Fix/autoformer

CONTRIBUTING.md

@@ -55,8 +55,8 @@ You can preview changes in your local browser before pushing by using the `nbdev
 ### Building the library
 The library is built using the notebooks contained in the `nbs` folder. If you want to make any changes to the library you have to find the relevant notebook, make your changes and then call:
 ```
- nbdev_export
- ```
+nbdev_export
+```
 
 ### Linters
 This project uses a couple of linters to validate different aspects of the code. Before opening a PR, please make sure that it passes all the linting tasks by following the next steps.

nbs/examples/Getting_Started.ipynb

@@ -206,8 +206,8 @@
     "                max_steps=100,               # Number of steps to train\n",
     "                n_freq_downsample=[2, 1, 1]) # Downsampling factors for each stack output\n",
     "          ]\n",
-    "fcst = NeuralForecast(models=models, freq='M')\n",
-    "fcst.fit(df=Y_df)"
+    "nf = NeuralForecast(models=models, freq='M')\n",
+    "nf.fit(df=Y_df)"
    ]
   },
   {
@@ -243,7 +243,7 @@
     }
    ],
    "source": [
-    "Y_hat_df = fcst.predict()"
+    "Y_hat_df = nf.predict()"
    ]
   },
   {

nbs/examples/LongHorizon_Probabilistic.ipynb

@@ -502,7 +502,7 @@
    "outputs": [],
    "source": [
     "# Fit and predict\n",
-    "fcst = NeuralForecast(\n",
+    "nf = NeuralForecast(\n",
     "    models=models,\n",
     "    freq='15min')"
    ]
@@ -525,8 +525,8 @@
    "outputs": [],
    "source": [
     "%%capture\n",
-    "fcst_df = fcst.cross_validation(df=Y_df, val_size=val_size,\n",
-    "                                test_size=test_size, n_windows=None)"
+    "Y_hat_df = nf.cross_validation(df=Y_df, val_size=val_size,\n",
+    "                               test_size=test_size, n_windows=None)"
    ]
   },
   {
@@ -549,7 +549,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "Y_hat_df = fcst_df.reset_index(drop=True)\n",
+    "Y_hat_df = Y_hat_df.reset_index(drop=True)\n",
     "Y_hat_df = Y_hat_df[(Y_hat_df['unique_id']=='OT') & (Y_hat_df['cutoff']=='2018-02-11 12:00:00')]\n",
     "Y_hat_df = Y_hat_df.drop(columns=['y','cutoff'])"
    ]
@@ -596,13 +596,19 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "metadata": {},
    "source": [
     "## References\n",
     "\n",
-    "[Cristian Challu, Kin G. Olivares, Boris N. Oreshkin, Federico Garza, Max Mergenthaler-Canseco, Artur Dubrawski (2021). N-HiTS: Neural Hierarchical Interpolation for Time Series Forecasting. Accepted at AAAI 2023.](https://arxiv.org/abs/2201.12886)"
+    "[Cristian Challu, Kin G. Olivares, Boris N. Oreshkin, Federico Garza, Max Mergenthaler-Canseco, Artur Dubrawski (2021). NHITS: Neural Hierarchical Interpolation for Time Series Forecasting. Accepted at AAAI 2023.](https://arxiv.org/abs/2201.12886)"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": []
   }
  ],
  "metadata": {

nbs/examples/LongHorizon_with_NHITS.ipynb

@@ -430,12 +430,12 @@
    ],
    "source": [
     "%%capture\n",
-    "fcst = NeuralForecast(\n",
+    "nf = NeuralForecast(\n",
     "    models=models,\n",
     "    freq='15min')\n",
     "\n",
-    "fcst_df = fcst.cross_validation(df=Y_df, val_size=val_size,\n",
-    "                                test_size=test_size, n_windows=None)"
+    "Y_hat_df = nf.cross_validation(df=Y_df, val_size=val_size,\n",
+    "                               test_size=test_size, n_windows=None)"
    ]
   },
   {
@@ -478,7 +478,7 @@
     }
    ],
    "source": [
-    "fcst.models[0].results.get_best_result().config"
+    "nf.models[0].results.get_best_result().config"
    ]
   },
   {
@@ -497,8 +497,8 @@
     }
    ],
    "source": [
-    "y_true = fcst_df.y.values\n",
-    "y_hat = fcst_df['AutoNHITS'].values\n",
+    "y_true = Y_hat_df.y.values\n",
+    "y_hat = Y_hat_df['AutoNHITS'].values\n",
     "\n",
     "n_series = len(Y_df.unique_id.unique())\n",
     "\n",
@@ -601,13 +601,19 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "metadata": {},
    "source": [
     "## References\n",
     "\n",
-    "[Cristian Challu, Kin G. Olivares, Boris N. Oreshkin, Federico Garza, Max Mergenthaler-Canseco, Artur Dubrawski (2021). N-HiTS: Neural Hierarchical Interpolation for Time Series Forecasting. Accepted at AAAI 2023.](https://arxiv.org/abs/2201.12886)"
+    "[Cristian Challu, Kin G. Olivares, Boris N. Oreshkin, Federico Garza, Max Mergenthaler-Canseco, Artur Dubrawski (2021). NHITS: Neural Hierarchical Interpolation for Time Series Forecasting. Accepted at AAAI 2023.](https://arxiv.org/abs/2201.12886)"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": []
   }
  ],
  "metadata": {

nbs/examples/Transfer_Learning.ipynb

@@ -159,8 +159,8 @@
     "                batch_size = 32,\n",
     "                scaler_type='standard',\n",
     "                n_freq_downsample=[12,4,1])]\n",
-    "fcst = NeuralForecast(models=models, freq='M')\n",
-    "fcst.fit(df=Y_df)"
+    "nf = NeuralForecast(models=models, freq='M')\n",
+    "nf.fit(df=Y_df)"
    ]
   },
   {
@@ -176,7 +176,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "fcst.save(path='./results/transfer/', model_index=None, overwrite=True, save_dataset=False)"
+    "nf.save(path='./results/transfer/', model_index=None, overwrite=True, save_dataset=False)"
    ]
   },
   {

nbs/examples/UncertaintyIntervals.ipynb

@@ -269,7 +269,7 @@
     "          NHITS(input_size=7*horizon, h=horizon,\n",
     "                n_freq_downsample=[24, 12, 1],\n",
     "                loss=MQLoss(level=levels), max_epochs=200),]\n",
-    "fcst = NeuralForecast(models=models, freq='H')"
+    "nf = NeuralForecast(models=models, freq='H')"
    ]
   },
   {
@@ -288,7 +288,7 @@
    "outputs": [],
    "source": [
     "%%capture\n",
-    "fcst.fit(df=Y_train_df)"
+    "nf.fit(df=Y_train_df)"
    ]
   },
   {
@@ -469,9 +469,9 @@
     }
    ],
    "source": [
-    "forecasts = fcst.predict()\n",
-    "forecasts = forecasts.reset_index()\n",
-    "forecasts.head()"
+    "Y_hat_df = nf.predict()\n",
+    "Y_hat_df = Y_hat_df.reset_index()\n",
+    "Y_hat_df.head()"
    ]
   },
   {
@@ -481,7 +481,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "Y_test_df = Y_test_df.merge(forecasts, how='left', on=['unique_id', 'ds'])"
+    "Y_test_df = Y_test_df.merge(Y_hat_df, how='left', on=['unique_id', 'ds'])"
    ]
   },
   {
@@ -561,14 +561,15 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "id": "cb637d8d-2823-4a03-85f8-221636861d2e",
    "metadata": {},
    "source": [
     "## References\n",
     "- [Roger Koenker and Gilbert Basset (1978). Regression Quantiles, Econometrica.](https://www.jstor.org/stable/1913643)<br>\n",
     "- [Jeffrey L. Elman (1990). \"Finding Structure in Time\".](https://onlinelibrary.wiley.com/doi/abs/10.1207/s15516709cog1402_1)<br>\n",
-    "- [Cristian Challu, Kin G. Olivares, Boris N. Oreshkin, Federico Garza, Max Mergenthaler-Canseco, Artur Dubrawski (2022). \"N-HiTS: Neural Hierarchical Interpolation for Time Series Forecasting\".](https://arxiv.org/abs/2201.12886)<br>"
+    "- [Cristian Challu, Kin G. Olivares, Boris N. Oreshkin, Federico Garza, Max Mergenthaler-Canseco, Artur Dubrawski (2022). \"NHITS: Neural Hierarchical Interpolation for Time Series Forecasting\".](https://arxiv.org/abs/2201.12886)<br>"
    ]
   },
   {

nbs/models.autoformer.ipynb

@@ -633,7 +633,7 @@
     "            x_mark_enc = None\n",
     "            x_mark_dec = None\n",
     "\n",
-    "        x_dec = torch.zeros(size=(len(insample_y),self.h,1))\n",
+    "        x_dec = torch.zeros(size=(len(insample_y),self.h,1)).to(insample_y.device)\n",
     "        x_dec = torch.cat([insample_y[:,-self.label_len:,:], x_dec], dim=1)\n",
     "\n",
     "        # decomp init\n",

nbs/models.informer.ipynb

@@ -442,7 +442,7 @@
     "            x_mark_enc = None\n",
     "            x_mark_dec = None\n",
     "\n",
-    "        x_dec = torch.zeros(size=(len(insample_y),self.h,1))\n",
+    "        x_dec = torch.zeros(size=(len(insample_y),self.h,1)).to(insample_y.device)\n",
     "        x_dec = torch.cat([insample_y[:,-self.label_len:,:], x_dec], dim=1)        \n",
     "\n",
     "        enc_out = self.enc_embedding(insample_y, x_mark_enc)\n",

nbs/models.nhits.ipynb

@@ -28,6 +28,7 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "metadata": {},
    "source": [
@@ -38,15 +39,16 @@
     "\n",
     "**References**<br>\n",
     "-[Boris N. Oreshkin, Dmitri Carpov, Nicolas Chapados, Yoshua Bengio (2019). \"N-BEATS: Neural basis expansion analysis for interpretable time series forecasting\".](https://arxiv.org/abs/1905.10437)<br>\n",
-    "-[Cristian Challu, Kin G. Olivares, Boris N. Oreshkin, Federico Garza, Max Mergenthaler-Canseco, Artur Dubrawski (2022). \"N-HiTS: Neural Hierarchical Interpolation for Time Series Forecasting\". Accepted at the Thirty-Seventh AAAI Conference on Artificial Intelligence.](https://arxiv.org/abs/2201.12886)<br>\n",
+    "-[Cristian Challu, Kin G. Olivares, Boris N. Oreshkin, Federico Garza, Max Mergenthaler-Canseco, Artur Dubrawski (2022). \"NHITS: Neural Hierarchical Interpolation for Time Series Forecasting\". Accepted at the Thirty-Seventh AAAI Conference on Artificial Intelligence.](https://arxiv.org/abs/2201.12886)<br>\n",
     "-[Zhou, H.; Zhang, S.; Peng, J.; Zhang, S.; Li, J.; Xiong, H.; and Zhang, W. (2020). \"Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting\". Association for the Advancement of Artificial Intelligence Conference 2021 (AAAI 2021).](https://arxiv.org/abs/2012.07436)"
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "![Figure 1. Neural Hierarchical Interpolation for Time Series (N-HITS).](imgs_models/nhits.png)"
+    "![Figure 1. Neural Hierarchical Interpolation for Time Series (NHITS).](imgs_models/nhits.png)"
    ]
   },
   {
@@ -162,7 +164,7 @@
     "\n",
     "class NHITSBlock(nn.Module):\n",
     "    \"\"\"\n",
-    "    N-HiTS block which takes a basis function as an argument.\n",
+    "    NHITS block which takes a basis function as an argument.\n",
     "    \"\"\"\n",
     "    def __init__(self, \n",
     "                 input_size: int,\n",
@@ -297,7 +299,7 @@
     "\n",
     "    **References:**<br>\n",
     "    -[Cristian Challu, Kin G. Olivares, Boris N. Oreshkin, Federico Garza, \n",
-    "    Max Mergenthaler-Canseco, Artur Dubrawski (2022). \"N-HiTS: Neural Hierarchical Interpolation for Time Series Forecasting\".\n",
+    "    Max Mergenthaler-Canseco, Artur Dubrawski (2022). \"NHITS: Neural Hierarchical Interpolation for Time Series Forecasting\".\n",
     "    Accepted at the Thirty-Seventh AAAI Conference on Artificial Intelligence.](https://arxiv.org/abs/2201.12886)\n",
     "    \"\"\"\n",
     "    def __init__(self, \n",
@@ -523,7 +525,7 @@
     "              pooling_mode='MaxPool1d')\n",
     "model.fit(dataset=dataset)\n",
     "y_hat = model.predict(dataset=dataset)\n",
-    "Y_test_df['N-HiTS'] = y_hat\n",
+    "Y_test_df['NHITS'] = y_hat\n",
     "\n",
     "pd.concat([Y_train_df, Y_test_df]).drop('unique_id', axis=1).set_index('ds').plot()"
    ]

neuralforecast/models/autoformer.py

@@ -655,7 +655,7 @@ def forward(self, windows_batch):
             x_mark_enc = None
             x_mark_dec = None
 
-        x_dec = torch.zeros(size=(len(insample_y), self.h, 1))
+        x_dec = torch.zeros(size=(len(insample_y), self.h, 1)).to(insample_y.device)
         x_dec = torch.cat([insample_y[:, -self.label_len :, :], x_dec], dim=1)
 
         # decomp init

neuralforecast/models/informer.py

@@ -392,7 +392,7 @@ def forward(self, windows_batch):
             x_mark_enc = None
             x_mark_dec = None
 
-        x_dec = torch.zeros(size=(len(insample_y), self.h, 1))
+        x_dec = torch.zeros(size=(len(insample_y), self.h, 1)).to(insample_y.device)
         x_dec = torch.cat([insample_y[:, -self.label_len :, :], x_dec], dim=1)
 
         enc_out = self.enc_embedding(insample_y, x_mark_enc)

neuralforecast/models/nhits.py

@@ -71,7 +71,7 @@ def forward(self, theta: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
 
 class NHITSBlock(nn.Module):
     """
-    N-HiTS block which takes a basis function as an argument.
+    NHITS block which takes a basis function as an argument.
     """
 
     def __init__(
@@ -216,7 +216,7 @@ class NHITS(BaseWindows):
 
     **References:**<br>
     -[Cristian Challu, Kin G. Olivares, Boris N. Oreshkin, Federico Garza,
-    Max Mergenthaler-Canseco, Artur Dubrawski (2022). "N-HiTS: Neural Hierarchical Interpolation for Time Series Forecasting".
+    Max Mergenthaler-Canseco, Artur Dubrawski (2022). "NHITS: Neural Hierarchical Interpolation for Time Series Forecasting".
     Accepted at the Thirty-Seventh AAAI Conference on Artificial Intelligence.](https://arxiv.org/abs/2201.12886)
     """