Adversarial Spectral Kernel Matching for Unsupervised Time Series Domain Adaptation

Installation

Datasets

Download data and convert to .tfrecord files for TensorFlow (./generate_tfrecords.sh)

Packages

We require the following packages. Adjust for your computer setup.

module load cuda/10.1.105 cudnn/7.6.4.38_cuda10.1 python3/3.7.4
pip install --user --upgrade pip
export PATH="$HOME/.local/bin:$PATH"
pip3 install --user --upgrade pip
pip3 install --user --upgrade numpy cython
pip3 install --user --upgrade tensorflow-gpu pillow lxml jupyter matplotlib pandas scikit-learn scipy tensorboard rarfile tqdm pyyaml grpcio absl-py

# If using --moving_average or F1 score metrics (typically tensorflow-addons, but that errors at the moment with TF 2.2)
pip3 install --user git+https://github.com/tensorflow/addons.git@r0.9

Usage

Example

Train our model on person 14 of the UCI HAR dataset and adapt to person 19.

python3 main.py \
    --logdir=example-logs --modeldir=example-models \
    --method=smd --dataset=ucihar --sources=14 \
    --target=19 --uid=0 --debugnum=0 --gpumem=0

Then evaluate that model on the holdout test data, outputting the results to a YAML file.

mkdir -p results
python3 main_eval.py \
    --logdir=example-logs --modeldir=example-models \
    --jobs=1 --gpus=1 --gpumem=0 \
    --match="ucihar-0-smd-[0-9]*" --selection="best_target" \
    --output_file=results/results_example_best_target-ucihar-0-daws.yaml

Note: there are a number of other models (e.g. --model=lstmfcn), methods (e.g. --method=dann), datasets (e.g. --dataset=wisdm_at), etc. implemented that you can experiment with beyond what was included in the paper.

Analysis

Then look at the resulting results/results_*.yaml files or analyze with analysis.py.

In addition, there are scripts for visualizing the datasets (datasets/view_datasets.py), viewing dataset statistics (dataset_statistics.py), and displaying or plotting the class balance of the data (class_balance.py, class_balance_plot.py).

Navigating the Code

Here is an outline of the key elements of the code.

Models

• --model=fcn base feature extractor in paper
• --model=lstmfcn fcn+lstm feature extractor

Methods

• dannMethods.py includes methods based on domain classifier (DANN, CoDATs)

• nannMethods.py includes methods based on non-learnable metrics (DDC, DAN, CORAL, etc.)

• nannMethods.py includes methods based on learnable metrics (our method)