How can Tensorflow be used to instantiate an estimator using Python?

TensorFlow Estimators provide a high-level API for building machine learning models. The DNNClassifier is a pre-made estimator that creates deep neural networks for classification tasks.

Read More: What is TensorFlow and how Keras work with TensorFlow to create Neural Networks?

An Estimator is TensorFlow's high-level representation of a complete model, designed for easy scaling and asynchronous training. We'll demonstrate using the classic Iris dataset for multi-class classification.

Setting Up Feature Columns

First, we need to define feature columns that describe how the model should use input data ?

import tensorflow as tf

# Define feature columns for Iris dataset
my_feature_columns = []
feature_names = ['sepal_length', 'sepal_width', 'petal_length', 'petal_width']

for key in feature_names:
    my_feature_columns.append(tf.feature_column.numeric_column(key=key))

print("Feature columns created:", len(my_feature_columns))

Feature columns created: 4

Creating the DNNClassifier

Now we can instantiate the DNNClassifier estimator with our feature columns ?

print("Build a DNN that has 2 hidden layers with 30 and 10 hidden nodes each")
classifier = tf.estimator.DNNClassifier(
    feature_columns=my_feature_columns,
    hidden_units=[30, 10],
    n_classes=3)

print("Classifier created successfully")

Build a DNN that has 2 hidden layers with 30 and 10 hidden nodes each
INFO:tensorflow:Using default config.
WARNING:tensorflow:Using temporary folder as model directory: /tmp/tmpdh8866zb
INFO:tensorflow:Using config: {'_model_dir': '/tmp/tmpdh8866zb', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_session_config': allow_soft_placement: true
graph_options {
  rewrite_options {
    meta_optimizer_iterations: ONE
  }
}
, '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_device_fn': None, '_protocol': None, '_eval_distribute': None, '_experimental_distribute': None, '_experimental_max_worker_delay_secs': None, '_session_creation_timeout_secs': 7200, '_checkpoint_save_graph_def': True, '_service': None, '_cluster_spec': ClusterSpec({}), '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1}
Classifier created successfully

Parameters

The DNNClassifier accepts several key parameters ?

Available Pre-made Estimators

TensorFlow provides several pre-made estimators for different use cases ?

• tf.estimator.DNNClassifier − Deep neural networks for multi-class classification
• tf.estimator.LinearClassifier − Linear models for classification
• tf.estimator.DNNLinearCombinedClassifier − Wide & deep models combining both approaches
• tf.estimator.DNNRegressor − Deep neural networks for regression tasks

Conclusion

TensorFlow's DNNClassifier provides an easy way to create deep neural network classifiers. Simply define your feature columns, specify the network architecture with hidden_units, and set the number of classes for your classification problem.