Clean modular torch training scripts

DLinearTrain.py → DLinearTrainDisagg.py

@@ -17,7 +17,6 @@
 model_summary = RichModelSummary(max_depth = -1)
 
 
-
 # Specify D-Linear model
 model_dlinear = DLinear(
   input_chunk_length = 30,
@@ -42,7 +41,6 @@
 
 # All covariates, future & past
 dlinear2_futcovars = ['tuesday', 'wednesday', 'thursday', 'friday', 'saturday', 'sunday', 'day_sin', 'day_cos', "month_sin", "month_cos", 'oil', 'oil_ma28', 'onpromotion', 'onp_ma28', 'local_holiday', 'regional_holiday', 'national_holiday', 'ny1', 'ny2', 'ny_eve31', 'ny_eve30', 'xmas_before', 'xmas_after', 'quake_after', 'dia_madre', 'futbol', 'black_friday', 'cyber_monday']
-
 dlinear2_pastcovars = ["sales_ema7", "transactions", "trns_ma7"]
 
 

DataPrepDisagg.py

@@ -1,3 +1,5 @@
+
+
 # Initialize list of disagg covariates
 disagg_covars = []
 

DataPrepMain.py

@@ -1,12 +1,28 @@
 # Import libraries
 import pandas as pd 
 import numpy as np
-import matplotlib.pyplot as plt
-import seaborn as sns
 import torch
 import tensorboard
 from tqdm import tqdm
 
+# Import transformers
+from sklearn.preprocessing import StandardScaler
+from darts.dataprocessing.transformers import Scaler
+from darts.dataprocessing.transformers import MissingValuesFiller
+from sktime.transformations.series.difference import Differencer
+
+# Import forecasting models
+from darts.models.forecasting.dlinear import DLinearModel as DLinear
+from darts.models.forecasting.rnn_model import RNNModel as RNN
+from darts.models.forecasting.tft_model import TFTModel
+
+# Import utils
+from darts import TimeSeries
+from darts.utils.timeseries_generation import datetime_attribute_timeseries
+from itertools import product
+from functools import reduce
+
+
 # Set printing options
 np.set_printoptions(suppress=True, precision=4)
 pd.options.display.float_format = '{:.4f}'.format
@@ -230,17 +246,12 @@
 category_store_nbr = category_store_nbr.pivot(columns="category_store_nbr", values="sales")
 
 # Merge all wide dataframes
-from functools import reduce
 wide_frames = [total, store_nbr, category_store_nbr]
 df_sales = reduce(lambda left, right: pd.merge(
   left, right, how="left", on="date"), wide_frames)
 df_sales = df_sales.rename(columns = {"sales":"TOTAL"})
 del total, store_nbr, wide_frames, category_store_nbr
 
-
-from darts import TimeSeries
-from itertools import product
-
 # Create multivariate time series with sales components
 ts_sales = TimeSeries.from_dataframe(df_sales, freq="D")
 
@@ -268,13 +279,12 @@
 
 
 # Fill gaps
-from darts.dataprocessing.transformers import MissingValuesFiller
 na_filler = MissingValuesFiller()
 ts_sales = na_filler.transform(ts_sales)
 
 
 # Create differencer
-from sktime.transformations.series.difference import Differencer
+
 diff = Differencer(lags = 1)
 
 
@@ -310,9 +320,6 @@
 total_covars1 = ts_totalcovars1.pd_dataframe()
 
 
-from darts.utils.timeseries_generation import datetime_attribute_timeseries
-
-
 # Retrieve copy of total_covars1, drop Fourier terms, trend knot (leaving daily predictors common to all categories).
 common_covars = total_covars1[total_covars1.columns[2:21].values.tolist()]
 
@@ -332,23 +339,3 @@
 
 
 
-# Import transformers
-from sklearn.preprocessing import StandardScaler
-from darts.dataprocessing.transformers import Scaler
-
-# Import baseline models
-from darts.models.forecasting.baselines import NaiveDrift, NaiveSeasonal
-from darts.models.forecasting.sf_ets import StatsForecastETS as ETS
-
-# Import forecasting models
-from darts.models.forecasting.linear_regression_model import LinearRegressionModel
-from darts.models.forecasting.auto_arima import AutoARIMA
-from darts.models.forecasting.random_forest import RandomForest
-from darts.models.forecasting.xgboost import XGBModel
-from darts.models.forecasting.dlinear import DLinearModel as DLinear
-from darts.models.forecasting.rnn_model import RNNModel as RNN
-
-# Import time decomposition functions
-from darts.utils.statistics import extract_trend_and_seasonality as decomposition
-from darts.utils.statistics import remove_from_series
-from darts.utils.utils import ModelMode, SeasonalityMode

DataPrepStore.py

@@ -1,7 +1,5 @@
 
 
-
-
 # Initialize list of store covariates
 store_covars = []
 

JunkCode.py

@@ -1,3 +1,7 @@
+exec(open("test2.py").read())
+
+
+
 Sys.setenv(QUARTO_PYTHON="./venv/Scripts/python.exe")
 
 print(np.isnan(series1.values()).sum())

TFTTrainStore.py

@@ -0,0 +1,198 @@
+
+from pytorch_lightning.callbacks.early_stopping import EarlyStopping
+from pytorch_lightning.callbacks import RichProgressBar, RichModelSummary
+
+# Create early stopper
+early_stopper = EarlyStopping(
+  monitor = "val_loss",
+  min_delta = 5000, # 1% of min. MSE of best model so far
+  patience = 10
+)
+
+# Progress bar
+progress_bar = RichProgressBar()
+
+# Rich model summary
+model_summary = RichModelSummary(max_depth = -1)
+
+
+# # Specify TFT model 2.0 (TFT specific params all default)
+# model_tft = TFTModel(
+#   input_chunk_length = 30,
+#   output_chunk_length = 15,
+#   hidden_size = 16,
+#   lstm_layers = 1,
+#   num_attention_heads = 4,
+#   dropout = 0.1,
+#   hidden_continuous_size = 8,
+#   batch_size = 32,
+#   n_epochs = 500,
+#   likelihood = None,
+#   loss_fn = torch.nn.MSELoss(),
+#   model_name = "TFTStoreX",
+#   log_tensorboard = True,
+#   save_checkpoints = True,
+#   show_warnings = True,
+#   optimizer_kwargs = {"lr": 0.002},
+#   lr_scheduler_cls = torch.optim.lr_scheduler.ReduceLROnPlateau,
+#   lr_scheduler_kwargs = {"patience": 5},
+#   pl_trainer_kwargs = {
+#     "callbacks": [early_stopper],
+#     "accelerator": "gpu",
+#     "devices": [0]
+#     }
+# )
+
+
+# # Specify TFT model 2.1 (TFT specific params all default, local-regional fix,
+# # higher initial LR)
+# model_tft = TFTModel(
+#   input_chunk_length = 30,
+#   output_chunk_length = 15,
+#   hidden_size = 16,
+#   lstm_layers = 1,
+#   num_attention_heads = 4,
+#   dropout = 0.1,
+#   hidden_continuous_size = 8,
+#   batch_size = 32,
+#   n_epochs = 500,
+#   likelihood = None,
+#   loss_fn = torch.nn.MSELoss(),
+#   model_name = "TFTStore2.1",
+#   log_tensorboard = True,
+#   save_checkpoints = True,
+#   show_warnings = True,
+#   optimizer_kwargs = {"lr": 0.005},
+#   lr_scheduler_cls = torch.optim.lr_scheduler.ReduceLROnPlateau,
+#   lr_scheduler_kwargs = {"patience": 5},
+#   pl_trainer_kwargs = {
+#     "callbacks": [early_stopper],
+#     "accelerator": "gpu",
+#     "devices": [0]
+#     }
+# )
+
+
+# # Specify TFT model 2.2 (TFT specific params all default, local-regional fix,
+# # initial LR 0.003)
+# model_tft = TFTModel(
+#   input_chunk_length = 30,
+#   output_chunk_length = 15,
+#   hidden_size = 16,
+#   lstm_layers = 1,
+#   num_attention_heads = 4,
+#   dropout = 0.1,
+#   hidden_continuous_size = 8,
+#   batch_size = 32,
+#   n_epochs = 500,
+#   likelihood = None,
+#   loss_fn = torch.nn.MSELoss(),
+#   model_name = "TFTStore2.2",
+#   log_tensorboard = True,
+#   save_checkpoints = True,
+#   show_warnings = True,
+#   optimizer_kwargs = {"lr": 0.003},
+#   lr_scheduler_cls = torch.optim.lr_scheduler.ReduceLROnPlateau,
+#   lr_scheduler_kwargs = {"patience": 5},
+#   pl_trainer_kwargs = {
+#     "callbacks": [early_stopper],
+#     "accelerator": "gpu",
+#     "devices": [0]
+#     }
+# )
+
+
+# Specify TFT model 2.3 (TFT specific params all default, local-regional fix,
+# # initial LR 0.002)
+model_tft = TFTModel(
+  input_chunk_length = 30,
+  output_chunk_length = 15,
+  hidden_size = 16,
+  lstm_layers = 1,
+  num_attention_heads = 4,
+  dropout = 0.1,
+  hidden_continuous_size = 8,
+  batch_size = 32,
+  n_epochs = 500,
+  likelihood = None,
+  loss_fn = torch.nn.MSELoss(),
+  model_name = "TFTStoreX",
+  log_tensorboard = True,
+  save_checkpoints = True,
+  show_warnings = True,
+  optimizer_kwargs = {"lr": 0.002},
+  lr_scheduler_cls = torch.optim.lr_scheduler.ReduceLROnPlateau,
+  lr_scheduler_kwargs = {"patience": 5},
+  pl_trainer_kwargs = {
+    "callbacks": [early_stopper],
+    "accelerator": "gpu",
+    "devices": [0]
+    }
+)
+
+
+# All covariates, future & past
+tft_futcovars = [
+  "trend", 'tuesday', 'wednesday', 'thursday', 'friday', 'saturday', 'sunday', 
+  'day_sin', 'day_cos', "month_sin", "month_cos", 'oil', 'oil_ma28', 'onpromotion', 
+  'onp_ma28', 'local_holiday', 'regional_holiday', 'national_holiday', 'ny1', 
+  'ny2', 'ny_eve31', 'ny_eve30', 'xmas_before', 'xmas_after', 'quake_after', 
+  'dia_madre', 'futbol', 'black_friday', 'cyber_monday']
+
+tft_pastcovars = ["sales_ema7", "transactions", "trns_ma7"]
+
+
+# Fit TFT model
+model_tft.fit(
+  series = [y[:-45] for y in y_train_store],
+  future_covariates = [x[tft_futcovars] for x in x_store],
+  past_covariates = [x[tft_pastcovars] for x in x_store],
+  val_series = [y[-45:] for y in y_train_store],
+  val_future_covariates = [x[tft_futcovars] for x in x_store],
+  val_past_covariates = [x[tft_pastcovars] for x in x_store],
+  verbose = True
+)
+
+
+# # Load best checkpoint
+# model_tft = TFTModel.load_from_checkpoint("TFTStore2.0", best = True)
+# 
+# 
+# 
+# 
+# # First fit & validate the first store to initialize series
+# pred_tft_store = model_tft.predict(
+#   n=15,
+#   series = y_train_store[0],
+#   future_covariates = x_store[0][tft_futcovars],
+#   past_covariates = x_store[0][tft_pastcovars]
+#   )
+# 
+# # Then loop over all categories except first
+# for i in tqdm(range(1, len(y_train_store))):
+# 
+#   # Predict validation data
+#   pred = model_tft.predict(
+#     n=15,
+#     series = y_train_store[i],
+#     future_covariates = x_store[i][tft_futcovars],
+#     past_covariates = x_store[i][tft_pastcovars]
+#   )
+# 
+#   # Stack predictions to multivariate series
+#   pred_tft_store = pred_tft_store.stack(pred)
+#   
+# del pred, i
+# 
+# 
+# 
+# # Score TFT
+# scores_hierarchy(
+#   ts_sales[stores][-15:],
+#   trafo_zeroclip(pred_tft_store),
+#   stores,
+#   "TFT (global, all features)"
+#   )
+
+
+