Shourya Bose
commited on
Commit
•
4ad6488
1
Parent(s):
a9073bb
update readme
Browse files
README.md
CHANGED
@@ -22,6 +22,30 @@ Executing the code only requires `numpy` and `torch` (PyTorch) packages. You can
|
|
22 |
|
23 |
In order to see how to use the model definitions and load the weights into them, see `example.py`.
|
24 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
## Credits
|
26 |
|
27 |
Some model definitions have been adapted from the code provided in the [TSLib Library](https://github.com/thuml/Time-Series-Library).
|
|
|
22 |
|
23 |
In order to see how to use the model definitions and load the weights into them, see `example.py`.
|
24 |
|
25 |
+
## Technical Details
|
26 |
+
|
27 |
+
In input layout of the models are as follows:
|
28 |
+
|
29 |
+
- The `forward()` functions of `LSTM`, `LSTNet`, and `PatchTST` take in two arguments: `forward(input, future_time_idx)`. They are laid out as follows:
|
30 |
+
|
31 |
+
- `input` is a tensor of shape `(B,L,num_features)` where `B` is the batch size, `L` is the lookback duration, and `num_features` is 8 for our current application.
|
32 |
+
- `future_time_idx` is a tensor of shape `(B,T,2)` where `T` is the lookahead and 2 is the number of time index features.
|
33 |
+
- The time indices in `input` as well as `fut_time_idx` are both normalized.
|
34 |
+
- Non-time features are normalized. The mean and standard deviation of the [companion dataset](https://huggingface.co/datasets/APPFL/Illinois_load_datasets) can be inferred by executing `example_dataset.py` there and looking at `Case 1` and `Case 4`.
|
35 |
+
- The output shape is `(B,1)` denoting the pointwise forecast `T` steps into the future.
|
36 |
+
- The `forward()` functions of `Transformer`, `Autoformer`, `Informer`, and `TimesNet` take in two arguments:` forward(input, future_time_idx)`. They are laid out as follows:
|
37 |
+
|
38 |
+
- `input` is a tensor of shape `(B,L,num_features)` where `B` is the batch size, `L` is the lookback duration, and `num_features` is 8 for our current application.
|
39 |
+
- `future_time_idx` is a tensor of shape `(B,T,2)` where `T` is the lookahead and 2 is the number of time index features.
|
40 |
+
- The time indices in `input` as well as `fut_time_idx` are un-normalized to allow for embedding.
|
41 |
+
- Non-time features are normalized. The mean and standard deviation of the [companion dataset](https://huggingface.co/datasets/APPFL/Illinois_load_datasets) can be inferred by executing `example_dataset.py` there and looking at `Case 2` and `Case 5`.
|
42 |
+
- The output shape is `(B,1)` denoting the pointwise forecast `T` steps into the future.
|
43 |
+
- The `forward()` functions of `TimesFM` takes in one argument:` forward(input)`. It is laid out as follows:
|
44 |
+
|
45 |
+
- `input` is a tensor of shape `(B,L)` where `B` is the batch size and `L` is the lookback duration. Since it is univariate, there is only one feature.
|
46 |
+
- The sole feature is normalized. The mean and standard deviation of the [companion dataset](https://huggingface.co/datasets/APPFL/Illinois_load_datasets) can be inferred by executing `example_dataset.py` there and looking at `Case 3` and `Case 6`.
|
47 |
+
- The output shape is `(B,T)` denoting the rolling horizon forecast `T` steps into the future.
|
48 |
+
|
49 |
## Credits
|
50 |
|
51 |
Some model definitions have been adapted from the code provided in the [TSLib Library](https://github.com/thuml/Time-Series-Library).
|