diff --git a/README.md b/README.md
index b3a7d90..ca0e223 100644
--- a/README.md
+++ b/README.md
@@ -1,11 +1,12 @@
-
tsl (Torch Spatiotemporal) is a library built to accelerate research on neural spatiotemporal data processing
+
tsl (Torch Spatiotemporal) is a library built to accelerate research on neural spatiotemporal data processing
methods, with a focus on Graph Neural Networks.
- tsl is built on several libraries of the Python scientific computing ecosystem, with the final objective of providing a straightforward process that goes from data preprocessing to model prototyping.
-In particular, tsl offers a wide range of utilities to develop neural networks in 
PyTorch for processing spatiotemporal data signals.
+ tsl is built on several libraries of the Python scientific computing ecosystem, with the final objective of providing a straightforward process that goes from data preprocessing to model prototyping.
+In particular, tsl offers a wide range of utilities to develop neural networks in 
PyTorch for processing spatiotemporal data signals.
## Getting Started
-Before you start using tsl, please review the documentation to get an understanding of the library and its capabilities.
+Before you start using tsl, please review the documentation to get an understanding of the library and its capabilities.
You can also explore the examples provided in the `examples` directory to see how train deep learning models working with spatiotemporal data.
## Installation
-Before installing tsl, make sure you have installed PyTorch (>=1.9.0) and 
PyG (>=2.0.3) in your virtual environment (see [PyG installation guidelines](https://pytorch-geometric.readthedocs.io/en/latest/install/installation.html)). tsl is available for Python>=3.8. We recommend installation from github to be up-to-date with the latest version:
+Before installing tsl, make sure you have installed PyTorch (>=1.9.0) and 
PyG (>=2.0.3) in your virtual environment (see [PyG installation guidelines](https://pytorch-geometric.readthedocs.io/en/latest/install/installation.html)). tsl is available for Python>=3.8. We recommend installation from github to be up-to-date with the latest version:
```bash
pip install git+https://github.com/TorchSpatiotemporal/tsl.git
@@ -50,7 +51,7 @@ conda env create -f conda_env.yml
## Tutorial
-The best way to start using tsl is by following the tutorial notebook in `examples/notebooks/a_gentle_introduction_to_tsl.ipynb`.
+The best way to start using tsl is by following the tutorial notebook in `examples/notebooks/a_gentle_introduction_to_tsl.ipynb`.
## Documentation
@@ -73,7 +74,7 @@ If you use Torch Spatiotemporal for your research, please consider citing the li
By [Andrea Cini](https://andreacini.github.io/) and [Ivan Marisca](https://marshka.github.io/).
-Thanks to all contributors! Check the [Contributing guidelines](https://github.com/TorchSpatiotemporal/tsl/blob/dev/.github/CONTRIBUTING.md) and help us build a better tsl.
+Thanks to all contributors! Check the [Contributing guidelines](https://github.com/TorchSpatiotemporal/tsl/blob/dev/.github/CONTRIBUTING.md) and help us build a better tsl.
diff --git a/examples/imputation/run_imputation_experiment.py b/examples/imputation/run_imputation_experiment.py
index 8232f3c..0017f52 100644
--- a/examples/imputation/run_imputation_experiment.py
+++ b/examples/imputation/run_imputation_experiment.py
@@ -164,12 +164,14 @@ def run_imputation(cfg: DictConfig):
mode='min',
)
- trainer = Trainer(max_epochs=cfg.epochs,
- default_root_dir=cfg.run.dir,
- logger=exp_logger,
- gpus=1 if torch.cuda.is_available() else None,
- gradient_clip_val=cfg.grad_clip_val,
- callbacks=[early_stop_callback, checkpoint_callback])
+ trainer = Trainer(
+ max_epochs=cfg.epochs,
+ default_root_dir=cfg.run.dir,
+ logger=exp_logger,
+ accelerator='gpu' if torch.cuda.is_available() else 'cpu',
+ devices=1,
+ gradient_clip_val=cfg.grad_clip_val,
+ callbacks=[early_stop_callback, checkpoint_callback])
trainer.fit(imputer, datamodule=dm)
diff --git a/tsl/datasets/pems_bay.py b/tsl/datasets/pems_bay.py
index 0a9b2a5..4c3162c 100644
--- a/tsl/datasets/pems_bay.py
+++ b/tsl/datasets/pems_bay.py
@@ -77,11 +77,8 @@ def load_raw(self):
# load traffic data
traffic_path = os.path.join(self.root_dir, 'pems_bay.h5')
df = pd.read_hdf(traffic_path)
- # add missing values
- datetime_idx = sorted(df.index)
- date_range = pd.date_range(datetime_idx[0],
- datetime_idx[-1],
- freq='5T')
+ # add missing values (index is sorted)
+ date_range = pd.date_range(df.index[0], df.index[-1], freq='5T')
df = df.reindex(index=date_range)
# load distance matrix
path = os.path.join(self.root_dir, 'pems_bay_dist.npy')
diff --git a/tsl/datasets/prototypes/casting.py b/tsl/datasets/prototypes/casting.py
index 45a3421..47bba37 100644
--- a/tsl/datasets/prototypes/casting.py
+++ b/tsl/datasets/prototypes/casting.py
@@ -99,4 +99,4 @@ def time_unit_to_nanoseconds(time_unit: str):
return 365.2425 * 24 * 60 * 60 * 10**9
elif time_unit == 'week':
time_unit = 'W'
- return pd.Timedelta('1' + time_unit).delta
+ return pd.Timedelta('1' + time_unit).value
diff --git a/tsl/metrics/torch/metric_base.py b/tsl/metrics/torch/metric_base.py
index 58fe948..f187f03 100644
--- a/tsl/metrics/torch/metric_base.py
+++ b/tsl/metrics/torch/metric_base.py
@@ -65,10 +65,9 @@ def __init__(self,
if metric_fn_kwargs is None:
metric_fn_kwargs = dict()
- if metric_fn is None:
- self.metric_fn = None
- else:
- self.metric_fn = partial(metric_fn, **metric_fn_kwargs)
+
+ self.metric_fn = partial(metric_fn, **metric_fn_kwargs)
+
self.mask_nans = mask_nans
self.mask_inf = mask_inf
if at is None:
diff --git a/tsl/nn/layers/base/embedding.py b/tsl/nn/layers/base/embedding.py
index 8e1ed68..2df8e8c 100644
--- a/tsl/nn/layers/base/embedding.py
+++ b/tsl/nn/layers/base/embedding.py
@@ -63,19 +63,18 @@ def get_emb(self):
def forward(self,
expand: Optional[List] = None,
- token_index: OptTensor = None,
- tokens_first: bool = True):
+ node_index: OptTensor = None,
+ nodes_first: bool = True):
""""""
emb = self.get_emb()
- if token_index is not None:
- emb = emb[token_index]
- if not tokens_first:
+ if node_index is not None:
+ emb = emb[node_index]
+ if not nodes_first:
emb = emb.T
if expand is None:
return emb
shape = [*emb.size()]
view = [
- 1 if d > 0 else shape.pop(0 if tokens_first else -1)
- for d in expand
+ 1 if d > 0 else shape.pop(0 if nodes_first else -1) for d in expand
]
return emb.view(*view).expand(*expand)
+ 
+
