[FEAT] TiDE model

action_files/test_models/src/evaluation.py

@@ -42,17 +42,18 @@ def evaluate(model: str, dataset: str, group: str):
 if __name__ == '__main__':
     groups = ['Monthly']
     models = ['AutoDilatedRNN', 'RNN', 'TCN', 'DeepAR',
-              'NHITS', 'TFT', 'AutoMLP', 'DLinear', 'VanillaTransformer']
+              'NHITS', 'TFT', 'AutoMLP', 'DLinear', 'VanillaTransformer',
+              'BiTCN', 'TiDE']
     datasets = ['M3']
     evaluation = [evaluate(model, dataset, group) for model, group in product(models, groups) for dataset in datasets]
     evaluation = [eval_ for eval_ in evaluation if eval_ is not None]
-    evaluation = pd.concat(evaluation)
-    evaluation = evaluation[['dataset', 'model', 'time', 'mae', 'smape']]
-    evaluation['time'] /= 60 #minutes
-    evaluation = evaluation.set_index(['dataset', 'model']).stack().reset_index()
-    evaluation.columns = ['dataset', 'model', 'metric', 'val']
-    evaluation = evaluation.set_index(['dataset', 'metric', 'model']).unstack().round(3)
-    evaluation = evaluation.droplevel(0, 1).reset_index()
-    evaluation['AutoARIMA'] = [666.82, 15.35, 3.000]
-    evaluation.to_csv('data/evaluation.csv')
-    print(evaluation.T)
+    df_evaluation = pd.concat(evaluation)
+    df_evaluation = df_evaluation.loc[:, ['dataset', 'model', 'time', 'mae', 'smape']]
+    df_evaluation['time'] /= 60 #minutes
+    df_evaluation = df_evaluation.set_index(['dataset', 'model']).stack().reset_index()
+    df_evaluation.columns = ['dataset', 'model', 'metric', 'val']
+    df_evaluation = df_evaluation.set_index(['dataset', 'metric', 'model']).unstack().round(3)
+    df_evaluation = df_evaluation.droplevel(0, 1).reset_index()
+    df_evaluation['AutoARIMA'] = [666.82, 15.35, 3.000]
+    df_evaluation.to_csv('data/evaluation.csv')
+    print(df_evaluation.T)

action_files/test_models/src/models.py

@@ -2,33 +2,39 @@
 import time
 
 import fire
-import numpy as np
+# import numpy as np
 import pandas as pd
-import pytorch_lightning as pl
-import torch
+# import pytorch_lightning as pl
+# import torch
 
-import neuralforecast
+# import neuralforecast
 from neuralforecast.core import NeuralForecast
 
-from neuralforecast.models.gru import GRU
+# from neuralforecast.models.gru import GRU
 from neuralforecast.models.rnn import RNN
 from neuralforecast.models.tcn import TCN
-from neuralforecast.models.lstm import LSTM
-from neuralforecast.models.dilated_rnn import DilatedRNN
+# from neuralforecast.models.lstm import LSTM
+# from neuralforecast.models.dilated_rnn import DilatedRNN
 from neuralforecast.models.deepar import DeepAR
-from neuralforecast.models.mlp import MLP
+# from neuralforecast.models.mlp import MLP
 from neuralforecast.models.nhits import NHITS
-from neuralforecast.models.nbeats import NBEATS
-from neuralforecast.models.nbeatsx import NBEATSx
+# from neuralforecast.models.nbeats import NBEATS
+# from neuralforecast.models.nbeatsx import NBEATSx
 from neuralforecast.models.tft import TFT
 from neuralforecast.models.vanillatransformer import VanillaTransformer
-from neuralforecast.models.informer import Informer
-from neuralforecast.models.autoformer import Autoformer
-from neuralforecast.models.patchtst import PatchTST
+# from neuralforecast.models.informer import Informer
+# from neuralforecast.models.autoformer import Autoformer
+# from neuralforecast.models.patchtst import PatchTST
 from neuralforecast.models.dlinear import DLinear
+from neuralforecast.models.bitcn import BiTCN   
+from neuralforecast.models.tide import TiDE
 
 from neuralforecast.auto import (
-    AutoMLP, AutoNHITS, AutoNBEATS, AutoDilatedRNN, AutoTFT
+    AutoMLP, 
+    # AutoNHITS, 
+    # AutoNBEATS, 
+    AutoDilatedRNN, 
+    # AutoTFT
 )
 
 from neuralforecast.losses.pytorch import SMAPE, MAE
@@ -43,13 +49,6 @@ def main(dataset: str = 'M3', group: str = 'Monthly') -> None:
     train, horizon, freq, seasonality = get_data('data/', dataset, group)
     train['ds'] = pd.to_datetime(train['ds'])
 
-    config_nbeats = {
-        "input_size": tune.choice([2 * horizon]),
-        "max_steps": 1000,
-        "val_check_steps": 300,
-        "scaler_type": "minmax1",
-        "random_seed": tune.choice([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]),
-    }
     config = {
         "hidden_size": tune.choice([256, 512]),
         "num_layers": tune.choice([2, 4]),
@@ -64,6 +63,7 @@ def main(dataset: str = 'M3', group: str = 'Monthly') -> None:
                    "max_steps": 300,
                    "val_check_steps": 100,
                    "random_seed": tune.choice([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]),}
+
     models = [
         AutoDilatedRNN(h=horizon, loss=MAE(), config=config_drnn, num_samples=2, cpus=1),
         RNN(h=horizon, input_size=2 * horizon, encoder_hidden_size=50, max_steps=300),
@@ -74,10 +74,12 @@ def main(dataset: str = 'M3', group: str = 'Monthly') -> None:
         TFT(h=horizon, input_size=2 * horizon, loss=SMAPE(), hidden_size=64, scaler_type='robust', windows_batch_size=512, max_steps=1500, val_check_steps=500),
         VanillaTransformer(h=horizon, input_size=2 * horizon, loss=MAE(), hidden_size=64, scaler_type='minmax1', windows_batch_size=512, max_steps=1500, val_check_steps=500),
         DeepAR(h=horizon, input_size=2 * horizon, scaler_type='minmax1', max_steps=1000),
+        BiTCN(h=horizon, input_size=2 * horizon, loss=MAE(), dropout=0.0, max_steps=1000, val_check_steps=500),
+        TiDE(h=horizon, input_size=2 * horizon, loss=MAE(), max_steps=1000, val_check_steps=500),
     ]
 
     # Models
-    for model in models[:-1]:
+    for model in models:
         model_name = type(model).__name__
         print(50*'-', model_name, 50*'-')
         start = time.time()
@@ -87,26 +89,13 @@ def main(dataset: str = 'M3', group: str = 'Monthly') -> None:
         end = time.time()
         print(end - start)
 
+        if model_name == 'DeepAR':
+            forecasts = forecasts[['unique_id', 'ds', 'DeepAR-median']]
+
         forecasts.columns = ['unique_id', 'ds', model_name]
         forecasts.to_csv(f'data/{model_name}-forecasts-{dataset}-{group}.csv', index=False)
         time_df = pd.DataFrame({'time': [end - start], 'model': [model_name]})
         time_df.to_csv(f'data/{model_name}-time-{dataset}-{group}.csv', index=False)
 
-    # DeepAR
-    model_name = type(models[-1]).__name__
-    start = time.time()
-    fcst = NeuralForecast(models=[models[-1]], freq=freq)
-    fcst.fit(train)
-    forecasts = fcst.predict()
-    end = time.time()
-    print(end - start)
-
-    forecasts = forecasts[['unique_id', 'ds', 'DeepAR-median']]
-    forecasts.columns = ['unique_id', 'ds', 'DeepAR']
-    forecasts.to_csv(f'data/{model_name}-forecasts-{dataset}-{group}.csv', index=False)
-    time_df = pd.DataFrame({'time': [end - start], 'model': [model_name]})
-    time_df.to_csv(f'data/{model_name}-time-{dataset}-{group}.csv', index=False)
-
-
 if __name__ == '__main__':
     fire.Fire(main)

nbs/core.ipynb

@@ -91,7 +91,7 @@
     "    Informer, Autoformer, FEDformer,\n",
     "    StemGNN, PatchTST, TimesNet, TimeLLM, TSMixer, TSMixerx,\n",
     "    MLPMultivariate, iTransformer,\n",
-    "    BiTCN,\n",
+    "    BiTCN, TiDE\n",
     ")"
    ]
   },
@@ -238,6 +238,7 @@
     "    'mlpmultivariate': MLPMultivariate, 'automlpmultivariate': MLPMultivariate,\n",
     "    'itransformer': iTransformer, 'autoitransformer': iTransformer,\n",
     "    'bitcn': BiTCN, 'autobitcn': BiTCN,\n",
+    "    'tide': TiDE, 'autotide': TiDE,\n",
     "}"
    ]
   },