Code Examples¶
This section describes the code examples found in objax/examples.
Classification¶
Image¶
Example code available at examples/image_classification.
Logistic Regression¶
Train and evaluate a logistic regression model for binary classification on horses or humans dataset.
# Run command
python3 examples/image_classification/horses_or_humans_logistic.py
Code |
|
Data |
horses_or_humans from tensorflow_datasets |
Network |
Custom single layer |
Loss |
|
Optimizer |
|
Accuracy |
~77% |
Hardware |
CPU or GPU or TPU |
Digit Classification with Deep Neural Network (DNN)¶
Train and evaluate a DNNet model for multiclass classification on the MNIST dataset.
# Run command
python3 examples/image_classification/mnist_dnn.py
Code |
|
Data |
MNIST from tensorflow_datasets |
Network |
Deep Neural Net |
Loss |
|
Optimizer |
|
Accuracy |
~98% |
Hardware |
CPU or GPU or TPU |
Techniques |
Model weight averaging for improved accuracy using
|
Digit Classification with Convolutional Neural Network (CNN)¶
Train and evaluate a simple custom CNN model for multiclass classification on the MNIST dataset.
# Run command
python3 examples/image_classification/mnist_cnn.py
Code |
|
Data |
MNIST from tensorflow_datasets |
Network |
Custom Convolution Neural Net using |
Loss |
|
Optimizer |
|
Accuracy |
~99.5% |
Hardware |
CPU or GPU or TPU |
Techniques |
|
Digit Classification using Differential Privacy¶
Train and evaluate a convNet model for MNIST dataset with differential privacy.
# Run command
python3 examples/image_classification/mnist_dp.py
# See available options with
python3 examples/image_classification/mnist_dp.py --help
Code |
|
Data |
MNIST from tensorflow_datasets |
Network |
Custom Convolution Neural Net using |
Loss |
|
Optimizer |
|
Accuracy |
|
Hardware |
GPU |
Techniques |
|
Image Classification on CIFAR-10 (Simple)¶
Train and evaluate a wide resnet model for multiclass classification on the CIFAR10 dataset.
# Run command
python3 examples/image_classification/cifar10_simple.py
Code |
|
Data |
CIFAR10 from tf.keras.datasets |
Network |
Wide ResNet using |
Loss |
|
Optimizer |
|
Accuracy |
~91% |
Hardware |
GPU or TPU |
Techniques |
|
Image Classification on CIFAR-10 (Advanced)¶
Train and evaluate convNet models for multiclass classification on the CIFAR10 dataset.
# Run command
python3 examples/image_classification/cifar10_advanced.py
# Run with custom settings
python3 examples/image_classification/cifar10_advanced.py --weight_decay=0.0001 --batch=64 --lr=0.03 --epochs=256
# See available options with
python3 examples/image_classification/cifar10_advanced.py --help
Code |
|
Data |
|
Network |
Configurable with |
Loss |
|
Optimizer |
|
Accuracy |
~94% |
Hardware |
GPU, Multi-GPU or TPU |
Techniques |
|
Image Classification on ImageNet¶
Train and evaluate a ResNet50 model on the ImageNet dataset. See README for additional information.
Code |
|
Data |
|
Network |
|
Loss |
|
Optimizer |
|
Accuracy |
|
Hardware |
GPU, Multi-GPU or TPU |
Techniques |
|
Semi-Supervised Learning¶
Example code available at examples/fixmatch.
Semi-Supervised Learning with FixMatch¶
Semi-supervised learning of image classification models with FixMatch.
# Run command
python3 examples/fixmatch/fixmatch.py
# Run with custom settings
python3 examples/fixmatch/fixmatch.py --dataset=cifar10.3@1000-0
# See available options with
python3 examples/fixmatch/fixmatch.py --help
Code |
|
Data |
|
Network |
Custom implementation of Wide ResNet. |
Loss |
|
Optimizer |
|
Accuracy |
See paper |
Hardware |
GPU, Multi-GPU, TPU |
Techniques |
|
GPT-2¶
Example code is available at examples/gpt-2.
Generating a Text Sequence using GPT-2¶
Load pretrained GPT-2 model (124M parameter) and demonstrate how to use the model to generate a text sequence. See README for additional information.
Code |
|
Hardware |
GPU or TPU |
Techniques |
|
RNN¶
Example code is available at examples/text_generation.
Train a Vanilla RNN to Predict Characters¶
Train and evaluate a vanilla RNN model on the Shakespeare corpus dataset. See README for additional information.
# Run command
python3 examples/text_generation/shakespeare_rnn.py
Code |
|
Data |
|
Network |
Custom implementation of vanilla RNN. |
Loss |
|
Optimizer |
|
Hardware |
GPU or TPU |
Techniques |
|
Optimization¶
Example codes available at examples/maml.
Jaxboard¶
Example code available at examples/jaxboard.