feature(Config): added transformation parameters to Config, removed expanding_window (it's ON now)

config/preprocess.py

@@ -9,6 +9,7 @@ from labeling.eventfilters_map import eventfilters_map
 from labeling.labellers_map import labellers_map
 from models.sklearn import SKLearnModel
 from sklearn.ensemble import VotingClassifier
+from transformations.retrieve import get_pca, get_rfe, get_scaler
 
 
 def preprocess_config(raw_config: RawConfig) -> Config:
@@ -18,6 +19,7 @@ def preprocess_config(raw_config: RawConfig) -> Config:
     config_dict = __preprocess_data_collections_config(config_dict)
     config_dict = __preprocess_event_filter_config(config_dict)
     config_dict = __preprocess_event_labeller_config(config_dict)
+    config_dict = __preprocess_transformations_config(config_dict)
 
     config_dict["no_of_classes"] = "two"
     config_dict["mode"] = "training"
@@ -89,3 +91,18 @@ def __preprocess_event_labeller_config(config_dict: dict) -> dict:
         config_dict["forecasting_horizon"]
     )
     return config_dict
+
+
+def __preprocess_transformations_config(config_dict: dict) -> dict:
+    transformations = [
+        get_scaler(config_dict["scaler"]),
+        get_pca(config_dict["dimensionality_reduction_ratio"], config_dict["sliding_window_size"]),
+        get_rfe(config_dict["n_features_to_select"]),
+    ]
+    transformations = [x for x in transformations if x is not None]
+    config_dict["transformations"] = transformations
+    config_dict.pop("scaler")
+    config_dict.pop("dimensionality_reduction_ratio")
+    config_dict.pop("n_features_to_select")
+
+    return config_dict

config/presets.py

@@ -1,36 +1,6 @@
 from .types import RawConfig, Config
 
 
-def get_dev_config() -> RawConfig:
-
-    classification_models = ["LogisticRegression_two_class"]
-
-    return RawConfig(
-        directional_models_meta=False,
-        dimensionality_reduction=False,
-        n_features_to_select=30,
-        expanding_window_base=False,
-        expanding_window_meta=False,
-        sliding_window_size_base=380,
-        sliding_window_size_meta=1,
-        retrain_every=20,
-        scaler="minmax",  # 'normalize' 'minmax' 'standardize'
-        assets=["daily_only_btc"],
-        target_asset="BTC_USD",
-        other_assets=[],
-        exogenous_data=[],
-        load_non_target_asset=True,
-        own_features=["level_2", "date_days"],
-        other_features=["single_mom"],
-        exogenous_features=["z_score"],
-        directional_models=classification_models,
-        meta_models=[],
-        event_filter="none",
-        labeling="two_class",
-        forecasting_horizon=100,
-    )
-
-
 def get_default_ensemble_config() -> RawConfig:
 
     classification_models = [
@@ -45,13 +15,9 @@ def get_default_ensemble_config() -> RawConfig:
     meta_models = ["LogisticRegression_two_class", "LGBM"]
 
     return RawConfig(
-        directional_models_meta=True,
+        dimensionality_reduction_ratio=0.5,
-        dimensionality_reduction=False,
         n_features_to_select=30,
-        expanding_window_base=False,
+        sliding_window_size=380,
-        expanding_window_meta=True,
-        sliding_window_size_base=380,
-        sliding_window_size_meta=240,
         retrain_every=10,
         scaler="minmax",  # 'normalize' 'minmax' 'standardize'
         assets=["daily_crypto"],
@@ -72,17 +38,20 @@ def get_default_ensemble_config() -> RawConfig:
 
 def get_lightweight_ensemble_config() -> RawConfig:
 
-    classification_models = ["LogisticRegression_two_class", "LGBM"]
+    classification_models = [
+        "LogisticRegression_two_class",
+        "LDA",
+        "NB",
+        "RFC",
+        "LGBM",
+        # "StaticMom",
+    ]
     meta_models = ["LogisticRegression_two_class", "LGBM"]
 
     return RawConfig(
-        directional_models_meta=True,
+        dimensionality_reduction_ratio=0.5,
-        dimensionality_reduction=True,
         n_features_to_select=30,
-        expanding_window_base=True,
+        sliding_window_size=3800,
-        expanding_window_meta=True,
-        sliding_window_size_base=3800,
-        sliding_window_size_meta=2400,
         retrain_every=1000,
         scaler="minmax",  # 'normalize' 'minmax' 'standardize'
         assets=["fivemin_crypto"],

config/sweep_ensemble.yaml

@@ -6,28 +6,20 @@ metric:
   goal: maximize
   name: sharpe
 parameters:
-    directional_models_meta:
-      value: True
     assets:
       value: ['daily_crypto']
     other_assets:
       value: ['daily_etf']
     exogenous_data:
       value: ['daily_glassnode']
-    expanding_window_base:
+    sliding_window_size: 
-      value: True
-    expanding_window_meta:
-      value: True
-    sliding_window_size_base: 
       value: 380
-    sliding_window_size_meta: 
-      values: [250, 300, 380]
       distribution: categorical
     n_features_to_select:
       values: [40, 50, 60]
       distribution: categorical
-    dimensionality_reduction:
+    dimensionality_reduction_ratio:
-      value: True
+      value: 0.5
     retrain_every: 
       value: 20
     scaler:

config/sweep_meta_labeling.yaml

@@ -1,55 +0,0 @@
-program: run_sweep.py
-method: bayes
-project: price-forecasting
-name: Level-2 models
-metric:
-  goal: maximize
-  name: sharpe
-parameters:
-    directional_models_meta:
-      value: True
-    assets:
-      value: ['daily_crypto']
-    other_assets:
-      value: ['daily_etf']
-    exogenous_data:
-      value: ['daily_glassnode']
-    expanding_window_base:
-      values: [True, False]
-      distribution: categorical
-    expanding_window_meta:
-      values: [True, False]
-      distribution: categorical
-    n_features_to_select:
-      values: [10, 20, 30]
-      distribution: categorical
-    dimensionality_reduction:
-      value: True
-    sliding_window_size_base: 
-      values: [180, 280, 380]
-      distribution: categorical
-    sliding_window_size_meta: 
-      values: [180, 280, 380]
-      distribution: categorical
-    retrain_every: 
-      values: [10, 20, 30]
-      distribution: categorical
-    scaler:
-      value: 'minmax'
-    no_of_classes:
-      values: ['two', 'three-balanced', 'three-imbalanced']
-      distribution: categorical
-    load_non_target_asset: 
-      values: [True, False]
-      distribution: categorical
-    directional_models: 
-      value: ["LogisticRegression_two_class", "LDA", "KNN", "CART", "NB", "AB", "RFC", "StaticMom"]
-    meta_models: 
-      values: ["LogisticRegression_two_class", "LDA", "KNN", "CART", "NB", "AB", "RFC"]
-      distribution: categorical
-    own_features:
-      values: [['single_mom', 'date_days'], [], ['level_1', 'date_days'], ['date_days', 'level_2']]
-      distribution: categorical
-    other_features:
-      values: [[], ['level_1']]
-      distribution: categorical

config/sweep_primary_models.yaml

@@ -1,49 +0,0 @@
-program: run_sweep.py
-method: grid
-project: price-forecasting
-name: Level-1 models
-metric:
-  goal: maximize
-  name: sharpe
-parameters:
-    directional_models_meta:
-      value: True
-    assets:
-      value: ['daily_crypto']
-    other_assets:
-      value: ['daily_etf']
-    exogenous_data:
-      value: ['daily_glassnode']
-    expanding_window_base:
-      values: [True, False]
-      distribution: categorical
-    expanding_window_meta:
-      value: False
-    n_features_to_select:
-      value: 50
-    dimensionality_reduction:
-      value: True
-    sliding_window_size_base: 
-      value: 380
-    sliding_window_size_meta: 
-      value: 380
-    retrain_every: 
-      values: [10, 20, 30]
-      distribution: categorical
-    scaler:
-      value: 'minmax'
-    no_of_classes:
-      value: 'two'
-    load_non_target_asset: 
-      value: True
-    directional_models: 
-      values: [['LogisticRegression_two_class'], ['SVC'], ['LDA'], ['KNN'], ['CART'], ['MNB'], ['NB'], ['AB'], ['RFC'], ['XGB_two_class'], ['LGBM']]
-      distribution: categorical
-    meta_models: 
-      value: ["LGBM", "LogisticRegression_two_class"]
-    own_features:
-      value: ['date_days', 'level_2', 'lags_up_to_5']
-    other_features:
-      value: ['level_2', 'lags_up_to_5']
-    exogenous_features:
-      value: ['z_score']

config/types.py

@@ -7,16 +7,13 @@ from feature_extractors.types import FeatureExtractor, ScalerTypes
 from labeling.types import EventFilter, EventLabeller
 from sklearn.base import BaseEstimator
 from dataclasses import dataclass
+from transformations.base import Transformation
 
 # RawConfig is needed to ensure we can declare config presets here with static typing, we then convert it to Config
 class RawConfig(BaseModel):
-    directional_models_meta: bool
+    dimensionality_reduction_ratio: float
-    dimensionality_reduction: bool
     n_features_to_select: int
-    expanding_window_base: bool
+    sliding_window_size: int
-    expanding_window_meta: bool
-    sliding_window_size_base: int
-    sliding_window_size_meta: int
     retrain_every: int
     scaler: Literal["normalize", "minmax", "standardize"]
 
@@ -38,15 +35,8 @@ class RawConfig(BaseModel):
 
 @dataclass
 class Config:
-    directional_models_meta: bool
+    sliding_window_size: int
-    dimensionality_reduction: bool
-    n_features_to_select: int
-    expanding_window_base: bool
-    expanding_window_meta: bool
-    sliding_window_size_base: int
-    sliding_window_size_meta: int
     retrain_every: int
-    scaler: Literal["normalize", "minmax", "standardize"]
 
     assets: DataCollection
     target_asset: DataSource
@@ -66,6 +56,8 @@ class Config:
     directional_model: Model
     meta_model: Model
 
+    transformations: list[Transformation]
+
     @validator("directional_model", "meta_model")
     def check_model(cls, v):
         assert isinstance(v, BaseEstimator)

data_loader/process.py

@@ -20,5 +20,5 @@ def has_enough_samples_to_train(X: XDataFrame, config: Config) -> bool:
     samples_to_train = len(X) - first_valid_index
     return (
         samples_to_train
-        > config.sliding_window_size_base + config.sliding_window_size_meta + 100
+        > (config.sliding_window_size *2) + 100
     )

run_inference.py

@@ -5,7 +5,6 @@ from reporting.saving import load_models
 from run_pipeline import run_pipeline
 from config.types import Config, RawConfig
 from config.presets import (
-    get_dev_config,
     get_default_ensemble_config,
     get_lightweight_ensemble_config,
 )
@@ -56,24 +55,26 @@ def __inference(config: Config, pipeline_outcome: PipelineOutcome):
 
     # 3. Train directional models
     directional_training_outcome = train_directional_model(
-        X,
+        X=X,
-        y,
+        y=y,
-        forward_returns,
+        forward_returns=forward_returns,
-        config,
+        config=config,
-        config.directional_model,
+        model=config.directional_model,
+        transformations=config.transformations,
         from_index=inference_from,
         preloaded_training_step=pipeline_outcome.directional_training,
     )
 
     # 4. Run bet sizing on primary model's output
     bet_sizing_outcome = bet_sizing_with_meta_model(
-        X,
+        X=X,
-        directional_training_outcome.training.predictions,
+        input_predictions=directional_training_outcome.training.predictions,
-        y,
+        y=y,
-        forward_returns,
+        forward_returns=forward_returns,
-        config.meta_model,
+        model=config.meta_model,
-        config,
+        transformations=config.transformations,
-        "meta",
+        config=config,
+        model_suffix="meta",
         from_index=inference_from,
         transformations_over_time=pipeline_outcome.bet_sizing.meta_transformations,
         preloaded_models=pipeline_outcome.bet_sizing.meta_training.model_over_time,

run_pipeline.py

@@ -81,6 +81,7 @@ def __run_training(config: Config) -> PipelineOutcome:
         forward_returns,
         config,
         config.directional_model,
+        config.transformations,
         from_index=None,
         preloaded_training_step=None,
     )
@@ -92,6 +93,7 @@ def __run_training(config: Config) -> PipelineOutcome:
         y,
         forward_returns,
         config.meta_model,
+        config.transformations,
         config,
         "meta",
         None,

training/bet_sizing.py

@@ -9,9 +9,7 @@ from typing import Optional
 from config.types import Config
 from .types import BetSizingWithMetaOutcome, ModelOverTime, TransformationsOverTime
 from training.walk_forward import walk_forward_process_transformations
-from transformations.scaler import get_scaler
+from transformations.base import Transformation
-from transformations.rfe import RFETransformation
-from transformations.pca import PCATransformation
 
 
 def bet_sizing_with_meta_model(
@@ -20,6 +18,7 @@ def bet_sizing_with_meta_model(
     y: ySeries,
     forward_returns: ForwardReturnSeries,
     model: Model,
+    transformations: list[Transformation],
     config: Config,
     model_suffix: str,
     from_index: Optional[pd.Timestamp],
@@ -44,21 +43,10 @@ def bet_sizing_with_meta_model(
             X=meta_X,
             y=meta_y,
             forward_returns=forward_returns,
-            expanding_window=config.expanding_window_meta,
+            window_size=config.sliding_window_size,
-            window_size=config.sliding_window_size_meta,
             retrain_every=config.retrain_every,
             from_index=from_index,
-            transformations=[
+            transformations=transformations,
-                get_scaler(config.scaler),
-                PCATransformation(
-                    ratio_components_to_keep=0.5,
-                    sliding_window_size=config.sliding_window_size_meta,
-                ),
-                RFETransformation(
-                    n_feature_to_select=40,
-                    model=default_feature_selector_classification,
-                ),
-            ],
         )
 
     meta_outcome = train_model(
@@ -67,8 +55,7 @@ def bet_sizing_with_meta_model(
         y=meta_y,
         forward_returns=forward_returns,
         model=model,
-        expanding_window=config.expanding_window_meta,
+        sliding_window_size=config.sliding_window_size,
-        sliding_window_size=config.sliding_window_size_meta,
         retrain_every=config.retrain_every,
         from_index=from_index,
         no_of_classes="two",

training/directional_training.py

@@ -1,5 +1,7 @@
 import pandas as pd
 
+from transformations.base import Transformation
+
 from .types import DirectionalTrainingOutcome, TrainingOutcome
 from training.train_model import train_model
 from training.walk_forward import walk_forward_process_transformations
@@ -7,11 +9,6 @@ from training.walk_forward import walk_forward_process_transformations
 from typing import Optional
 from config.types import Config
 from models.base import Model
-from models.model_map import default_feature_selector_classification
-
-from transformations.scaler import get_scaler
-from transformations.rfe import RFETransformation
-from transformations.pca import PCATransformation
 
 
 def train_directional_model(
@@ -20,6 +17,7 @@ def train_directional_model(
     forward_returns: pd.Series,
     config: Config,
     model: Model,
+    transformations: list[Transformation],
     from_index: Optional[pd.Timestamp],
     preloaded_training_step: Optional[DirectionalTrainingOutcome] = None,
 ) -> DirectionalTrainingOutcome:
@@ -30,21 +28,10 @@ def train_directional_model(
             X=X,
             y=y,
             forward_returns=forward_returns,
-            expanding_window=config.expanding_window_base,
+            window_size=config.sliding_window_size,
-            window_size=config.sliding_window_size_base,
             retrain_every=config.retrain_every,
             from_index=from_index,
-            transformations=[
+            transformations=transformations,
-                get_scaler(config.scaler),
-                PCATransformation(
-                    ratio_components_to_keep=0.5,
-                    sliding_window_size=config.sliding_window_size_base,
-                ),
-                RFETransformation(
-                    n_feature_to_select=40,
-                    model=default_feature_selector_classification,
-                ),
-            ],
         )
     else:
         transformations_over_time = preloaded_training_step.transformations
@@ -55,8 +42,7 @@ def train_directional_model(
         y=y,
         forward_returns=forward_returns,
         model=model,
-        expanding_window=config.expanding_window_base,
+        sliding_window_size=config.sliding_window_size,
-        sliding_window_size=config.sliding_window_size_base,
         retrain_every=config.retrain_every,
         from_index=from_index,
         no_of_classes=config.no_of_classes,

training/train_model.py

@@ -16,7 +16,6 @@ def train_model(
     y: pd.Series,
     forward_returns: pd.Series,
     model: Model,
-    expanding_window: bool,
     sliding_window_size: int,
     retrain_every: int,
     from_index: Optional[pd.Timestamp],
@@ -34,7 +33,7 @@ def train_model(
             X=X,
             y=y,
             forward_returns=forward_returns,
-            expanding_window=expanding_window,
+            expanding_window=True,
             window_size=sliding_window_size,
             retrain_every=retrain_every,
             from_index=from_index,
@@ -57,7 +56,7 @@ def train_model(
         model_over_time=model_over_time,
         transformations_over_time=transformations_over_time,
         X=X,
-        expanding_window=expanding_window,
+        expanding_window=True,
         window_size=sliding_window_size,
         retrain_every=retrain_every,
         from_index=from_index,

training/walk_forward/process_transformations.py

@@ -11,7 +11,6 @@ def walk_forward_process_transformations(
     X: XDataFrame,
     y: ySeries,
     forward_returns: ForwardReturnSeries,
-    expanding_window: bool,
     window_size: int,
     retrain_every: int,
     from_index: Optional[pd.Timestamp],
@@ -36,11 +35,7 @@ def walk_forward_process_transformations(
     iterations_before_retrain = 0
 
     for index in tqdm(range(train_from, train_till)):
-        train_window_start = (
+        train_window_start = X.index[first_nonzero_return]
-            X.index[first_nonzero_return]
-            if expanding_window
-            else X.index[index - window_size - 1]
-        )
 
         if iterations_before_retrain <= 0 or pd.isna(
             transformations_over_time[0][index - 1]

training/walk_forward/process_transformations_parallel.py

@@ -13,7 +13,6 @@ def walk_forward_process_transformations(
     X: XDataFrame,
     y: ySeries,
     forward_returns: ForwardReturnSeries,
-    expanding_window: bool,
     window_size: int,
     retrain_every: int,
     from_index: Optional[pd.Timestamp],
@@ -40,7 +39,6 @@ def walk_forward_process_transformations(
             preprocess_transformations_window.remote(
                 X,
                 y,
-                expanding_window,
                 window_size,
                 transformations,
                 first_nonzero_return,
@@ -63,17 +61,12 @@ def walk_forward_process_transformations(
 def preprocess_transformations_window(
     X: XDataFrame,
     y: ySeries,
-    expanding_window: bool,
     window_size: int,
     transformations: list[Transformation],
     first_nonzero_return: int,
     index: int,
 ) -> tuple[list[Transformation], int]:
-    train_window_start = (
+    train_window_start = X.index[first_nonzero_return]
-        X.index[first_nonzero_return]
-        if expanding_window
-        else X.index[index - window_size - 1]
-    )
     train_window_end = X.index[index - 1]
 
     X_expanding_window = X[train_window_start:train_window_end]

transformations/retrieve.py

@@ -0,0 +1,42 @@
+from .rfe import RFETransformation
+from .pca import PCATransformation
+from .sklearn import SKLearnTransformation
+from typing import Literal, Optional
+from models.model_map import default_feature_selector_classification
+from sklearn.preprocessing import MinMaxScaler, Normalizer, StandardScaler
+
+ScalerTypes = Literal["normalize", "minmax", "standardize"]
+
+
+def get_rfe(n_feature_to_select: int) -> Optional[RFETransformation]:
+    if n_feature_to_select > 0:
+        return RFETransformation(
+            n_feature_to_select=n_feature_to_select,
+            model=default_feature_selector_classification,
+        )
+    else:
+        return None
+
+
+def get_pca(
+    ratio_components_to_keep: float, sliding_window_size: int
+) -> Optional[PCATransformation]:
+
+    if ratio_components_to_keep > 0:
+        return PCATransformation(
+            ratio_components_to_keep=ratio_components_to_keep,
+            sliding_window_size=sliding_window_size,
+        )
+    else:
+        return None
+
+
+def get_scaler(type: ScalerTypes) -> SKLearnTransformation:
+    if type == "normalize":
+        return SKLearnTransformation(Normalizer())
+    elif type == "minmax":
+        return SKLearnTransformation(MinMaxScaler(feature_range=(-1, 1)))
+    elif type == "standardize":
+        return SKLearnTransformation(StandardScaler())
+    else:
+        raise Exception("Scaler type not supported")

transformations/scaler.py

@@ -1,16 +0,0 @@
-from sklearn.preprocessing import MinMaxScaler, Normalizer, StandardScaler
-from .sklearn import SKLearnTransformation
-from typing import Literal
-
-ScalerTypes = Literal["normalize", "minmax", "standardize"]
-
-
-def get_scaler(type: ScalerTypes) -> SKLearnTransformation:
-    if type == "normalize":
-        return SKLearnTransformation(Normalizer())
-    elif type == "minmax":
-        return SKLearnTransformation(MinMaxScaler(feature_range=(-1, 1)))
-    elif type == "standardize":
-        return SKLearnTransformation(StandardScaler())
-    else:
-        raise Exception("Scaler type not supported")