Skip to content

TensorFlow

Google's end-to-end machine learning platform, strongest at production deployment.

AI & Machine LearningML/AIPython
Official docs ↗GitHub ↗

What it is

TensorFlow is an end-to-end open-source platform for machine learning. It allows you to build and deploy machine learning models easily, from training to inference, across multiple platforms and devices.

TensorFlow provides APIs for building neural networks, performing automatic differentiation, training models, and serving models for production. It supports both eager execution (imperative programming) and graph execution (declarative programming).

Best known for
Deployment reach — server, mobile, browser and TPU
Licence
Apache 2.0

When to use it

The question documentation cannot answer for you — because it cannot recommend something else.

Reach for it when

  • Deploying models to mobile, browsers or embedded devices via TFLite and TensorFlow.js
  • You need TPU support or the surrounding production tooling — TFX, Serving, TensorBoard
  • Maintaining an existing TensorFlow codebase

Look elsewhere when

  • Starting fresh research — PyTorch has won most of that ground and the ecosystem follows it

Installation

pip install tensorflow

Getting started

The smallest useful thing you can do with it, and what each part means.

Simple Tensor operations
import tensorflow as tf
x = tf.constant([[1., 2.],[3., 4.]])
y = tf.constant([[5., 6.],[7., 8.]])
print(tf.matmul(x, y))
Defines two constant tensors and performs matrix multiplication.
Creating a simple neural network
from tensorflow import keras
model = keras.Sequential([
    keras.layers.Dense(10, activation='relu', input_shape=(5,)),
    keras.layers.Dense(1, activation='sigmoid')
])
model.compile(optimizer='adam', loss='binary_crossentropy')
Creates a simple feedforward neural network with one hidden layer using the Keras API integrated in TensorFlow.

Advanced usage

Where the library earns its place over a simpler alternative.

Training a model
import numpy as np
X_train = np.random.rand(100,5)
y_train = np.random.randint(0,2, size=(100,1))
model.fit(X_train, y_train, epochs=10, batch_size=8)
Trains the previously defined model on synthetic data for 10 epochs with a batch size of 8.
Saving and loading a model
model.save('my_model')
new_model = keras.models.load_model('my_model')
Demonstrates saving a trained model and loading it later for inference or further training.
Using TensorFlow Datasets
import tensorflow_datasets as tfds
dataset = tfds.load('mnist', split='train')
for example in dataset.take(1):
    image, label = example['image'], example['label']
    print(image.shape, label)
Shows how to load standard datasets from TensorFlow Datasets for training or testing.
Custom training loop with GradientTape
x = tf.random.normal((10,3))
y_true = tf.random.normal((10,1))
weights = tf.Variable(tf.random.normal((3,1)))
bias = tf.Variable(tf.zeros(1))
optimizer = tf.optimizers.SGD(0.01)
for i in range(100):
    with tf.GradientTape() as tape:
        y_pred = tf.matmul(x, weights) + bias
        loss = tf.reduce_mean(tf.square(y_true - y_pred))
    grads = tape.gradient(loss, [weights, bias])
    optimizer.apply_gradients(zip(grads, [weights, bias]))
Implements a custom training loop to optimize weights using TensorFlow’s GradientTape and optimizer API.

Errors and fixes

The failures you are most likely to hit, and what actually resolves them.

InvalidArgumentError
Check input shapes and dtypes to ensure compatibility with model layers.
ResourceExhaustedError
Reduce batch size or use GPU/TPU memory more efficiently.
ModuleNotFoundError: No module named 'tensorflow'
Ensure TensorFlow is installed in the current Python environment using pip or conda.

Best practices

  • Use tf.data API for efficient data loading and preprocessing.
  • Use eager execution for debugging and graph execution for production performance.
  • Leverage TensorBoard for visualizing training metrics.
  • Use mixed precision and hardware acceleration (GPU/TPU) for faster training.
  • Organize models and code using the Keras API for simplicity and modularity.

Alternatives

Comparable options, and the reason you would pick one over the other.

Background

Why it exists, and what it was reacting to.

TensorFlow was developed by the Google Brain team and released in 2015. It was designed to provide a comprehensive ecosystem for developing, training, and deploying machine learning models. TensorFlow supports deep learning, neural networks, and large-scale numerical computations and has become one of the most widely used ML frameworks in both industry and academia.

Related libraries

Others in AI & Machine Learning for Python.

← All Python librariesPython guide