forCar

time-series-forecasting & anomaly_detection

Installation

git clone https://github.com/KNUAI/forCar.git

cd forCar && pip install -r requirements.txt

If you want to forecaste, put your data in datasets/forecasting/
If you want to sample data for forecasting, you can download this ETDataset
ex)

If you want to anomaly detection, put your data in datasets/Anomaly_Detection/
If you want to sample data for anomaly detection, you can download this ST-AWFD
ex)
Notice anomaly detection dataset must have 'MaterialID', 'is_test', 'target' columns

Usage

If you want to train and test forecasting at once

python forecasting.py --model SCINet --data ETTh1.csv --cols HUFL HULL MUFL MULL LUFL LULL OT --seq_len 96 --pred_len 48

If you want to train forecasting

python forecasting_train.py --model SCINet --data ETTh1.csv --cols HUFL HULL MUFL MULL LUFL LULL OT --seq_len 96 --pred_len 48

If you want to test forecasting

python forecasting_test.py --model SCINet --data ETTh1.csv --cols HUFL HULL MUFL MULL LUFL LULL OT --seq_len 96 --pred_len 48

Write the model you want to use in model!
Write the file_name you want to use in data!
Write the variables you want to use in cols!
Write the length you want to input in seq_len!
Write the length you want to predict in pred_len!

If you want to train and test anomaly detection at once

python FD.py --data D1.csv --cols feature_1 feature_2 feature_3 feature_4 feature_5 feature_6 feature_7 feature_8 feature_9 feature_10 feature_11 feature_12 feature_13 feature_14 feature_15

If you want to train anomaly detection

python FD_train.py --data D1.csv --cols feature_1 feature_2 feature_3 feature_4 feature_5 feature_6 feature_7 feature_8 feature_9 feature_10 feature_11 feature_12 feature_13 feature_14 feature_15

If you want to test anomaly detection

python FD_test.py --data D1.csv --cols feature_1 feature_2 feature_3 feature_4 feature_5 feature_6 feature_7 feature_8 feature_9 feature_10 feature_11 feature_12 feature_13 feature_14 feature_15

Write the file_name you want to use in data!
Write the variables you want to use in cols!

The detailed descriptions for forecasting

Parameter name	Description of parameter
model	The model of experiment. This can be set to SCINet, Informer, LSTM
data	The data_file_name
inverse	Whether to inverse output data, using this argument means inversing output data (defaults to True)
gpu	The gpu no, used for training and inference (defaults to 0)
seq_len	Input sequence length of Informer encoder (defaults to 96)
pred_len	Prediction sequence length (defaults to 48)
cols	Certain cols from the data files as the input features
train_epochs	Train epochs (defaults to 100)
batch_size	The batch size of training input data (defaults to 32)
patience	Early stopping patience (defaults to 5)
learning_rate	Optimizer learning rate (defaults to 0.0001)
loss	Loss function: mse, mae (defaults to mae)
lradj	Ways to adjust the learning rate (defaults to type1)
evaluate	Evaluate the trained model
hidden_size	N_channel of module (defaults to 4)
kernel	Window_size: 3, 5, 7 (defaults to 5)
dropout	Dropout (defaults to 0.5)
num_decoder_layer	Evaluate the trained model (defaults to 1)

The detailed descriptions for anomaly detection

Parameter name	Description of parameter
data	The data_file_name
cols	Certain cols from the data files as the input features
fold	5-fold: 1, 2, 3, 4, 5 (defaults to 5)
latent_size	Dimension of latent vector (defaults to 128)
threshold_rate	Threshold_rate (defaults to 5)
n_layer	n_layers of rnn model (defaults to 1)
epoch	Train epochs (defaults to 200)
batch_size	The batch size of training input data (defaults to 32)
lr	Optimizer learning rate (defaults to 0.0001)
r_model	RNN model: LSTM, GRU (defaults to LSTM)
evaluate	Evaluate the trained model
patience	Early stopping patience (defaults to 3)
gpus	The gpu no, used for training and inference (defaults to 0)

Results

If you finish the experiment, you can see experiment results in picture directory