A tensorflow implementation of Meta-Learning with Temporal Convolutions
Embedding for Omniglot dataset is only available now.
- Python 3.6+
- Tensorflow 1.3
- Image dataset for training/validation (Python dictionary-like object)
python train.py --dataset omniglot --n 5 --k 1 --dilation 1 2 1 2 4 8 16 --lr 5e-4 --batch_size 64
It will print valiation loss and accuracy.
Checkpoints and summaries for other metrics are saved in ./runs/tcml_{input_dim}_{num_dense_filter}_{attention_value_dim}_{lr}
| Omniglot | 5-way, 1-shot | 5-way, 5-shot |
|---|---|---|
| Accuracy | 95.12% | 95.01% |
Since the paper didn't share exact hyperparameters, it is hard to make same results with the paper. This code performs ~95% accuracy for 5-way environment with way smaller numbers of dilation stacks.
When I tried to use same dilation stack with paper(1 2 1 2 1 2 1 2 1 2 4 8 16), it is too deep, thus the loss converges at 1.60 with poor accuracy.
Still trying to find better hyperparameters to get higher performance.
MIT
Donghwa Kim (@storykim)