss: 0.5525 - val_accuracy: 0.7390
Epoch 5/20
25/25 [==============================] - 0s 11ms/step - loss: 0.5358 - accuracy: 0.7500 - val_loss: 0.5376 - val_accuracy: 0.7392
Epoch 6/20
25/25 [==============================] - 0s 11ms/step - loss: 0.5211 - accuracy: 0.7500 - val_loss: 0.5253 - val_accuracy: 0.7392
Epoch 7/20
25/25 [==============================] - 0s 11ms/step - loss: 0.5086 - accuracy: 0.7500 - val_loss: 0.5143 - val_accuracy: 0.7392
Epoch 8/20
25/25 [==============================] - 0s 11ms/step - loss: 0.4973 - accuracy: 0.7500 - val_loss: 0.5041 - val_accuracy: 0.7392
Epoch 9/20
25/25 [==============================] - 0s 12ms/step - loss: 0.4867 - accuracy: 0.7500 - val_loss: 0.4944 - val_accuracy: 0.7392
Epoch 10/20
25/25 [==============================] - 0s 11ms/step - loss: 0.4765 - accuracy: 0.7496 - val_loss: 0.4850 - val_accuracy: 0.7392
Epoch 11/20
25/25 [==============================] - 0s 12ms/step - loss: 0.4670 - accuracy: 0.7504 - val_loss: 0.4761 - val_accuracy: 0.7390
Epoch 12/20
25/25 [==============================] - 0s 11ms/step - loss: 0.4580 - accuracy: 0.7500 - val_loss: 0.4678 - val_accuracy: 0.7408
Epoch 13/20
25/25 [==============================] - 0s 11ms/step - loss: 0.4500 - accuracy: 0.7536 - val_loss: 0.4603 - val_accuracy: 0.7448
Epoch 14/20
25/25 [==============================] - 0s 12ms/step - loss: 0.4427 - accuracy: 0.7592 - val_loss: 0.4538 - val_accuracy: 0.7528
Epoch 15/20
25/25 [==============================] - 0s 12ms/step - loss: 0.4363 - accuracy: 0.7660 - val_loss: 0.4481 - val_accuracy: 0.7607
Epoch 16/20
25/25 [==============================] - 0s 11ms/step - loss: 0.4308 - accuracy: 0.7736 - val_loss: 0.4433 - val_accuracy: 0.7688
Epoch 17/20
25/25 [==============================] - 0s 11ms/step - loss: 0.4260 - accuracy: 0.7820 - val_loss: 0.4391 - val_accuracy: 0.7765
Epoch 18/20
25/25 [==============================] - 0s 12ms/step - loss: 0.4216 - accuracy: 0.7880 - val_loss: 0.4353 - val_accuracy: 0.7812
Epoch 19/20
25/25 [==============================] - 0s 11ms/step - loss: 0.4177 - accuracy: 0.7888 - val_loss: 0.4320 - val_accuracy: 0.7837
Epoch 20/20
25/25 [==============================] - 0s 11ms/step - loss: 0.4142 - accuracy: 0.7948 - val_loss: 0.4288 - val_accuracy: 0.7864
CPU times: user 8.85 s, sys: 2.03 s, total: 10.9 s
Wall time: 6.83 s
<keras.callbacks.History at 0x7f6759fc25d0>

Let's evaluate the preprocessing and the part with the two neural networks only:

evaluation_nn_only = ensemble_nn_only.evaluate(test_dataset, return_dict=True)
print("Accuracy (NN #1 and #2 only): ", evaluation_nn_only["accuracy"])
print("Loss (NN #1 and #2 only): ", evaluation_nn_only["loss"])

100/100 [==============================] - 0s 2ms/step - loss: 0.4288 - accuracy: 0.7864
Accuracy (NN #1 and #2 only):  0.7864000201225281
Loss (NN #1 and #2 only):  0.4288266599178314

Let's train the two Decision Forest components (one after another).

%%time
train_dataset_with_preprocessing = train_dataset.map(lambda x,y: (preprocessor(x), y))
test_dataset_with_preprocessing = test_dataset.map(lambda x,y: (preprocessor(x), y))

model_3.fit(train_dataset_with_preprocessing)
model_4.fit(train_dataset_with_preprocessing)

WARNING:tensorflow:AutoGraph could not transform <function <lambda> at 0x7f674878eef0> and will run it as-is.
Cause: could not parse the source code of <function <lambda> at 0x7f674878eef0>: no matching AST found among candidates:

To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
WARNING:tensorflow:AutoGraph could not transform <function <lambda> at 0x7f674878eef0> and will run it as-is.
Cause: could not parse the source code of <function <lambda> at 0x7f674878eef0>: no matching AST found among candidates:

To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
WARNING: AutoGraph could not transform <function <lambda> at 0x7f674878eef0> and will run it as-is.
Cause: could not parse the source code of <function <lambda> at 0x7f674878eef0>: no matching AST found among candidates:

To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
WARNING:tensorflow:AutoGraph could not transform <function <lambda> at 0x7f674878ef80> and will run it as-is.
Cause: could not parse the source code of <function <lambda> at 0x7f674878ef80>: no matching AST found among candidates:

To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
WARNING:tensorflow:AutoGraph could not transform <function <lambda> at 0x7f674878ef80> and will run it as-is.
Cause: could not parse the source code of <function <lambda> at 0x7f674878ef80>: no matching AST found among candidates:

To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
WARNING: AutoGraph could not transform <function <lambda> at 0x7f674878ef80> and will run it as-is.
Cause: could not parse the source code of <function <lambda> at 0x7f674878ef80>: no matching AST found among candidates:

To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
Starting reading the dataset
21/25 [========================>.....] - ETA: 0s  
Dataset read in 0:00:03.646528
Training model
Model trained in 0:00:01.597792
Compiling model
[INFO kernel.cc:1153] Loading model from path
25/25 [==============================] - 6s 111ms/step
[INFO abstract_model.cc:1063] Engine "RandomForestOptPred" built
[INFO kernel.cc:1001] Use fast generic engine
WARNING:tensorflow:AutoGraph could not transform <function simple_ml_inference_op_with_handle at 0x7f675b2c9680> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, `export AUTOGRAPH_VERBOSITY=10`) and attach the full output.
Cause: could not get source code
To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
WARNING:tensorflow:AutoGraph could not transform <function simple_ml_inference_op_with_handle at 0x7f675b2c9680> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, `export AUTOGRAPH_VERBOSITY=10`) and attach the full output.
Cause: could not get source code
To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
WARNING: AutoGraph could not transform <function simple_ml_inference_op_with_handle at 0x7f675b2c9680> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, `export AUTOGRAPH_VERBOSITY=10`) and attach the full output.
Cause: could not get source code
To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
Starting reading the dataset
21/25 [========================>.....] - ETA: 0s
Dataset read in 0:00:00.187466
Training model
Model trained in 0:00:01.585278
Compiling model
[INFO kernel.cc:1153] Loading model from path
25/25 [==============================] - 3s 107ms/step
[INFO kernel.cc:1001] Use fast generic engine
CPU times: user 25.1 s, sys: 742 ms, total: 25.9 s
Wall time: 10.7 s
<keras.callbacks.History at 0x7f6787502550>

And let's evaluate the Decision Forests individually.

model_3.compile(["accuracy"])
model_4.compile(["accuracy"])

evaluation_df3_only = model_3.evaluate(
    test_dataset_with_preprocessing, return_dict=True)
evaluation_df4_only = model_4.evaluate(
    test_dataset_with_preprocessing, return_dict=True)

print("Accuracy (DF #3 only): ", evaluation_df3_only["accuracy"])
print("Accuracy (DF #4 only): ", evaluation_df4_only["accuracy"])

100/100 [==============================] - 1s 11ms/step - loss: 0.0000e+00 - accuracy: 0.8127
100/100 [==============================] - 1s 11ms/step - loss: 0.0000e+00 - accuracy: 0.8123
Accuracy (DF #3 only):  0.8126999735832214
Accuracy (DF #4 only):  0.8123000264167786

Let's evaluate the entire model composition:

ensemble_nn_and_df.compile(
    loss=tf.keras.losses.BinaryCrossentropy(), metrics=["accuracy"])

evaluation_nn_and_df = ensemble_nn_and_df.evaluate(
    test_dataset, return_dict=True)

print("Accuracy (2xNN and 2xDF): ", evaluation_nn_and_df["accuracy"])
print("Loss (2xNN and 2xDF): ", evaluation_nn_and_df["loss"])

100/100 [==============================] - 1s 12ms/step - loss: 0.3953 - accuracy: 0.8100
Accuracy (2xNN and 2xDF):  0.8100000023841858
Loss (2xNN and 2xDF):  0.3953089416027069

To finish, let's finetune the neural network layer a bit more. Note that we do not finetune the pre-trained embedding as the DF models depends on it (unless we would also retrain them after).

In summary, you have:

Toggle code

print(f"Accuracy (NN #1 and #2 only):\t{evaluation_nn_only['accuracy']:.6f}")
print(f"Accuracy (DF #3 only):\t\t{evaluation_df3_only['accuracy']:.6f}")
print(f"Accuracy (DF #4 only):\t\t{evaluation_df4_only['accuracy']:.6f}")
print("----------------------------------------")
print(f"Accuracy (2xNN and 2xDF):\t{evaluation_nn_and_df['accuracy']:.6f}")


def delta_percent(src_eval, key):
  src_acc = src_eval["accuracy"]
  final_acc = evaluation_nn_and_df["accuracy"]
  increase = final_acc - src_acc
  print(f"\t\t\t\t  {increase:+.6f} over {key}")


delta_percent(evaluation_nn_only, "NN #1 and #2 only")
delta_percent(evaluation_df3_only, "DF #3 only")
delta_percent(evaluation_df4_only, "DF #4 only")

Accuracy (NN #1 and #2 only): 0.786400
Accuracy (DF #3 only):        0.812700
Accuracy (DF #4 only):        0.812300
----------------------------------------
Accuracy (2xNN and 2xDF): 0.810000
                  +0.023600 over NN #1 and #2 only
                  -0.002700 over DF #3 only
                  -0.002300 over DF #4 only

Here, you can see that the composed model performs better than its individual parts. This is why ensembles work so well.

What's next?

In this example, you saw how to combine decision forests with neural networks. An extra step would be to further train the neural network and the decision forests together.

In addition, for the sake of clarrity, the decision forests received only the preprocessed input. However, decision forests are generally great are consuming raw data. The model would be improved by also feeding the raw features to the decision forest models.

In this example, the final model is the average of the predictions of the individual models. This solution works well if all of the model perform more of less with the same. However, if one of the sub-models is very good, aggregating it with other models might actually be detrimental (or vice-versa; for example try to reduce the number of examples from 1k and see how it hurts the neural networks a lot; or enable the SPARSE_OBLIQUE split in the second Random Forest model).