CIFAR-10 CNN Notebook
Budget: $30 – $250 USD
I need a single, self-contained Jupyter Notebook (Python 3) that walks from data import to final evaluation for the CIFAR-10 image-classification task. The scope is deliberately focused on a Convolutional Neural Network so that the notebook can serve both as a clear reference implementation and a teaching aid.
Within one runnable file, please:
• Load CIFAR-10 directly from a standard library (torchvision or Keras) and perform the usual train/validation/test split, including basic normalization and a couple of lightweight augmentations.
• Build and document a CNN architecture of sensible depth for this dataset, implemented in either PyTorch or TensorFlow/Keras. Hyper-parameters (learning rate, batch size, epoch count) should be easy to tweak in a single cell.
• Train the network with real-time progress reporting and save the best weights.
• Evaluate the final model on the held-out test set, chart the training curves, and briefly interpret the results in markdown cells.
• Add an “inference” section that lets a user drop in a single image and see the predicted label with class probabilities.
Keep external dependencies to the usual scientific-Python stack so the notebook runs straight away on Google Colab or any GPU-enabled environment. Make sure every cell runs top-to-bottom without manual edits, and annotate the key steps so someone new to deep learning can follow the flow.
Deliverable: the .ipynb file, ready to execute, with clear cell structure and clean comments throughout.
Within one runnable file, please:
• Load CIFAR-10 directly from a standard library (torchvision or Keras) and perform the usual train/validation/test split, including basic normalization and a couple of lightweight augmentations.
• Build and document a CNN architecture of sensible depth for this dataset, implemented in either PyTorch or TensorFlow/Keras. Hyper-parameters (learning rate, batch size, epoch count) should be easy to tweak in a single cell.
• Train the network with real-time progress reporting and save the best weights.
• Evaluate the final model on the held-out test set, chart the training curves, and briefly interpret the results in markdown cells.
• Add an “inference” section that lets a user drop in a single image and see the predicted label with class probabilities.
Keep external dependencies to the usual scientific-Python stack so the notebook runs straight away on Google Colab or any GPU-enabled environment. Make sure every cell runs top-to-bottom without manual edits, and annotate the key steps so someone new to deep learning can follow the flow.
Deliverable: the .ipynb file, ready to execute, with clear cell structure and clean comments throughout.