diff --git a/mlforecast/_modidx.py b/mlforecast/_modidx.py index a802a118..d8627a8c 100644 --- a/mlforecast/_modidx.py +++ b/mlforecast/_modidx.py @@ -5,15 +5,35 @@ 'doc_host': 'https://Nixtla.github.io', 'git_url': 'https://github.com/Nixtla/mlforecast', 'lib_path': 'mlforecast'}, - 'syms': { 'mlforecast.auto': { 'mlforecast.auto.AutoMLForecast': ('auto.html#automlforecast', 'mlforecast/auto.py'), + 'syms': { 'mlforecast.auto': { 'mlforecast.auto.AutoCatboost': ('auto.html#autocatboost', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoCatboost.__init__': ('auto.html#autocatboost.__init__', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoElasticNet': ('auto.html#autoelasticnet', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoElasticNet.__init__': ('auto.html#autoelasticnet.__init__', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoLasso': ('auto.html#autolasso', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoLasso.__init__': ('auto.html#autolasso.__init__', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoLightGBM': ('auto.html#autolightgbm', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoLightGBM.__init__': ('auto.html#autolightgbm.__init__', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoLinearRegression': ('auto.html#autolinearregression', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoLinearRegression.__init__': ( 'auto.html#autolinearregression.__init__', + 'mlforecast/auto.py'), + 'mlforecast.auto.AutoMLForecast': ('auto.html#automlforecast', 'mlforecast/auto.py'), 'mlforecast.auto.AutoMLForecast.__init__': ('auto.html#automlforecast.__init__', 'mlforecast/auto.py'), 'mlforecast.auto.AutoMLForecast._seasonality_based_config': ( 'auto.html#automlforecast._seasonality_based_config', 'mlforecast/auto.py'), 'mlforecast.auto.AutoMLForecast.fit': ('auto.html#automlforecast.fit', 'mlforecast/auto.py'), 'mlforecast.auto.AutoMLForecast.predict': ('auto.html#automlforecast.predict', 'mlforecast/auto.py'), 'mlforecast.auto.AutoMLForecast.save': ('auto.html#automlforecast.save', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoModel': ('auto.html#automodel', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoModel.__init__': ('auto.html#automodel.__init__', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoModel.__repr__': ('auto.html#automodel.__repr__', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoRandomForest': ('auto.html#autorandomforest', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoRandomForest.__init__': ('auto.html#autorandomforest.__init__', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoRidge': ('auto.html#autoridge', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoRidge.__init__': ('auto.html#autoridge.__init__', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoXGBoost': ('auto.html#autoxgboost', 'mlforecast/auto.py'), + 'mlforecast.auto.AutoXGBoost.__init__': ('auto.html#autoxgboost.__init__', 'mlforecast/auto.py'), 'mlforecast.auto.catboost_space': ('auto.html#catboost_space', 'mlforecast/auto.py'), - 'mlforecast.auto.elasticnet_space': ('auto.html#elasticnet_space', 'mlforecast/auto.py'), + 'mlforecast.auto.elastic_net_space': ('auto.html#elastic_net_space', 'mlforecast/auto.py'), 'mlforecast.auto.lasso_space': ('auto.html#lasso_space', 'mlforecast/auto.py'), 'mlforecast.auto.lightgbm_space': ('auto.html#lightgbm_space', 'mlforecast/auto.py'), 'mlforecast.auto.linear_regression_space': ('auto.html#linear_regression_space', 'mlforecast/auto.py'), diff --git a/mlforecast/auto.py b/mlforecast/auto.py index 71a845f1..be91d4bd 100644 --- a/mlforecast/auto.py +++ b/mlforecast/auto.py @@ -2,31 +2,26 @@ # %% auto 0 __all__ = ['lightgbm_space', 'xgboost_space', 'catboost_space', 'linear_regression_space', 'ridge_space', 'lasso_space', - 'elasticnet_space', 'random_forest_space', 'AutoMLForecast'] + 'elastic_net_space', 'random_forest_space', 'AutoModel', 'AutoLightGBM', 'AutoXGBoost', 'AutoCatboost', + 'AutoLinearRegression', 'AutoRidge', 'AutoLasso', 'AutoElasticNet', 'AutoRandomForest', 'AutoMLForecast'] # %% ../nbs/auto.ipynb 2 -from collections import defaultdict -from typing import Any, Callable, Dict, List, Optional, Tuple, Union +from pathlib import Path +from typing import Any, Callable, Dict, List, Optional, Union import numpy as np import optuna -import pandas as pd from sklearn.base import BaseEstimator, clone -from sklearn.pipeline import Pipeline from sklearn.preprocessing import FunctionTransformer from utilsforecast.compat import DataFrame -from utilsforecast.losses import mase, smape +from utilsforecast.losses import smape from utilsforecast.processing import counts_by_id from utilsforecast.validation import validate_freq from . import MLForecast -from .core import Freq, Models, _get_model_name, _name_models -from mlforecast.lag_transforms import ( - ExpandingMean, - ExponentiallyWeightedMean, - RollingMean, -) -from .optimization import mlforecast_objective +from .core import Freq, _get_model_name, _name_models +from .lag_transforms import ExponentiallyWeightedMean, RollingMean +from .optimization import _TrialToConfig, mlforecast_objective from mlforecast.target_transforms import ( Differences, LocalStandardScaler, @@ -34,7 +29,7 @@ ) # %% ../nbs/auto.ipynb 3 -def lightgbm_space(trial): +def lightgbm_space(trial: optuna.Trial): return { "bagging_freq": 1, "learning_rate": 0.05, @@ -49,7 +44,7 @@ def lightgbm_space(trial): } -def xgboost_space(trial): +def xgboost_space(trial: optuna.Trial): return { "n_estimators": trial.suggest_int("n_estimators", 20, 1000), "max_depth": trial.suggest_int("max_depth", 1, 10), @@ -64,8 +59,9 @@ def xgboost_space(trial): } -def catboost_space(trial): +def catboost_space(trial: optuna.Trial): return { + "silent": True, "n_estimators": trial.suggest_int("n_estimators", 50, 1000), "depth": trial.suggest_int("depth", 1, 10), "learning_rate": trial.suggest_float("learning_rate", 1e-3, 0.2, log=True), @@ -75,25 +71,25 @@ def catboost_space(trial): } -def linear_regression_space(trial): +def linear_regression_space(trial: optuna.Trial): return {"fit_intercept": trial.suggest_categorical("fit_intercept", [True, False])} -def ridge_space(trial): +def ridge_space(trial: optuna.Trial): return { "fit_intercept": trial.suggest_categorical("fit_intercept", [True, False]), "alpha": trial.suggest_float("alpha", 0.001, 10.0), } -def lasso_space(trial): +def lasso_space(trial: optuna.Trial): return { "fit_intercept": trial.suggest_categorical("fit_intercept", [True, False]), "alpha": trial.suggest_float("alpha", 0.001, 10.0), } -def elasticnet_space(trial): +def elastic_net_space(trial: optuna.Trial): return { "fit_intercept": trial.suggest_categorical("fit_intercept", [True, False]), "alpha": trial.suggest_float("alpha", 0.001, 10.0), @@ -101,7 +97,7 @@ def elasticnet_space(trial): } -def random_forest_space(trial): +def random_forest_space(trial: optuna.Trial): return { "n_estimators": trial.suggest_int("n_estimators", 50, 1000), "max_depth": trial.suggest_int("max_depth", 1, 10), @@ -112,56 +108,171 @@ def random_forest_space(trial): ), } -# %% ../nbs/auto.ipynb 4 -_default_model_spaces = { - "LGBMRegressor": lightgbm_space, - "XGBRegressor": xgboost_space, - "CatBoostRegressor": catboost_space, - "LinearRegression": linear_regression_space, - "Ridge": ridge_space, - "Lasso": lasso_space, - "ElasticNet": elasticnet_space, - "RandomForest": random_forest_space, -} - -_ModelWithConfig = Tuple[BaseEstimator, Optional[Dict[str, Callable]]] - -# %% ../nbs/auto.ipynb 5 + +class AutoModel: + """Structure to hold a model and its search space + + Parameters + ---------- + model : BaseEstimator + scikit-learn compatible regressor + config : callable + function that takes an optuna trial and produces a configuration + """ + + def __init__( + self, + model: BaseEstimator, + config: _TrialToConfig, + ): + self.model = model + self.config = config + + def __repr__(self): + return f"AutoModel(model={_get_model_name(self.model)})" + + +class AutoLightGBM(AutoModel): + def __init__( + self, + config: Optional[_TrialToConfig] = None, + ): + from mlforecast.compat import LGBMRegressor + + super().__init__( + LGBMRegressor(), + config if config is not None else lightgbm_space, + ) + + +class AutoXGBoost(AutoModel): + def __init__( + self, + config: Optional[_TrialToConfig] = None, + ): + from mlforecast.compat import XGBRegressor + + super().__init__( + XGBRegressor(), + config if config is not None else xgboost_space, + ) + + +class AutoCatboost(AutoModel): + def __init__( + self, + config: Optional[_TrialToConfig] = None, + ): + from mlforecast.compat import CatBoostRegressor + + super().__init__( + CatBoostRegressor(), + config if config is not None else catboost_space, + ) + + +class AutoLinearRegression(AutoModel): + def __init__( + self, + config: Optional[_TrialToConfig] = None, + ): + from sklearn.linear_model import LinearRegression + + super().__init__( + LinearRegression(), + config if config is not None else linear_regression_space, + ) + + +class AutoRidge(AutoModel): + def __init__( + self, + config: Optional[_TrialToConfig] = None, + ): + from sklearn.linear_model import Ridge + + super().__init__( + Ridge(), + config if config is not None else ridge_space, + ) + + +class AutoLasso(AutoModel): + def __init__( + self, + config: Optional[_TrialToConfig] = None, + ): + from sklearn.linear_model import Lasso + + super().__init__( + Lasso(), + config if config is not None else lasso_space, + ) + + +class AutoElasticNet(AutoModel): + def __init__( + self, + config: Optional[_TrialToConfig] = None, + ): + from sklearn.linear_model import ElasticNet + + super().__init__( + ElasticNet(), + config if config is not None else elastic_net_space, + ) + + +class AutoRandomForest(AutoModel): + def __init__( + self, + config: Optional[_TrialToConfig] = None, + ): + from sklearn.ensemble import RandomForestRegressor + + super().__init__( + RandomForestRegressor(), + config if config is not None else random_forest_space, + ) + +# %% ../nbs/auto.ipynb 6 class AutoMLForecast: + """Hyperparameter optimization helper + + Parameters + ---------- + models : list or dict + Auto models to be optimized. + freq : str or int + pandas' or polars' offset alias or integer denoting the frequency of the series. + season_length : int + Length of the seasonal period. This is used for producing the feature space. + init_config : callable, optional (default=None) + Function that takes an optuna trial and produces a configuration passed to the MLForecast constructor. + fit_config : callable, optional (default=None) + Function that takes an optuna trial and produces a configuration passed to the MLForecast fit method. + num_threads : int (default=1) + Number of threads to use when computing the features. + """ def __init__( self, - models_with_configs: Union[List[_ModelWithConfig], Dict[str, _ModelWithConfig]], + models: Union[List[AutoModel], Dict[str, AutoModel]], freq: Freq, season_length: int, - init_config: Optional[Callable] = None, - fit_config: Optional[Callable] = None, + init_config: Optional[_TrialToConfig] = None, + fit_config: Optional[_TrialToConfig] = None, num_threads: int = 1, ): self.freq = freq self.season_length = season_length self.num_threads = num_threads - if isinstance(models_with_configs, list): - names = _name_models([_get_model_name(m) for m, _ in models_with_configs]) - self.models_with_configs = { - name: (model, config) - for name, (model, config) in zip(names, models_with_configs) - } - elif isinstance(models_with_configs, dict): - self.models_with_configs = models_with_configs + if isinstance(models, list): + model_names = _name_models([_get_model_name(m) for m in models]) + models_with_names = dict(zip(model_names, models)) else: - raise ValueError( - "`models_with_configs` should be a list of tuples " - "or a dict from str to tuple." - ) - for name, (model, config) in self.models_with_configs.items(): - if config is not None: - continue - if name not in _default_model_spaces: - raise NotImplementedError( - f"{name} does not have a default config. Please provide one." - ) - self.models_with_configs[name] = (model, _default_model_spaces[name]) + models_with_names = models + self.models = models_with_names self.init_config = init_config if fit_config is not None: self.fit_config = fit_config @@ -173,7 +284,7 @@ def _seasonality_based_config( h: int, min_samples: int, min_value: float, - ) -> Callable: + ) -> _TrialToConfig: # target transforms candidate_targ_tfms = [ None, @@ -205,9 +316,7 @@ def _seasonality_based_config( ) # lags - candidate_lags = [None] - if self.season_length > 1: - candidate_lags.append([self.season_length]) + candidate_lags = [None, [self.season_length]] seasonality2extra_candidate_lags = { 7: [ [7, 14], @@ -218,6 +327,9 @@ def _seasonality_based_config( range(1, 25), range(24, 24 * 7 + 1, 24), ], + 52: [ + range(4, 53, 4), + ], } candidate_lags.extend( seasonality2extra_candidate_lags.get(self.season_length, []) @@ -265,13 +377,9 @@ def _seasonality_based_config( 52: ["week", "year"], 60: ["weekday", "hour", "second"], } - date_features = seasonality2date_features.get(self.season_length, None) + candidate_date_features = seasonality2date_features.get(self.season_length, []) if isinstance(self.freq, int): - date_features = None - if date_features is not None: - use_date_features = trial.suggest_int("use_date_features", 0, 1) - if not use_date_features: - date_features = None + candidate_date_features = [] def config(trial): # target transforms @@ -293,6 +401,16 @@ def config(trial): else: lag_transforms = None + # date features + if candidate_date_features: + use_date_features = trial.suggest_int("use_date_features", 0, 1) + if use_date_features: + date_features = candidate_date_features + else: + date_features = None + else: + date_features = None + return { "lags": lags, "target_transforms": target_transforms, @@ -308,13 +426,45 @@ def fit( n_windows: int, h: int, num_samples: int, - loss: Optional[Callable] = None, + loss: Optional[Callable[[DataFrame, DataFrame], float]] = None, id_col: str = "unique_id", time_col: str = "ds", target_col: str = "y", study_kwargs: Optional[Dict[str, Any]] = None, optimize_kwargs: Optional[Dict[str, Any]] = None, ) -> "AutoMLForecast": + """Carry out the optimization process. + Each model is optimized independently and the best one is trained on all data + + Parameters + ---------- + df : pandas or polars DataFrame + Series data in long format. + n_windows : int + Number of windows to evaluate. + h : int + Forecast horizon. + num_samples : int + Number of trials to run + loss : callable, optional (default=None) + Function that takes the validation and train dataframes and produces a float. + If `None` will use the average SMAPE across series. + id_col : str (default='unique_id') + Column that identifies each serie. + time_col : str (default='ds') + Column that identifies each timestep, its values can be timestamps or integers. + target_col : str (default='y') + Column that contains the target. + study_kwargs : dict, optional (default=None) + Keyword arguments to be passed to the optuna.Study constructor. + optimize_kwargs : dict, optional (default=None) + Keyword arguments to be passed to the Study.optimize method. + + Returns + ------- + AutoMLForecast + object with best models and optimization results + """ validate_freq(df[time_col], self.freq) if self.init_config is not None: init_config = self.init_config @@ -337,18 +487,14 @@ def loss(df, train_df): if "sampler" not in study_kwargs: # for reproducibility study_kwargs["sampler"] = optuna.samplers.TPESampler(seed=0) - if optimize_kwargs is None: - optimize_kwargs = {} - if "n_jobs" not in optimize_kwargs: - optimize_kwargs["n_jobs"] = 1 self.results_ = [] self.models_ = {} - for name, (model, model_config) in self.models_with_configs.items(): + for name, auto_model in self.models.items(): def config_fn(trial: optuna.Trial) -> float: return { - "model_params": model_config(trial), + "model_params": auto_model.config(trial), "mlf_init_params": { **init_config(trial), "num_threads": self.num_threads, @@ -359,8 +505,8 @@ def config_fn(trial: optuna.Trial) -> float: objective = mlforecast_objective( df=df, config_fn=config_fn, - eval_fn=loss, - model=model, + loss=loss, + model=auto_model.model, freq=self.freq, n_windows=n_windows, h=h, @@ -372,7 +518,7 @@ def config_fn(trial: optuna.Trial) -> float: study.optimize(objective, n_trials=num_samples, **optimize_kwargs) self.results_.append(study) best_config = study.best_trial.user_attrs["config"] - best_model = clone(model) + best_model = clone(auto_model.model) best_model.set_params(**best_config["model_params"]) self.models_[name] = MLForecast( models={name: best_model}, @@ -387,6 +533,20 @@ def predict( h: int, X_df: Optional[DataFrame] = None, ) -> DataFrame: + """ "Compute forecasts + + Parameters + ---------- + h : int + Number of periods to predict. + X_df : pandas or polars DataFrame, optional (default=None) + Dataframe with the future exogenous features. Should have the id column and the time column. + + Returns + ------- + pandas or polars DataFrame + Predictions for each serie and timestep, with one column per model. + """ all_preds = None for name, model in self.models_.items(): preds = model.predict(h=h, X_df=X_df) @@ -396,6 +556,12 @@ def predict( all_preds[name] = preds[name] return all_preds - def save(self, path: str) -> None: + def save(self, path: Union[str, Path]) -> None: + """Save MLForecast objects + + Parameters + ---------- + path : str or pathlib.Path + Directory where artifacts will be stored.""" for name, model in self.models_.items(): model.save(f"{path}/{name}") diff --git a/mlforecast/compat.py b/mlforecast/compat.py index 2cfe2bfd..dd29519a 100644 --- a/mlforecast/compat.py +++ b/mlforecast/compat.py @@ -6,10 +6,34 @@ # %% ../nbs/compat.ipynb 1 try: from catboost import CatBoostRegressor +except ImportError: + + class CatBoostRegressor: + def __init__(self, *args, **kwargs): + raise ImportError("Please install catboost to use this model.") + + +try: + from lightgbm import LGBMRegressor +except ImportError: + + class LGBMRegressor: + def __init__(self, *args, **kwargs): + raise ImportError("Please install lightgbm to use this model.") + + +try: from window_ops.shift import shift_array except ImportError: def shift_array(*_args, **_kwargs): raise Exception - class CatBoostRegressor: ... + +try: + from xgboost import XGBRegressor +except ImportError: + + class XGBRegressor: + def __init__(self, *args, **kwargs): + raise ImportError("Please install xgboost to use this model.") diff --git a/mlforecast/optimization.py b/mlforecast/optimization.py index 4cbf50b9..3e4006bb 100644 --- a/mlforecast/optimization.py +++ b/mlforecast/optimization.py @@ -5,24 +5,26 @@ # %% ../nbs/optimization.ipynb 2 import copy -from typing import Callable, List, Optional +from typing import Any, Callable, Dict, Optional import numpy as np import optuna import utilsforecast.processing as ufp from sklearn.base import BaseEstimator, clone from utilsforecast.compat import DataFrame -from utilsforecast.losses import smape from . import MLForecast from .compat import CatBoostRegressor from .core import Freq # %% ../nbs/optimization.ipynb 3 +_TrialToConfig = Callable[[optuna.Trial], Dict[str, Any]] + +# %% ../nbs/optimization.ipynb 4 def mlforecast_objective( df: DataFrame, - config_fn: Callable, - eval_fn: Callable, + config_fn: _TrialToConfig, + loss: Callable[[DataFrame, DataFrame], float], model: BaseEstimator, freq: Freq, n_windows: int, @@ -30,7 +32,35 @@ def mlforecast_objective( id_col: str = "unique_id", time_col: str = "ds", target_col: str = "y", -) -> Callable: +) -> _TrialToConfig: + """optuna objective function for the MLForecast class + + Parameters + ---------- + config_fn : callable + Function that takes an optuna trial and produces a configuration with the following keys: + - model_params + - mlf_init_params + - mlf_fit_params + loss : callable + Function that takes the validation and train dataframes and produces a float. + model : BaseEstimator + scikit-learn compatible model to be trained + freq : str or int + pandas' or polars' offset alias or integer denoting the frequency of the series. + n_windows : int + Number of windows to evaluate. + h : int + Forecast horizon. + id_col : str (default='unique_id') + Column that identifies each serie. + time_col : str (default='ds') + Column that identifies each timestep, its values can be timestamps or integers. + target_col : str (default='y') + Column that contains the target. + study_kwargs : dict, optional (default=None) + """ + def objective(trial: optuna.Trial) -> float: config = config_fn(trial) trial.set_user_attr("config", copy.deepcopy(config)) @@ -95,7 +125,7 @@ def objective(trial: optuna.Trial) -> float: "Please verify that the passed frequency (freq) matches your series' " "and that there aren't any missing periods." ) - metric = eval_fn(result, train_df=train) + metric = loss(result, train_df=train) metrics.append(metric) trial.report(metric, step=i) if trial.should_prune(): diff --git a/nbs/auto.ipynb b/nbs/auto.ipynb index 8d7080e0..cc851a36 100644 --- a/nbs/auto.ipynb +++ b/nbs/auto.ipynb @@ -30,24 +30,22 @@ "outputs": [], "source": [ "#| export\n", - "from collections import defaultdict\n", - "from typing import Any, Callable, Dict, List, Optional, Tuple, Union\n", + "from pathlib import Path\n", + "from typing import Any, Callable, Dict, List, Optional, Union\n", "\n", "import numpy as np\n", "import optuna\n", - "import pandas as pd\n", "from sklearn.base import BaseEstimator, clone\n", - "from sklearn.pipeline import Pipeline\n", "from sklearn.preprocessing import FunctionTransformer\n", "from utilsforecast.compat import DataFrame\n", - "from utilsforecast.losses import mase, smape\n", + "from utilsforecast.losses import smape\n", "from utilsforecast.processing import counts_by_id\n", "from utilsforecast.validation import validate_freq\n", "\n", "from mlforecast import MLForecast\n", - "from mlforecast.core import Freq, Models, _get_model_name, _name_models\n", - "from mlforecast.lag_transforms import ExpandingMean, ExponentiallyWeightedMean, RollingMean\n", - "from mlforecast.optimization import mlforecast_objective\n", + "from mlforecast.core import Freq, _get_model_name, _name_models\n", + "from mlforecast.lag_transforms import ExponentiallyWeightedMean, RollingMean\n", + "from mlforecast.optimization import _TrialToConfig, mlforecast_objective\n", "from mlforecast.target_transforms import Differences, LocalStandardScaler, GlobalSklearnTransformer" ] }, @@ -59,37 +57,38 @@ "outputs": [], "source": [ "#| export\n", - "def lightgbm_space(trial):\n", + "def lightgbm_space(trial: optuna.Trial):\n", " return {\n", - " \"bagging_freq\": 1,\n", - " \"learning_rate\": 0.05,\n", - " \"verbosity\": -1, \n", - " \"n_estimators\": trial.suggest_int(\"n_estimators\", 20, 1000, log=True),\n", + " 'bagging_freq': 1,\n", + " 'learning_rate': 0.05,\n", + " 'verbosity': -1, \n", + " 'n_estimators': trial.suggest_int('n_estimators', 20, 1000, log=True),\n", " 'lambda_l1': trial.suggest_float('lambda_l1', 1e-8, 10.0, log=True),\n", " 'lambda_l2': trial.suggest_float('lambda_l2', 1e-8, 10.0, log=True),\n", " 'num_leaves': trial.suggest_int('num_leaves', 2, 4096, log=True),\n", " 'feature_fraction': trial.suggest_float('feature_fraction', 0.5, 1.0),\n", " 'bagging_fraction': trial.suggest_float('bagging_fraction', 0.5, 1.0),\n", - " \"objective\": trial.suggest_categorical(\"objective\", ['l1', 'l2']),\n", + " 'objective': trial.suggest_categorical('objective', ['l1', 'l2']),\n", " }\n", "\n", - "def xgboost_space(trial):\n", + "def xgboost_space(trial: optuna.Trial):\n", " return {\n", - " \"n_estimators\": trial.suggest_int(\"n_estimators\", 20, 1000),\n", + " 'n_estimators': trial.suggest_int('n_estimators', 20, 1000),\n", " 'max_depth': trial.suggest_int('max_depth', 1, 10),\n", " 'learning_rate': trial.suggest_float('learning_rate', 1e-3, 0.2, log=True),\n", " 'subsample': trial.suggest_float('subsample', 0.1, 1.0),\n", " 'bagging_freq': trial.suggest_float('bagging_freq', 0.1, 1.0),\n", " 'colsample_bytree': trial.suggest_float('colsample_bytree', 0.1, 1.0),\n", " 'min_data_in_leaf': trial.suggest_float('min_data_in_leaf', 1, 100),\n", - " \"reg_lambda\": trial.suggest_float(\"reg_lambda\", 1e-8, 1.0, log=True),\n", - " \"reg_alpha\": trial.suggest_float(\"reg_alpha\", 1e-8, 1.0, log=True),\n", + " 'reg_lambda': trial.suggest_float('reg_lambda', 1e-8, 1.0, log=True),\n", + " 'reg_alpha': trial.suggest_float('reg_alpha', 1e-8, 1.0, log=True),\n", " 'min_child_weight': trial.suggest_int('min_child_weight', 2, 10),\n", " }\n", " \n", - "def catboost_space(trial):\n", + "def catboost_space(trial: optuna.Trial):\n", " return {\n", - " \"n_estimators\": trial.suggest_int(\"n_estimators\", 50, 1000),\n", + " 'silent': True,\n", + " 'n_estimators': trial.suggest_int('n_estimators', 50, 1000),\n", " 'depth': trial.suggest_int('depth', 1, 10),\n", " 'learning_rate': trial.suggest_float('learning_rate', 1e-3, 0.2, log=True),\n", " 'subsample': trial.suggest_float('subsample', 0.1, 1.0),\n", @@ -97,60 +96,210 @@ " 'min_data_in_leaf': trial.suggest_float('min_data_in_leaf', 1, 100),\n", " }\n", " \n", - "def linear_regression_space(trial):\n", + "def linear_regression_space(trial: optuna.Trial):\n", " return {\n", - " \"fit_intercept\": trial.suggest_categorical(\"fit_intercept\", [True, False])\n", + " 'fit_intercept': trial.suggest_categorical('fit_intercept', [True, False])\n", " }\n", " \n", - "def ridge_space(trial):\n", + "def ridge_space(trial: optuna.Trial):\n", " return {\n", - " \"fit_intercept\": trial.suggest_categorical(\"fit_intercept\", [True, False]),\n", + " 'fit_intercept': trial.suggest_categorical('fit_intercept', [True, False]),\n", " 'alpha': trial.suggest_float('alpha', 0.001, 10.0)\n", " }\n", " \n", - "def lasso_space(trial):\n", + "def lasso_space(trial: optuna.Trial):\n", " return {\n", - " \"fit_intercept\": trial.suggest_categorical(\"fit_intercept\", [True, False]),\n", + " 'fit_intercept': trial.suggest_categorical('fit_intercept', [True, False]),\n", " 'alpha': trial.suggest_float('alpha', 0.001, 10.0)\n", " }\n", " \n", - "def elasticnet_space(trial):\n", + "def elastic_net_space(trial: optuna.Trial):\n", " return {\n", - " \"fit_intercept\": trial.suggest_categorical(\"fit_intercept\", [True, False]),\n", + " 'fit_intercept': trial.suggest_categorical('fit_intercept', [True, False]),\n", " 'alpha': trial.suggest_float('alpha', 0.001, 10.0),\n", " 'l1_ratio': trial.suggest_float('l1_ratio', 0.0, 1.0)\n", " }\n", "\n", - "def random_forest_space(trial):\n", + "def random_forest_space(trial: optuna.Trial):\n", " return {\n", - " \"n_estimators\": trial.suggest_int(\"n_estimators\", 50, 1000),\n", + " 'n_estimators': trial.suggest_int('n_estimators', 50, 1000),\n", " 'max_depth': trial.suggest_int('max_depth', 1, 10),\n", " 'min_samples_split': trial.suggest_int('min_child_samples', 1, 100),\n", " 'max_features': trial.suggest_float('max_features', 0.5, 1.0),\n", - " \"criterion\": trial.suggest_categorical(\"criterion\", ['squared_error', 'poisson']),\n", - " }" + " 'criterion': trial.suggest_categorical('criterion', ['squared_error', 'poisson']),\n", + " }\n", + "\n", + "class AutoModel:\n", + " \"\"\"Structure to hold a model and its search space\n", + " \n", + " Parameters\n", + " ----------\n", + " model : BaseEstimator\n", + " scikit-learn compatible regressor\n", + " config : callable \n", + " function that takes an optuna trial and produces a configuration\n", + " \"\"\"\n", + " def __init__(\n", + " self,\n", + " model: BaseEstimator,\n", + " config: _TrialToConfig,\n", + " ):\n", + " self.model = model\n", + " self.config = config\n", + "\n", + " def __repr__(self):\n", + " return f'AutoModel(model={_get_model_name(self.model)})'\n", + "\n", + "class AutoLightGBM(AutoModel):\n", + " def __init__(\n", + " self,\n", + " config: Optional[_TrialToConfig] = None,\n", + " ):\n", + " from mlforecast.compat import LGBMRegressor\n", + " super().__init__(\n", + " LGBMRegressor(),\n", + " config if config is not None else lightgbm_space,\n", + " )\n", + "\n", + "class AutoXGBoost(AutoModel):\n", + " def __init__(\n", + " self,\n", + " config: Optional[_TrialToConfig] = None,\n", + " ):\n", + " from mlforecast.compat import XGBRegressor\n", + " super().__init__(\n", + " XGBRegressor(),\n", + " config if config is not None else xgboost_space,\n", + " )\n", + "\n", + "class AutoCatboost(AutoModel):\n", + " def __init__(\n", + " self,\n", + " config: Optional[_TrialToConfig] = None,\n", + " ):\n", + " from mlforecast.compat import CatBoostRegressor\n", + " super().__init__(\n", + " CatBoostRegressor(),\n", + " config if config is not None else catboost_space,\n", + " )\n", + "\n", + "class AutoLinearRegression(AutoModel):\n", + " def __init__(\n", + " self,\n", + " config: Optional[_TrialToConfig] = None,\n", + " ):\n", + " from sklearn.linear_model import LinearRegression\n", + " super().__init__(\n", + " LinearRegression(),\n", + " config if config is not None else linear_regression_space,\n", + " )\n", + "\n", + "class AutoRidge(AutoModel):\n", + " def __init__(\n", + " self,\n", + " config: Optional[_TrialToConfig] = None,\n", + " ):\n", + " from sklearn.linear_model import Ridge\n", + " super().__init__(\n", + " Ridge(),\n", + " config if config is not None else ridge_space,\n", + " )\n", + "\n", + "class AutoLasso(AutoModel):\n", + " def __init__(\n", + " self,\n", + " config: Optional[_TrialToConfig] = None,\n", + " ):\n", + " from sklearn.linear_model import Lasso\n", + " super().__init__(\n", + " Lasso(),\n", + " config if config is not None else lasso_space,\n", + " )\n", + "\n", + "class AutoElasticNet(AutoModel):\n", + " def __init__(\n", + " self,\n", + " config: Optional[_TrialToConfig] = None,\n", + " ):\n", + " from sklearn.linear_model import ElasticNet\n", + " super().__init__(\n", + " ElasticNet(),\n", + " config if config is not None else elastic_net_space,\n", + " )\n", + "\n", + "class AutoRandomForest(AutoModel):\n", + " def __init__(\n", + " self,\n", + " config: Optional[_TrialToConfig] = None,\n", + " ):\n", + " from sklearn.ensemble import RandomForestRegressor\n", + " super().__init__(\n", + " RandomForestRegressor(),\n", + " config if config is not None else random_forest_space,\n", + " )" ] }, { "cell_type": "code", "execution_count": null, - "id": "c01bfb2f-43ab-4848-add3-3e4c01cf3c49", + "id": "b2819ca2", "metadata": {}, "outputs": [], "source": [ - "#| exporti\n", - "_default_model_spaces = {\n", - " 'LGBMRegressor': lightgbm_space,\n", - " 'XGBRegressor': xgboost_space,\n", - " 'CatBoostRegressor': catboost_space,\n", - " 'LinearRegression': linear_regression_space,\n", - " 'Ridge': ridge_space,\n", - " 'Lasso': lasso_space,\n", - " 'ElasticNet': elasticnet_space,\n", - " 'RandomForest': random_forest_space,\n", - "}\n", - "\n", - "_ModelWithConfig = Tuple[BaseEstimator, Optional[Dict[str, Callable]]]" + "#| hide\n", + "from nbdev import show_doc" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "ea9f3693", + "metadata": {}, + "outputs": [ + { + "data": { + "text/markdown": [ + "---\n", + "\n", + "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/auto.py#L115){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n", + "\n", + "### AutoModel\n", + "\n", + "> AutoModel (model:sklearn.base.BaseEstimator,\n", + "> config:Callable[[optuna.trial._trial.Trial],Dict[str,Any]])\n", + "\n", + "Structure to hold a model and its search space\n", + "\n", + "| | **Type** | **Details** |\n", + "| -- | -------- | ----------- |\n", + "| model | BaseEstimator | scikit-learn compatible regressor |\n", + "| config | Callable | function that takes an optuna trial and produces a configuration |" + ], + "text/plain": [ + "---\n", + "\n", + "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/auto.py#L115){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n", + "\n", + "### AutoModel\n", + "\n", + "> AutoModel (model:sklearn.base.BaseEstimator,\n", + "> config:Callable[[optuna.trial._trial.Trial],Dict[str,Any]])\n", + "\n", + "Structure to hold a model and its search space\n", + "\n", + "| | **Type** | **Details** |\n", + "| -- | -------- | ----------- |\n", + "| model | BaseEstimator | scikit-learn compatible regressor |\n", + "| config | Callable | function that takes an optuna trial and produces a configuration |" + ] + }, + "execution_count": null, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "show_doc(AutoModel)" ] }, { @@ -162,40 +311,41 @@ "source": [ "#| export\n", "class AutoMLForecast:\n", - "\n", + " \"\"\"Hyperparameter optimization helper\n", + " \n", + " Parameters\n", + " ----------\n", + " models : list or dict\n", + " Auto models to be optimized.\n", + " freq : str or int\n", + " pandas' or polars' offset alias or integer denoting the frequency of the series.\n", + " season_length : int\n", + " Length of the seasonal period. This is used for producing the feature space.\n", + " init_config : callable, optional (default=None)\n", + " Function that takes an optuna trial and produces a configuration passed to the MLForecast constructor.\n", + " fit_config : callable, optional (default=None)\n", + " Function that takes an optuna trial and produces a configuration passed to the MLForecast fit method.\n", + " num_threads : int (default=1)\n", + " Number of threads to use when computing the features.\n", + " \"\"\"\n", " def __init__(\n", " self,\n", - " models_with_configs: Union[List[_ModelWithConfig], Dict[str, _ModelWithConfig]],\n", + " models: Union[List[AutoModel], Dict[str, AutoModel]],\n", " freq: Freq,\n", " season_length: int,\n", - " init_config: Optional[Callable] = None,\n", - " fit_config: Optional[Callable] = None,\n", + " init_config: Optional[_TrialToConfig] = None,\n", + " fit_config: Optional[_TrialToConfig] = None,\n", " num_threads: int = 1,\n", " ):\n", " self.freq = freq\n", " self.season_length = season_length\n", " self.num_threads = num_threads\n", - " if isinstance(models_with_configs, list):\n", - " names = _name_models([_get_model_name(m) for m, _ in models_with_configs])\n", - " self.models_with_configs = {\n", - " name: (model, config)\n", - " for name, (model, config) in zip(names, models_with_configs)\n", - " }\n", - " elif isinstance(models_with_configs, dict):\n", - " self.models_with_configs = models_with_configs\n", + " if isinstance(models, list):\n", + " model_names = _name_models([_get_model_name(m) for m in models])\n", + " models_with_names = dict(zip(model_names, models))\n", " else:\n", - " raise ValueError(\n", - " '`models_with_configs` should be a list of tuples '\n", - " 'or a dict from str to tuple.'\n", - " )\n", - " for name, (model, config) in self.models_with_configs.items():\n", - " if config is not None:\n", - " continue\n", - " if name not in _default_model_spaces:\n", - " raise NotImplementedError(\n", - " f\"{name} does not have a default config. Please provide one.\"\n", - " )\n", - " self.models_with_configs[name] = (model, _default_model_spaces[name])\n", + " models_with_names = models\n", + " self.models = models_with_names\n", " self.init_config = init_config\n", " if fit_config is not None:\n", " self.fit_config = fit_config\n", @@ -207,7 +357,7 @@ " h: int,\n", " min_samples: int,\n", " min_value: float,\n", - " ) -> Callable:\n", + " ) -> _TrialToConfig:\n", " # target transforms \n", " candidate_targ_tfms = [\n", " None,\n", @@ -233,9 +383,7 @@ " )\n", "\n", " # lags\n", - " candidate_lags = [None]\n", - " if self.season_length > 1:\n", - " candidate_lags.append([self.season_length])\n", + " candidate_lags = [None, [self.season_length]]\n", " seasonality2extra_candidate_lags = {\n", " 7: [\n", " [7, 14],\n", @@ -246,6 +394,9 @@ " range(1, 25),\n", " range(24, 24 * 7 + 1, 24),\n", " ],\n", + " 52: [\n", + " range(4, 53, 4),\n", + " ]\n", " }\n", " candidate_lags.extend(\n", " seasonality2extra_candidate_lags.get(self.season_length, [])\n", @@ -293,13 +444,9 @@ " 52: ['week', 'year'],\n", " 60: ['weekday', 'hour', 'second'],\n", " }\n", - " date_features = seasonality2date_features.get(self.season_length, None)\n", + " candidate_date_features = seasonality2date_features.get(self.season_length, [])\n", " if isinstance(self.freq, int):\n", - " date_features = None\n", - " if date_features is not None:\n", - " use_date_features = trial.suggest_int('use_date_features', 0, 1)\n", - " if not use_date_features:\n", - " date_features = None \n", + " candidate_date_features = []\n", "\n", " def config(trial):\n", " # target transforms\n", @@ -320,6 +467,16 @@ " lag_transforms = candidate_lag_tfms[lag_tfms_idx]\n", " else:\n", " lag_transforms = None\n", + "\n", + " # date features\n", + " if candidate_date_features:\n", + " use_date_features = trial.suggest_int('use_date_features', 0, 1)\n", + " if use_date_features:\n", + " date_features = candidate_date_features\n", + " else:\n", + " date_features = None \n", + " else:\n", + " date_features = None\n", " \n", " return {\n", " 'lags': lags,\n", @@ -336,13 +493,45 @@ " n_windows: int,\n", " h: int,\n", " num_samples: int, \n", - " loss: Optional[Callable] = None,\n", + " loss: Optional[Callable[[DataFrame, DataFrame], float]] = None,\n", " id_col: str = 'unique_id',\n", " time_col: str = 'ds',\n", " target_col: str = 'y',\n", " study_kwargs: Optional[Dict[str, Any]] = None,\n", " optimize_kwargs: Optional[Dict[str, Any]] = None,\n", " ) -> 'AutoMLForecast':\n", + " \"\"\"Carry out the optimization process.\n", + " Each model is optimized independently and the best one is trained on all data\n", + " \n", + " Parameters\n", + " ----------\n", + " df : pandas or polars DataFrame\n", + " Series data in long format.\n", + " n_windows : int\n", + " Number of windows to evaluate.\n", + " h : int\n", + " Forecast horizon.\n", + " num_samples : int\n", + " Number of trials to run\n", + " loss : callable, optional (default=None)\n", + " Function that takes the validation and train dataframes and produces a float.\n", + " If `None` will use the average SMAPE across series.\n", + " id_col : str (default='unique_id')\n", + " Column that identifies each serie.\n", + " time_col : str (default='ds')\n", + " Column that identifies each timestep, its values can be timestamps or integers.\n", + " target_col : str (default='y')\n", + " Column that contains the target. \n", + " study_kwargs : dict, optional (default=None)\n", + " Keyword arguments to be passed to the optuna.Study constructor.\n", + " optimize_kwargs : dict, optional (default=None)\n", + " Keyword arguments to be passed to the Study.optimize method.\n", + "\n", + " Returns\n", + " -------\n", + " AutoMLForecast\n", + " object with best models and optimization results\n", + " \"\"\"\n", " validate_freq(df[time_col], self.freq)\n", " if self.init_config is not None:\n", " init_config = self.init_config\n", @@ -363,17 +552,13 @@ " if 'sampler' not in study_kwargs:\n", " # for reproducibility\n", " study_kwargs['sampler'] = optuna.samplers.TPESampler(seed=0)\n", - " if optimize_kwargs is None:\n", - " optimize_kwargs = {}\n", - " if 'n_jobs' not in optimize_kwargs:\n", - " optimize_kwargs['n_jobs'] = 1\n", "\n", " self.results_ = []\n", " self.models_ = {}\n", - " for name, (model, model_config) in self.models_with_configs.items():\n", + " for name, auto_model in self.models.items():\n", " def config_fn(trial: optuna.Trial) -> float:\n", " return {\n", - " 'model_params': model_config(trial),\n", + " 'model_params': auto_model.config(trial),\n", " 'mlf_init_params': {\n", " **init_config(trial),\n", " 'num_threads': self.num_threads,\n", @@ -384,8 +569,8 @@ " objective = mlforecast_objective(\n", " df=df,\n", " config_fn=config_fn,\n", - " eval_fn=loss,\n", - " model=model,\n", + " loss=loss,\n", + " model=auto_model.model,\n", " freq=self.freq,\n", " n_windows=n_windows,\n", " h=h,\n", @@ -397,7 +582,7 @@ " study.optimize(objective, n_trials=num_samples, **optimize_kwargs)\n", " self.results_.append(study)\n", " best_config = study.best_trial.user_attrs['config'] \n", - " best_model = clone(model)\n", + " best_model = clone(auto_model.model)\n", " best_model.set_params(**best_config['model_params'])\n", " self.models_[name] = MLForecast(\n", " models={name: best_model},\n", @@ -415,6 +600,20 @@ " h: int,\n", " X_df: Optional[DataFrame] = None,\n", " ) -> DataFrame:\n", + " \"\"\"\"Compute forecasts\n", + "\n", + " Parameters\n", + " ----------\n", + " h : int\n", + " Number of periods to predict.\n", + " X_df : pandas or polars DataFrame, optional (default=None)\n", + " Dataframe with the future exogenous features. Should have the id column and the time column.\n", + "\n", + " Returns\n", + " -------\n", + " pandas or polars DataFrame\n", + " Predictions for each serie and timestep, with one column per model.\n", + " \"\"\"\n", " all_preds = None\n", " for name, model in self.models_.items():\n", " preds = model.predict(h=h, X_df=X_df)\n", @@ -424,11 +623,87 @@ " all_preds[name] = preds[name]\n", " return all_preds\n", "\n", - " def save(self, path: str) -> None:\n", + " def save(self, path: Union[str, Path]) -> None:\n", + " \"\"\"Save MLForecast objects\n", + "\n", + " Parameters\n", + " ----------\n", + " path : str or pathlib.Path\n", + " Directory where artifacts will be stored.\"\"\"\n", " for name, model in self.models_.items():\n", " model.save(f'{path}/{name}')" ] }, + { + "cell_type": "code", + "execution_count": null, + "id": "2e311237", + "metadata": {}, + "outputs": [ + { + "data": { + "text/markdown": [ + "---\n", + "\n", + "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/auto.py#L242){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n", + "\n", + "### AutoMLForecast\n", + "\n", + "> AutoMLForecast\n", + "> (models:Union[List[__main__.AutoModel],Dict[str,__main__.\n", + "> AutoModel]], freq:Union[int,str], season_length:int, init\n", + "> _config:Optional[Callable[[optuna.trial._trial.Trial],Dic\n", + "> t[str,Any]]]=None, fit_config:Optional[Callable[[optuna.t\n", + "> rial._trial.Trial],Dict[str,Any]]]=None,\n", + "> num_threads:int=1)\n", + "\n", + "Hyperparameter optimization helper\n", + "\n", + "| | **Type** | **Default** | **Details** |\n", + "| -- | -------- | ----------- | ----------- |\n", + "| models | Union | | Auto models to be optimized. |\n", + "| freq | Union | | pandas' or polars' offset alias or integer denoting the frequency of the series. |\n", + "| season_length | int | | Length of the seasonal period. This is used for producing the feature space. |\n", + "| init_config | Optional | None | Function that takes an optuna trial and produces a configuration passed to the MLForecast constructor. |\n", + "| fit_config | Optional | None | Function that takes an optuna trial and produces a configuration passed to the MLForecast fit method. |\n", + "| num_threads | int | 1 | Number of threads to use when computing the features. |" + ], + "text/plain": [ + "---\n", + "\n", + "[source](https://github.com/Nixtla/mlforecast/blob/main/mlforecast/auto.py#L242){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n", + "\n", + "### AutoMLForecast\n", + "\n", + "> AutoMLForecast\n", + "> (models:Union[List[__main__.AutoModel],Dict[str,__main__.\n", + "> AutoModel]], freq:Union[int,str], season_length:int, init\n", + "> _config:Optional[Callable[[optuna.trial._trial.Trial],Dic\n", + "> t[str,Any]]]=None, fit_config:Optional[Callable[[optuna.t\n", + "> rial._trial.Trial],Dict[str,Any]]]=None,\n", + "> num_threads:int=1)\n", + "\n", + "Hyperparameter optimization helper\n", + "\n", + "| | **Type** | **Default** | **Details** |\n", + "| -- | -------- | ----------- | ----------- |\n", + "| models | Union | | Auto models to be optimized. |\n", + "| freq | Union | | pandas' or polars' offset alias or integer denoting the frequency of the series. |\n", + "| season_length | int | | Length of the seasonal period. This is used for producing the feature space. |\n", + "| init_config | Optional | None | Function that takes an optuna trial and produces a configuration passed to the MLForecast constructor. |\n", + "| fit_config | Optional | None | Function that takes an optuna trial and produces a configuration passed to the MLForecast fit method. |\n", + "| num_threads | int | 1 | Number of threads to use when computing the features. |" + ] + }, + "execution_count": null, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "show_doc(AutoMLForecast)" + ] + }, { "cell_type": "code", "execution_count": null, @@ -436,9 +711,6 @@ "metadata": {}, "outputs": [], "source": [ - "import warnings\n", - "\n", - "import lightgbm as lgb\n", "from datasetsforecast.m4 import M4, M4Evaluation, M4Info\n", "from sklearn.linear_model import Ridge\n", "from sklearn.compose import ColumnTransformer\n", @@ -465,73 +737,185 @@ { "cell_type": "code", "execution_count": null, - "id": "f131da88-6243-4b26-92d0-c0a1f559234b", + "id": "7888cb6d", "metadata": {}, "outputs": [], "source": [ - "import pickle" + "ridge_pipeline = make_pipeline(\n", + " ColumnTransformer(\n", + " [('encoder', OneHotEncoder(), ['unique_id'])],\n", + " remainder='passthrough',\n", + " ),\n", + " Ridge()\n", + ")\n", + "auto_ridge = AutoModel(ridge_pipeline, lambda trial: {f'ridge__{k}': v for k, v in ridge_space(trial).items()})" ] }, { "cell_type": "code", "execution_count": null, - "id": "49ccb9b1-484a-4938-a762-1a843a716983", + "id": "88b2aa47", "metadata": {}, - "outputs": [], + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + " | unique_id | \n", + "ds | \n", + "lgb | \n", + "ridge | \n", + "
---|---|---|---|---|
0 | \n", + "W1 | \n", + "2180 | \n", + "35529.435224 | \n", + "36111.103819 | \n", + "
1 | \n", + "W1 | \n", + "2181 | \n", + "35521.764894 | \n", + "36195.480607 | \n", + "
2 | \n", + "W1 | \n", + "2182 | \n", + "35537.417268 | \n", + "36107.924714 | \n", + "
3 | \n", + "W1 | \n", + "2183 | \n", + "35538.058206 | \n", + "36027.610599 | \n", + "
4 | \n", + "W1 | \n", + "2184 | \n", + "35614.611211 | \n", + "36093.419431 | \n", + "
... | \n", + "... | \n", + "... | \n", + "... | \n", + "... | \n", + "
4662 | \n", + "W99 | \n", + "2292 | \n", + "15071.536978 | \n", + "15319.461233 | \n", + "
4663 | \n", + "W99 | \n", + "2293 | \n", + "15058.145278 | \n", + "15299.892278 | \n", + "
4664 | \n", + "W99 | \n", + "2294 | \n", + "15042.493434 | \n", + "15272.111331 | \n", + "
4665 | \n", + "W99 | \n", + "2295 | \n", + "15042.144846 | \n", + "15250.457610 | \n", + "
4666 | \n", + "W99 | \n", + "2296 | \n", + "15038.729044 | \n", + "15232.536072 | \n", + "
4667 rows × 4 columns
\n", + "