Better random forests

.gitignore

@@ -1,5 +1,7 @@
 data_very_raw/
 zzz_raman_spectroscopy-main/
+presentations/
+to_ignore/
 
 # ---> JupyterNotebooks
 # gitignore template for Jupyter Notebooks

classifiers/__init__.py

@@ -0,0 +1 @@
+from classifiers.evaluation import *

classifiers/evaluation.py

@@ -0,0 +1,35 @@
+from data.data_processing import process_train_test
+from sklearn.model_selection import cross_validate, ParameterGrid
+
+
+def crossvalidate_params(classifier, params, experiments_train, metadata_train, y_train, cv=5):
+    process_params = {key: params[key] for key in ['baseline_lam', 'baseline_p', 'smooth_window_length', 'smooth_polyorder']}
+    classifier_params = {key: params[key] for key in params.keys() if key not in ['baseline_lam', 'baseline_p', 'smooth_window_length', 'smooth_polyorder']}
+    X_train, _ = process_train_test(process_params, experiments_train, metadata_train, scale=True)
+    clf = classifier(**classifier_params)
+    return cross_validate(clf, X_train, y_train.to_numpy().ravel(), cv=cv, return_estimator=True)
+
+
+def param_grid_search(classifier, param_grid, experiments_train, metadata_train, y_train, cv=5):
+    results = []
+    for params in ParameterGrid(param_grid):
+        try:
+            results.append([params, crossvalidate_params(classifier, params, experiments_train, metadata_train, y_train, cv=cv)])
+            print(results[-1])
+        except Exception as e:
+            pass  # print(params, e)
+    return results
+
+
+def evaluate_classifier_params(classifier, params, X_train, y_train, X_test, y_test, iters=10):
+    train_score_mean = 0
+    test_score_mean = 0
+    for i in range(iters):
+        clf = classifier(**params)
+        clf.fit(X_train, y_train.to_numpy().ravel())
+        train_score_mean += clf.score(X_train, y_train.to_numpy().ravel())
+        test_score_mean += clf.score(X_test, y_test.to_numpy().ravel())
+    return train_score_mean / iters, test_score_mean / iters
+
+
+

data/data_loading.py

@@ -57,5 +57,5 @@ def load_raw_data() -> tuple[pd.DataFrame, list[pd.DataFrame]]:
 
 def load_data(name: str, path: os.path = os.path.join("data")) -> tuple[pd.DataFrame, pd.DataFrame]:
     metadata = pd.read_csv(os.path.join(path, name, "metadata.csv"))
-    experiments = pd.read_csv(os.path.join(path, name, "experiments.csv"))
+    experiments = pd.read_csv(os.path.join(path, name, "experiments.csv"), dtype=float)
     return metadata, experiments

data/data_processing.py

@@ -80,9 +80,26 @@ def categorize_metadata(metadata: pd.DataFrame) -> tuple[pd.DataFrame, pd.DataFr
     return truth, encoded
 
 
-def process_experiments(experiments: pd.DataFrame, baseline_lam=10, baseline_p=1e-2,
-                        smooth_window_length=7, smooth_polyorder=3) -> pd.DataFrame:
+def process_experiments(experiments: pd.DataFrame, baseline_lam: int = 10, baseline_p: float = 1e-2,
+                        smooth_window_length: int = 7, smooth_polyorder: int = 3) -> pd.DataFrame:
     experiments = adjust_all_baselines(experiments, lam=baseline_lam, p=baseline_p)
     experiments = scale_experiments(experiments)
     experiments = smooth_experiments(experiments, window_length=smooth_window_length, polyorder=smooth_polyorder)
     return experiments
+
+
+def process_train_test(params: dict, experiments_train: pd.DataFrame, metadata_train: pd.DataFrame, experiments_test: pd.DataFrame = None, metadata_test: pd.DataFrame = None, scale: bool=True) -> tuple[pd.DataFrame, pd.DataFrame]:
+    processed_train = process_experiments(experiments_train, **params)
+    X_train = pd.concat([metadata_train, processed_train], axis=1)
+    if experiments_test is not None:
+        processed_test = process_experiments(experiments_test, **params)
+        X_test = pd.concat([metadata_test, processed_test], axis=1)
+    else:
+        X_test = None
+    if scale:
+        scaler = StandardScaler()
+        scaler.fit(X_train)
+        X_train = scaler.transform(X_train)
+        if X_test is not None:
+            X_test = scaler.transform(X_test)
+    return X_train, X_test