Merge pull request #25 from JulesBelveze/feat/hydra-parser

feat: use hydra config parser

Author: JulesBelveze

README.md

@@ -6,35 +6,40 @@
 </p>
 
 This repository contains an autoencoder for multivariate time series forecasting.
-It features two attention mechanisms described in *[A Dual-Stage Attention-Based Recurrent Neural Network for Time Series Prediction](https://arxiv.org/abs/1704.02971)* and was inspired by [Seanny123's repository](https://github.com/Seanny123/da-rnn).
+It features two attention mechanisms described
+in *[A Dual-Stage Attention-Based Recurrent Neural Network for Time Series Prediction](https://arxiv.org/abs/1704.02971)*
+and was inspired by [Seanny123's repository](https://github.com/Seanny123/da-rnn).
 
 ![Autoencoder architecture](autoenc_architecture.png)
+
 ## Download and dependencies
+
 To clone the repository please run:
+
 ```
 git clone https://github.com/JulesBelveze/time-series-autoencoder.git
 ```
 
 To install all the required dependencies please run:
+
 ```
 python3 -m venv .venv/tsa
 source .venv/tsa/bin/active
 poetry install
 ```
 
 ## Usage
+
+The project uses [Hydra](https://hydra.cc/docs/intro/) as a configuration parser. You can simply change the parameters
+directly within your `.yaml` file or you can override/set parameter using flags (for a complete guide please refer to
+the docs).
+
 ```
-python main.py [-h] [--batch-size BATCH_SIZE] [--output-size OUTPUT_SIZE]
-               [--label-col LABEL_COL] [--input-att INPUT_ATT]
-               [--temporal-att TEMPORAL_ATT] [--seq-len SEQ_LEN]
-               [--hidden-size-encoder HIDDEN_SIZE_ENCODER]
-               [--hidden-size-decoder HIDDEN_SIZE_DECODER]
-               [--reg-factor1 REG_FACTOR1] [--reg-factor2 REG_FACTOR2]
-               [--reg1 REG1] [--reg2 REG2] [--denoising DENOISING]
-               [--do-train DO_TRAIN] [--do-eval DO_EVAL]
-               [--data-path DATA_PATH] [--output-dir OUTPUT_DIR] [--ckpt CKPT]
+python3 main.py -cn=[PATH_TO_FOLDER_CONFIG] -cp=[CONFIG_NAME]
 ```
+
 Optional arguments:
+
 ```  
   -h, --help            show this help message and exit
   --batch-size BATCH_SIZE
@@ -71,16 +76,19 @@ Optional arguments:
                         name of folder to output files
   --ckpt CKPT           checkpoint path for evaluation 
   ```
-  
-  ## Features
-  * handles multivariate time series 
-  * attention mechanisms
-  * denoising autoencoder
-  * sparse autoencoder
-  
-  ## Examples
-  You can find under the `examples` scripts to train the model in both cases:
-  * reconstruction: the dataset can be found [here](https://gist.github.com/JulesBelveze/99ecdbea62f81ce647b131e7badbb24a)
-  * forecasting: the dataset can be found [here](https://gist.github.com/JulesBelveze/e9997b9b0b68101029b461baf698bd72)
+
+## Features
+
+* handles multivariate time series
+* attention mechanisms
+* denoising autoencoder
+* sparse autoencoder
+
+## Examples
+
+You can find under the `examples` scripts to train the model in both cases:
+
+* reconstruction: the dataset can be found [here](https://gist.github.com/JulesBelveze/99ecdbea62f81ce647b131e7badbb24a)
+* forecasting: the dataset can be found [here](https://gist.github.com/JulesBelveze/e9997b9b0b68101029b461baf698bd72)
 
 

examples/forecasting/config.yaml

@@ -0,0 +1,41 @@
+data:
+  _target_: tsa.dataset.TimeSeriesDataset
+  batch_size: 16
+  categorical_cols: [ ]
+  index_col: "Date_Time"
+  target_col: [ "AH" ]
+  data_path: "../data/AirQualityUCI.csv"
+  prediction_window: 1
+  seq_length: 3
+  task:
+    _target_: tsa.dataset.Tasks
+    value: prediction
+
+training:
+  batch_size: ${data.batch_size}
+  denoising: False
+  directions: 1
+  gradient_accumulation_steps: 1
+  hidden_size_encoder: 64
+  hidden_size_decoder: 64
+  input_att: True
+  lr: 1e-5
+  lrs_step_size: 5000
+  max_grad_norm: 0.1
+  num_epochs: 100
+  output_size: 1
+  reg1: True
+  reg2: False
+  reg_factor1: 1e-4
+  reg_factor2: 1e-4
+  seq_len: ${data.seq_length}
+  temporal_att: True
+
+general:
+  do_eval: True
+  do_train: True
+  logging_steps: 100
+
+  output_dir: "output"
+  save_steps: 5000
+  eval_during_training: True

examples/forecasting/config_forecasting.py

@@ -1,72 +1,29 @@
+import hydra
 import torch
-import argparse
-import pandas as pd
 import torch.nn as nn
-from tsa import TimeSeriesDataset, AutoEncForecast, train, evaluate
-from .config_forecasting import config
+from hydra.utils import instantiate
 
-device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-
-def parse_args():
-    """
-    Parse command line arguments.
+from tsa import AutoEncForecast, train, evaluate
+from tsa.utils import load_checkpoint
 
-    Args:
-    """
-    parser = argparse.ArgumentParser()
-    parser.add_argument("--batch-size", default=config["batch_size"], type=int, help="batch size")
-    parser.add_argument("--output-size", default=config["output_size"], type=int,
-                        help="size of the ouput: default value to 1 for forecasting")
-    parser.add_argument("--label-col", default=config["label_col"], type=str, help="name of the target column")
-    parser.add_argument("--input-att", default=config["input_att"], type=lambda x: (str(x).lower() == "true"),
-                        help="whether or not activate the input attention mechanism")
-    parser.add_argument("--temporal-att", default=config["temporal_att"], type=lambda x: (str(x).lower() == "true"),
-                        help="whether or not activate the temporal attention mechanism")
-    parser.add_argument("--seq-len", default=config["seq_len"], type=int, help="window length to use for forecasting")
-    parser.add_argument("--hidden-size-encoder", default=config["hidden_size_encoder"], type=int,
-                        help="size of the encoder's hidden states")
-    parser.add_argument("--hidden-size-decoder", default=config["hidden_size_decoder"], type=int,
-                        help="size of the decoder's hidden states")
-    parser.add_argument("--reg-factor1", default=config["reg_factor1"], type=float,
-                        help="contribution factor of the L1 regularization if using a sparse autoencoder")
-    parser.add_argument("--reg-factor2", default=config["reg_factor2"], type=float,
-                        help="contribution factor of the L2 regularization if using a sparse autoencoder")
-    parser.add_argument("--reg1", default=config["reg1"], type=lambda x: (str(x).lower() == "true"),
-                        help="activate/deactivate L1 regularization")
-    parser.add_argument("--reg2", default=config["reg2"], type=lambda x: (str(x).lower() == "true"),
-                        help="activate/deactivate L2 regularization")
-    parser.add_argument("--denoising", default=config["denoising"], type=lambda x: (str(x).lower() == "true"),
-                        help="whether or not to use a denoising autoencoder")
-    parser.add_argument("--do-train", default=True, type=lambda x: (str(x).lower() == "true"),
-                        help="whether or not to train the model")
-    parser.add_argument("--do-eval", default=False, type=lambda x: (str(x).lower() == "true"),
-                        help="whether or not evaluating the mode")
-    parser.add_argument("--output-dir", default=config["output_dir"], help="name of folder to output files")
-    parser.add_argument("--ckpt", default=None, help="checkpoint path for evaluation")
-    return parser.parse_args()
+device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 
 
-if __name__ == "__main__":
-    args = vars(parse_args())
-    config.update(args)
+@hydra.main(config_path="./", config_name="config")
+def run(cfg):
+    ts = instantiate(cfg.data)
+    train_iter, test_iter, nb_features = ts.get_loaders()
 
-    df = pd.read_csv("data/AirQualityUCI.csv", index_col=config["index_col"])
+    model = AutoEncForecast(cfg.training, input_size=nb_features).to(device)
+    criterion = nn.MSELoss()
+    optimizer = torch.optim.Adam(model.parameters(), lr=cfg.training.lr)
 
-    ts = TimeSeriesDataset(
-        data=df,
-        categorical_cols=config["categorical_cols"],
-        target_col=config["label_col"],
-        seq_length=config["seq_len"],
-        prediction_window=config["prediction_window"]
-    )
-    train_iter, test_iter, nb_features = ts.get_loaders(batch_size=config["batch_size"])
+    if cfg.general.do_train:
+        train(train_iter, test_iter, model, criterion, optimizer, cfg, ts)
+    if cfg.general.do_eval and cfg.general.get("ckpt", False):
+        model, _, loss, epoch = load_checkpoint(cfg.general.ckpt, model, optimizer, device)
+        evaluate(test_iter, loss, model, cfg, ts)
 
-    model = AutoEncForecast(config, input_size=nb_features).to(config["device"])
-    criterion = nn.MSELoss()
-    optimizer = torch.optim.Adam(model.parameters(), lr=config["lr"])
 
-    if config["do_eval"] and config["ckpt"]:
-        model, _, loss, epoch = load_checkpoint(config["ckpt"], model, optimizer, config["device"])
-        evaluate(test_iter, loss, model, config)
-    elif config["do_train"]:
-        train(train_iter, test_iter, model, criterion, optimizer, config)
+if __name__ == "__main__":
+    run()

examples/forecasting/run_forecasting.py

@@ -0,0 +1,40 @@
+data:
+  _target_: tsa.dataset.TimeSeriesDataset
+  batch_size: 16
+  categorical_cols: [ ]
+  index_col: "Date_Time"
+  target_col: [ ]
+  data_path: "../data/AirQualityUCI.csv"
+  prediction_window: 1
+  seq_length: 3
+  task:
+    _target_: tsa.dataset.Tasks
+    value: reconstruction
+
+training:
+  denoising: False
+  directions: 1
+  gradient_accumulation_steps: 1
+  hidden_size_encoder: 64
+  hidden_size_decoder: 64
+  input_att: True
+  lr: 1e-5
+  lrs_step_size: 5000
+  max_grad_norm: 0.1
+  num_epochs: 100
+  output_size: 13
+  reg1: True
+  reg2: False
+  reg_factor1: 1e-4
+  reg_factor2: 1e-4
+  seq_len: ${data.seq_length}
+  temporal_att: True
+
+general:
+  do_eval: True
+  do_train: True
+  logging_steps: 100
+
+  output_dir: "output"
+  save_steps: 5000
+  eval_during_training: True