dslighting 1.7.1__py3-none-any.whl → 1.7.8__py3-none-any.whl

mlebench/competitions/dog-breed-identification/dogs.py

@@ -0,0 +1,124 @@
+_dogs_str = """
+affenpinscher
+afghan_hound
+african_hunting_dog
+airedale
+american_staffordshire_terrier
+appenzeller
+australian_terrier
+basenji
+basset
+beagle
+bedlington_terrier
+bernese_mountain_dog
+black-and-tan_coonhound
+blenheim_spaniel
+bloodhound
+bluetick
+border_collie
+border_terrier
+borzoi
+boston_bull
+bouvier_des_flandres
+boxer
+brabancon_griffon
+briard
+brittany_spaniel
+bull_mastiff
+cairn
+cardigan
+chesapeake_bay_retriever
+chihuahua
+chow
+clumber
+cocker_spaniel
+collie
+curly-coated_retriever
+dandie_dinmont
+dhole
+dingo
+doberman
+english_foxhound
+english_setter
+english_springer
+entlebucher
+eskimo_dog
+flat-coated_retriever
+french_bulldog
+german_shepherd
+german_short-haired_pointer
+giant_schnauzer
+golden_retriever
+gordon_setter
+great_dane
+great_pyrenees
+greater_swiss_mountain_dog
+groenendael
+ibizan_hound
+irish_setter
+irish_terrier
+irish_water_spaniel
+irish_wolfhound
+italian_greyhound
+japanese_spaniel
+keeshond
+kelpie
+kerry_blue_terrier
+komondor
+kuvasz
+labrador_retriever
+lakeland_terrier
+leonberg
+lhasa
+malamute
+malinois
+maltese_dog
+mexican_hairless
+miniature_pinscher
+miniature_poodle
+miniature_schnauzer
+newfoundland
+norfolk_terrier
+norwegian_elkhound
+norwich_terrier
+old_english_sheepdog
+otterhound
+papillon
+pekinese
+pembroke
+pomeranian
+pug
+redbone
+rhodesian_ridgeback
+rottweiler
+saint_bernard
+saluki
+samoyed
+schipperke
+scotch_terrier
+scottish_deerhound
+sealyham_terrier
+shetland_sheepdog
+shih-tzu
+siberian_husky
+silky_terrier
+soft-coated_wheaten_terrier
+staffordshire_bullterrier
+standard_poodle
+standard_schnauzer
+sussex_spaniel
+tibetan_mastiff
+tibetan_terrier
+toy_poodle
+toy_terrier
+vizsla
+walker_hound
+weimaraner
+welsh_springer_spaniel
+west_highland_white_terrier
+whippet
+wire-haired_fox_terrier
+yorkshire_terrier
+"""
+DOGS_LIST = _dogs_str.strip().split("\n")
+assert len(DOGS_LIST) == 120

mlebench/competitions/dog-breed-identification/grade.py

@@ -0,0 +1,42 @@
+import numpy as np
+from pandas import DataFrame
+from sklearn.metrics import log_loss
+
+from mlebench.competitions.utils import one_hot_dfs_to_log_loss_inputs
+from mlebench.grade_helpers import InvalidSubmissionError
+
+from .dogs import DOGS_LIST
+
+
+def prepare_for_metric(submission: DataFrame, answers: DataFrame) -> dict:
+    if not all(dog in submission.columns for dog in DOGS_LIST):
+        raise InvalidSubmissionError(f"Submission must have columns for all dogs: {DOGS_LIST}")
+    if "id" not in submission.columns:
+        raise InvalidSubmissionError("Submission must have an `id` column")
+    if len(submission) != len(answers):
+        raise InvalidSubmissionError("Submission should be the same length as the answers")
+
+    assert "id" in answers.columns, "Answers must have an `id` column"
+    assert all(
+        dog in answers.columns for dog in DOGS_LIST
+    ), f"Answers must have columns for all dogs: {DOGS_LIST}"
+
+    tolerance = 1e-6
+    if not np.all(np.isclose(submission[DOGS_LIST].sum(axis=1), 1, atol=tolerance)):
+        raise InvalidSubmissionError(
+            "Dog probabilities in each row in submission should sum to one, as probabilities."
+        )
+    if not ((submission[DOGS_LIST] >= 0) & (submission[DOGS_LIST] <= 1)).all().all():
+        raise InvalidSubmissionError(
+            "All probabilities in submission DataFrame must be between 0 and 1."
+        )
+
+    log_loss_inputs = one_hot_dfs_to_log_loss_inputs(
+        submission, answers, id_column="id", apply_softmax=False
+    )
+    return log_loss_inputs
+
+
+def grade(submission: DataFrame, answers: DataFrame) -> float:
+    log_loss_inputs = prepare_for_metric(submission, answers)
+    return log_loss(**log_loss_inputs)

mlebench/competitions/dog-breed-identification/prepare.py

@@ -0,0 +1,55 @@
+import shutil
+from pathlib import Path
+
+from sklearn.model_selection import train_test_split
+
+from mlebench.competitions.utils import df_to_one_hot
+from mlebench.utils import read_csv
+
+from .dogs import DOGS_LIST
+
+
+def to_one_hot(df):
+    return df_to_one_hot(df, id_column="id", target_column="breed", classes=DOGS_LIST)
+
+
+def prepare(raw: Path, public: Path, private: Path):
+
+    # Create train, test from train split
+    old_train = read_csv(raw / "labels.csv")
+    new_train, new_test = train_test_split(old_train, test_size=0.1, random_state=0)
+    # one hot the private test because we will one-hot the submission, as per kaggle.com
+    new_test = to_one_hot(new_test)
+
+    # Copy over files
+    new_train.to_csv(public / "labels.csv", index=False)
+    new_test.to_csv(private / "test.csv", index=False)
+
+    (public / "train").mkdir(exist_ok=True)
+
+    for file_id in new_train["id"]:
+        shutil.copyfile(
+            src=raw / "train" / f"{file_id}.jpg",
+            dst=public / "train" / f"{file_id}.jpg",
+        )
+
+    (public / "test").mkdir(exist_ok=True)
+
+    for file_id in new_test["id"]:
+        shutil.copyfile(
+            src=raw / "train" / f"{file_id}.jpg",
+            dst=public / "test" / f"{file_id}.jpg",
+        )
+
+    # Check integrity of the files copied
+    assert len(list(public.glob("train/*.jpg"))) == len(new_train)
+    assert len(list(public.glob("test/*.jpg"))) == len(new_test)
+
+    # Create a sample submission file
+    submission_df = new_test.copy()
+    for col in submission_df.columns[1:]:
+        submission_df[col] = submission_df[col].astype("float64")
+    submission_df.iloc[:, 1:] = 1 / 120
+    submission_df.to_csv(public / "sample_submission.csv", index=False)
+
+    assert submission_df.shape == (len(new_test), 121)  # 1 id column + 120 breeds

mlebench/competitions/dog-breed-identification/prepare_val.py

@@ -0,0 +1,104 @@
+import shutil
+from pathlib import Path
+
+import pandas as pd
+from sklearn.model_selection import train_test_split
+
+from mlebench.competitions.utils import df_to_one_hot
+from mlebench.utils import read_csv
+
+from .dogs import DOGS_LIST
+
+
+def to_one_hot(df: pd.DataFrame) -> pd.DataFrame:
+    return df_to_one_hot(df, id_column="id", target_column="breed", classes=DOGS_LIST)
+
+
+def _save_split(
+    raw_data_path: Path,
+    public_dir: Path,
+    private_dir: Path,
+    train_df: pd.DataFrame,
+    test_df: pd.DataFrame,
+):
+    """Saves a given train/test split to the specified public and private directories."""
+    # Ensure target directories exist
+    public_dir.mkdir(exist_ok=True)
+    private_dir.mkdir(exist_ok=True)
+
+    # one hot the private test because we will one-hot the submission, as per kaggle.com
+    test_labels_private = to_one_hot(test_df.copy())
+
+    # Copy over files
+    train_df.to_csv(public_dir / "labels.csv", index=False)
+    test_labels_private.to_csv(private_dir / "test.csv", index=False)
+
+    (public_dir / "train").mkdir(exist_ok=True)
+    for file_id in train_df["id"]:
+        shutil.copyfile(
+            src=raw_data_path / "train" / f"{file_id}.jpg",
+            dst=public_dir / "train" / f"{file_id}.jpg",
+        )
+
+    (public_dir / "test").mkdir(exist_ok=True)
+    for file_id in test_df["id"]:
+        shutil.copyfile(
+            src=raw_data_path / "train" / f"{file_id}.jpg",
+            dst=public_dir / "test" / f"{file_id}.jpg",
+        )
+
+    # Check integrity of the files copied
+    assert len(list(public_dir.glob("train/*.jpg"))) == len(train_df)
+    assert len(list(public_dir.glob("test/*.jpg"))) == len(test_df)
+
+    # Create a sample submission file
+    submission_df = test_labels_private.copy()
+    for col in submission_df.columns[1:]:
+        submission_df[col] = submission_df[col].astype("float64")
+    submission_df.iloc[:, 1:] = 1 / 120
+    submission_df.to_csv(public_dir / "sample_submission.csv", index=False)
+
+    assert submission_df.shape == (len(test_df), 121)  # 1 id column + 120 breeds
+
+
+def prepare(raw: Path, public: Path, private: Path):
+    # Read the full raw dataset labels
+    all_labels = read_csv(raw / "labels.csv")
+
+    # --- 1. Original Split: Create main train and test sets ---
+    # This split and its outputs must remain identical to the original script.
+    train_df, test_df = train_test_split(all_labels, test_size=0.1, random_state=0)
+
+    # Save the original split to the 'public' and 'private' directories
+    _save_split(
+        raw_data_path=raw,
+        public_dir=public,
+        private_dir=private,
+        train_df=train_df,
+        test_df=test_df,
+    )
+
+    # --- 2. New Split: Create a validation set from the main train set ---
+    # Define paths for the new validation directories, parallel to the original ones.
+    public_val = public.parent / "public_val"
+    private_val = private.parent / "private_val"
+
+    # To ensure the new validation set ('test_val_df') is the same size as the
+    # original test set, we calculate the required test_size for the second split.
+    test_size_for_val_split = len(test_df) / len(train_df)
+
+    # Split the main training data (train_df) into a smaller training set
+    # and a validation set, using the same random_state for reproducibility.
+    train_val_df, test_val_df = train_test_split(
+        train_df, test_size=test_size_for_val_split, random_state=0
+    )
+
+    # Save the new validation split to the 'public_val' and 'private_val' directories
+    # using the same helper function to ensure identical structure and filenames.
+    _save_split(
+        raw_data_path=raw,
+        public_dir=public_val,
+        private_dir=private_val,
+        train_df=train_val_df,
+        test_df=test_val_df,
+    )

mlebench/competitions/dogs-vs-cats-redux-kernels-edition/grade.py

@@ -0,0 +1,43 @@
+from typing import Tuple
+
+import numpy as np
+import pandas as pd
+from sklearn.metrics import log_loss
+
+from mlebench.grade_helpers import InvalidSubmissionError
+
+
+def prepare_for_metric(
+    submission: pd.DataFrame, answers: pd.DataFrame
+) -> Tuple[np.ndarray, np.ndarray]:
+    # answers checks
+    assert "id" in answers.columns, f"Answers is missing `id` column"
+    assert "label" in answers.columns, f"Answers is missing `label` column"
+
+    # submission checks
+    if "id" not in submission.columns:
+        raise InvalidSubmissionError(f"Submission is missing `id` column")
+    if "label" not in submission.columns:
+        raise InvalidSubmissionError(f"Submission is missing `label` column")
+    if set(submission["id"]) != set(answers["id"]):
+        raise InvalidSubmissionError("Submission and answers have different id's")
+    if len(submission) != len(answers):
+        raise InvalidSubmissionError("Submission and answers have different lengths")
+    if not ((submission["label"] >= 0) & (submission["label"] <= 1)).all():
+        raise InvalidSubmissionError(
+            "All values in submission `label` column must be between 0 and 1."
+        )
+
+    # sort by id to ensure correct order
+    submission = submission.sort_values("id")
+    answers = answers.sort_values("id")
+
+    y_true = answers["label"]
+    y_pred = submission["label"]
+    return y_true, y_pred
+
+
+def grade(submission: pd.DataFrame, answers: pd.DataFrame) -> float:
+    y_true, y_pred = prepare_for_metric(submission, answers)
+    score = log_loss(y_true=y_true, y_pred=y_pred)
+    return score

mlebench/competitions/dogs-vs-cats-redux-kernels-edition/prepare.py

@@ -0,0 +1,70 @@
+import shutil
+from pathlib import Path
+
+import pandas as pd
+from sklearn.model_selection import train_test_split
+from tqdm import tqdm
+
+from mlebench.utils import compress, extract, read_csv
+
+
+def prepare(raw: Path, public: Path, private: Path):
+    extract(raw / "train.zip", raw)
+    extract(raw / "test.zip", raw)
+
+    all_train_images = sorted(list((raw / "train").glob("*.jpg")))
+    # Original test ratio has Train set - 25,000 samples; Test set - 12,500 samples (33% ratio)
+    # We use 0.1 ratio to avoid removing too many samples from train
+    train_images, test_images = train_test_split(all_train_images, test_size=0.1, random_state=0)
+
+    # Copy over train images. Rename cat files to cat.0.jpg, cat.1.jpg, etc.
+    # Rename dog files to dog.0.jpg, dog.1.jpg, etc.
+    cat_ctr = 0
+    dog_ctr = 0
+    (public / "train").mkdir(exist_ok=True)
+    for img in tqdm(train_images):
+        if "cat" in img.name:
+            shutil.copy(img, public / "train" / f"cat.{cat_ctr}.jpg")
+            cat_ctr += 1
+        else:
+            shutil.copy(img, public / "train" / f"dog.{dog_ctr}.jpg")
+            dog_ctr += 1
+    assert cat_ctr + dog_ctr == len(
+        train_images
+    ), f"Expected {len(train_images)} train images but got {cat_ctr + dog_ctr} images."
+
+    # Copy over test images. Rename files to 1.jpg, 2.jpg, etc.
+    (public / "test").mkdir(exist_ok=True)
+    for i, img in enumerate(tqdm(test_images), start=1):
+        shutil.copy(img, public / "test" / f"{i}.jpg")
+    assert i == len(test_images), f"Expected {len(test_images)} test images but got {i} images."
+
+    # Compress train and test images, then remove the raw images
+    compress(public / "train", public / "train.zip", exist_ok=True)
+    compress(public / "test", public / "test.zip", exist_ok=True)
+    shutil.rmtree(raw / "train")
+    shutil.rmtree(raw / "test")
+
+    # Make answers
+    answers = pd.DataFrame(
+        {
+            "id": [i for i in range(1, len(test_images) + 1)],
+            "label": [int("dog" in img.name) for img in test_images],
+        }
+    )
+    answers.to_csv(private / "answers.csv", index=False)
+    assert len(answers) == len(
+        test_images
+    ), f"Expected {len(test_images)} answers but got {len(answers)} answers."
+
+    # Make sample submission
+    sample_submission = pd.DataFrame(
+        {
+            "id": [i for i in range(1, len(test_images) + 1)],
+            "label": [0.5 for _ in range(1, len(test_images) + 1)],
+        }
+    )
+    sample_submission.to_csv(public / "sample_submission.csv", index=False)
+    assert len(sample_submission) == len(
+        test_images
+    ), f"Expected {len(test_images)} sample submission rows but got {len(sample_submission)} rows."

mlebench/competitions/dogs-vs-cats-redux-kernels-edition/prepare_val.py

@@ -0,0 +1,143 @@
+import shutil
+from pathlib import Path
+from typing import List
+
+import pandas as pd
+from sklearn.model_selection import train_test_split
+from tqdm import tqdm
+
+from mlebench.utils import compress, extract
+
+
+def _process_split(
+    image_paths: List[Path],
+    public_dir: Path,
+    private_dir: Path,
+    test_size: float,
+    random_state: int,
+) -> List[Path]:
+    """
+    Splits a list of image paths into train/test sets and generates all required files.
+
+    This function encapsulates the logic for:
+    1. Splitting data.
+    2. Copying and renaming images to public train/test subdirectories.
+    3. Compressing the public train/test image directories.
+    4. Creating the private answer key for the test set.
+    5. Creating a public sample submission file.
+
+    Args:
+        image_paths: A list of Path objects for the images to be split.
+        public_dir: The public output directory (e.g., 'data/public').
+        private_dir: The private output directory (e.g., 'data/private').
+        test_size: The proportion of the dataset to allocate to the test set.
+        random_state: The seed for the random number generator.
+
+    Returns:
+        A list of Path objects corresponding to the training set of this split.
+    """
+    public_dir.mkdir(exist_ok=True, parents=True)
+    private_dir.mkdir(exist_ok=True, parents=True)
+
+    # Perform the split
+    train_images, test_images = train_test_split(
+        image_paths, test_size=test_size, random_state=random_state
+    )
+
+    # Copy over train images. Rename cat files to cat.0.jpg, cat.1.jpg, etc.
+    # Rename dog files to dog.0.jpg, dog.1.jpg, etc.
+    cat_ctr = 0
+    dog_ctr = 0
+    (public_dir / "train").mkdir(exist_ok=True)
+    for img in tqdm(train_images, desc=f"Processing train set for {public_dir.name}"):
+        if "cat" in img.name:
+            shutil.copy(img, public_dir / "train" / f"cat.{cat_ctr}.jpg")
+            cat_ctr += 1
+        else:
+            shutil.copy(img, public_dir / "train" / f"dog.{dog_ctr}.jpg")
+            dog_ctr += 1
+    assert cat_ctr + dog_ctr == len(
+        train_images
+    ), f"Expected {len(train_images)} train images but got {cat_ctr + dog_ctr} images."
+
+    # Copy over test images. Rename files to 1.jpg, 2.jpg, etc.
+    (public_dir / "test").mkdir(exist_ok=True)
+    for i, img in enumerate(tqdm(test_images, desc=f"Processing test set for {public_dir.name}"), start=1):
+        shutil.copy(img, public_dir / "test" / f"{i}.jpg")
+    assert i == len(test_images), f"Expected {len(test_images)} test images but got {i} images."
+
+    # Compress train and test images
+    compress(public_dir / "train", public_dir / "train.zip", exist_ok=True)
+    compress(public_dir / "test", public_dir / "test.zip", exist_ok=True)
+
+    # Make answers
+    answers = pd.DataFrame(
+        {
+            "id": [i for i in range(1, len(test_images) + 1)],
+            "label": [int("dog" in img.name) for img in test_images],
+        }
+    )
+    answers.to_csv(private_dir / "answers.csv", index=False)
+    assert len(answers) == len(
+        test_images
+    ), f"Expected {len(test_images)} answers but got {len(answers)} answers."
+
+    # Make sample submission
+    sample_submission = pd.DataFrame(
+        {
+            "id": [i for i in range(1, len(test_images) + 1)],
+            "label": [0.5 for _ in range(1, len(test_images) + 1)],
+        }
+    )
+    sample_submission.to_csv(public_dir / "sample_submission.csv", index=False)
+    assert len(sample_submission) == len(
+        test_images
+    ), f"Expected {len(test_images)} sample submission rows but got {len(sample_submission)} rows."
+
+    return train_images
+
+
+def prepare(raw: Path, public: Path, private: Path):
+    # This part remains from the original script
+    extract(raw / "train.zip", raw)
+    extract(raw / "test.zip", raw)
+
+    all_train_images = sorted(list((raw / "train").glob("*.jpg")))
+    
+    # --- 1. Original Split (train -> train + test) ---
+    # This call generates the original competition files in `public` and `private`.
+    # The logic and outputs of this step are unchanged.
+    # Original test ratio has Train set - 25,000 samples; Test set - 12,500 samples (33% ratio)
+    # We use 0.1 ratio to avoid removing too many samples from train
+    original_test_size = 0.1
+    main_train_set = _process_split(
+        image_paths=all_train_images,
+        public_dir=public,
+        private_dir=private,
+        test_size=original_test_size,
+        random_state=0,
+    )
+
+    # --- 2. New Validation Split (main_train_set -> train_val + test_val) ---
+    # This call generates a new, parallel set of files for validation purposes.
+    # It operates *only* on the training data from the first split.
+    public_val = public.parent / "public_val"
+    private_val = private.parent / "private_val"
+
+    # To make the new test_val set have the same size as the original test set,
+    # we adjust the test_size for the second split.
+    # new_size = original_test_size / (1 - original_test_size)
+    # e.g., 0.1 / (1 - 0.1) = 0.1 / 0.9 = 1/9
+    validation_test_size = original_test_size / (1 - original_test_size)
+
+    _process_split(
+        image_paths=main_train_set,
+        public_dir=public_val,
+        private_dir=private_val,
+        test_size=validation_test_size,
+        random_state=0,  # Use the same random state for consistency
+    )
+
+    # Final cleanup is done after all splits are complete
+    shutil.rmtree(raw / "train")
+    shutil.rmtree(raw / "test")

mlebench/competitions/ethanol-concentration/grade.py

@@ -0,0 +1,23 @@
+import pandas as pd
+import numpy as np
+from sklearn.metrics import accuracy_score
+
+
+def grade(submission: pd.DataFrame, answers: pd.DataFrame) -> float:
+    """
+    Grade the submission using accuracy metric.
+
+    Args:
+        submission: DataFrame with columns ['id', 'label']
+        answers: DataFrame with columns ['id', 'label']
+
+    Returns:
+        Accuracy score (float between 0 and 1)
+    """
+    # Merge on id to ensure alignment
+    merged = pd.merge(answers, submission, on='id', suffixes=('_true', '_pred'))
+
+    # Calculate accuracy
+    accuracy = accuracy_score(merged['label_true'], merged['label_pred'])
+
+    return accuracy