公榜62
私榜650，嘤嘤嘤
用了三个notebook合成，代码放下面
第一个notebook
https://www.kaggle.com/code/richolson/isic-2024-magic-noise-for-lb-overfit第二个notebook
导入包
import os
from pathlib import Path
import numpy as np
import pandas as pd
import polars as pl
from sklearn.model_selection import StratifiedGroupKFold
from sklearn.model_selection import cross_val_score
from sklearn.preprocessing import OneHotEncoder
from sklearn.metrics import roc_auc_score
from sklearn.ensemble import VotingClassifier
from imblearn.under_sampling import RandomUnderSampler
from imblearn.pipeline import Pipeline
from imblearn.over_sampling import RandomOverSampler
import lightgbm as lgb
import catboost as cb
import xgboost as xgb
import optuna
from optuna.samplers import TPESampler
#downsampling techniques
# they took long time, so we use RandomUnderSampler
from imblearn.under_sampling import NearMiss
from imblearn.under_sampling import ClusterCentroids
from imblearn.under_sampling import TomekLinks
from imblearn.under_sampling import EditedNearestNeighbours
from imblearn.pipeline import Pipeline as ImbPipeline
from imblearn.under_sampling import NearMiss, TomekLinks
from sklearn.impute import SimpleImputer
import time
from sklearn.feature_selection import SelectKBest, chi2, mutual_info_classif, VarianceThreshold
!python /kaggle/input/isic-2024-pl-submission-script-and-preds/pl_submission.py
!mv submission.csv submission_image3.csv
!python /kaggle/input/ser-sub/main.py
!mv submission.csv submission_ser.csv
!python /kaggle/input/fb-sub/main.py
!mv submission.csv submission_fb.csv
!python /kaggle/input/isic-script-inference-effnetv1b0-f313ae/main.py /kaggle/input/isic-pytorch-training-baseline-image-only/AUROC0.5171_Loss0.3476_epoch35.bin
!mv submission.csv submission_effnetv1b0.csv
!python /kaggle/input/dense-sub-cat/main.py
!mv submission.csv submission_den.csv代码
root = Path('/kaggle/input/isic-2024-challenge')
train_path = root / 'train-metadata.csv'
test_path = root / 'test-metadata.csv'
subm_path = root / 'sample_submission.csv'
id_col = 'isic_id'
target_col = 'target'
group_col = 'patient_id'
err = 1e-5
sampling_ratio = 0.01
seed = 42
num_cols = [
'age_approx', # Approximate age of patient at time of imaging.
'clin_size_long_diam_mm', # Maximum diameter of the lesion (mm).+
'tbp_lv_A', # A inside lesion.+
'tbp_lv_Aext', # A outside lesion.+
'tbp_lv_B', # B inside lesion.+
'tbp_lv_Bext', # B outside lesion.+
'tbp_lv_C', # Chroma inside lesion.+
'tbp_lv_Cext', # Chroma outside lesion.+
'tbp_lv_H', # Hue inside the lesion; calculated as the angle of A* and B* in LAB* color space. Typical values range from 25 (red) to 75 (brown).+
'tbp_lv_Hext', # Hue outside lesion.+
'tbp_lv_L', # L inside lesion.+
'tbp_lv_Lext', # L outside lesion.+
'tbp_lv_areaMM2', # Area of lesion (mm^2).+
'tbp_lv_area_perim_ratio', # Border jaggedness, the ratio between lesions perimeter and area. Circular lesions will have low values; irregular shaped lesions will have higher values. Values range 0-10.+
'tbp_lv_color_std_mean', # Color irregularity, calculated as the variance of colors within the lesion's boundary.
'tbp_lv_deltaA', # Average A contrast (inside vs. outside lesion).+
'tbp_lv_deltaB', # Average B contrast (inside vs. outside lesion).+
'tbp_lv_deltaL', # Average L contrast (inside vs. outside lesion).+
'tbp_lv_deltaLB', #
'tbp_lv_deltaLBnorm', # Contrast between the lesion and its immediate surrounding skin. Low contrast lesions tend to be faintly visible such as freckles; high contrast lesions tend to be those with darker pigment. Calculated as the average delta LB of the lesion relative to its immediate background in LAB* color space. Typical values range from 5.5 to 25.+
'tbp_lv_eccentricity', # Eccentricity.+
'tbp_lv_minorAxisMM', # Smallest lesion diameter (mm).+
'tbp_lv_nevi_confidence', # Nevus confidence score (0-100 scale) is a convolutional neural network classifier estimated probability that the lesion is a nevus. The neural network was trained on approximately 57,000 lesions that were classified and labeled by a dermatologist.+,++
'tbp_lv_norm_border', # Border irregularity (0-10 scale); the normalized average of border jaggedness and asymmetry.+
'tbp_lv_norm_color', # Color variation (0-10 scale); the normalized average of color asymmetry and color irregularity.+
'tbp_lv_perimeterMM', # Perimeter of lesion (mm).+
'tbp_lv_radial_color_std_max', # Color asymmetry, a measure of asymmetry of the spatial distribution of color within the lesion. This score is calculated by looking at the average standard deviation in LAB* color space within concentric rings originating from the lesion center. Values range 0-10.+
'tbp_lv_stdL', # Standard deviation of L inside lesion.+
'tbp_lv_stdLExt', # Standard deviation of L outside lesion.+
'tbp_lv_symm_2axis', # Border asymmetry; a measure of asymmetry of the lesion's contour about an axis perpendicular to the lesion's most symmetric axis. Lesions with two axes of symmetry will therefore have low scores (more symmetric), while lesions with only one or zero axes of symmetry will have higher scores (less symmetric). This score is calculated by comparing opposite halves of the lesion contour over many degrees of rotation. The angle where the halves are most similar identifies the principal axis of symmetry, while the second axis of symmetry is perpendicular to the principal axis. Border asymmetry is reported as the asymmetry value about this second axis. Values range 0-10.+
'tbp_lv_symm_2axis_angle', # Lesion border asymmetry angle.+
'tbp_lv_x', # X-coordinate of the lesion on 3D TBP.+
'tbp_lv_y', # Y-coordinate of the lesion on 3D TBP.+
'tbp_lv_z', # Z-coordinate of the lesion on 3D TBP.+
]
new_num_cols = [
'lesion_size_ratio', # tbp_lv_minorAxisMM / clin_size_long_diam_mm
'lesion_shape_index', # tbp_lv_areaMM2 / tbp_lv_perimeterMM **2
'hue_contrast', # tbp_lv_H - tbp_lv_Hext abs
'luminance_contrast', # tbp_lv_L - tbp_lv_Lext abs
'lesion_color_difference', # tbp_lv_deltaA **2 + tbp_lv_deltaB **2 + tbp_lv_deltaL **2 sqrt
'border_complexity', # tbp_lv_norm_border + tbp_lv_symm_2axis
'color_uniformity', # tbp_lv_color_std_mean / tbp_lv_radial_color_std_max
'position_distance_3d', # tbp_lv_x **2 + tbp_lv_y **2 + tbp_lv_z **2 sqrt
'perimeter_to_area_ratio', # tbp_lv_perimeterMM / tbp_lv_areaMM2
'area_to_perimeter_ratio', # tbp_lv_areaMM2 / tbp_lv_perimeterMM
'lesion_visibility_score', # tbp_lv_deltaLBnorm + tbp_lv_norm_color
'symmetry_border_consistency', # tbp_lv_symm_2axis * tbp_lv_norm_border
'consistency_symmetry_border', # tbp_lv_symm_2axis * tbp_lv_norm_border / (tbp_lv_symm_2axis + tbp_lv_norm_border)
'color_consistency', # tbp_lv_stdL / tbp_lv_Lext
'consistency_color', # tbp_lv_stdL*tbp_lv_Lext / tbp_lv_stdL + tbp_lv_Lext
'size_age_interaction', # clin_size_long_diam_mm * age_approx
'hue_color_std_interaction', # tbp_lv_H * tbp_lv_color_std_mean
'lesion_severity_index', # tbp_lv_norm_border + tbp_lv_norm_color + tbp_lv_eccentricity / 3
'shape_complexity_index', # border_complexity + lesion_shape_index
'color_contrast_index', # tbp_lv_deltaA + tbp_lv_deltaB + tbp_lv_deltaL + tbp_lv_deltaLBnorm
'log_lesion_area', # tbp_lv_areaMM2 + 1 np.log
'normalized_lesion_size', # clin_size_long_diam_mm / age_approx
'mean_hue_difference', # tbp_lv_H + tbp_lv_Hext / 2
'std_dev_contrast', # tbp_lv_deltaA **2 + tbp_lv_deltaB **2 + tbp_lv_deltaL **2 / 3 np.sqrt
'color_shape_composite_index', # tbp_lv_color_std_mean + bp_lv_area_perim_ratio + tbp_lv_symm_2axis / 3
'lesion_orientation_3d', # tbp_lv_y , tbp_lv_x np.arctan2
'overall_color_difference', # tbp_lv_deltaA + tbp_lv_deltaB + tbp_lv_deltaL / 3
'symmetry_perimeter_interaction',# tbp_lv_symm_2axis * tbp_lv_perimeterMM
'comprehensive_lesion_index', # tbp_lv_area_perim_ratio + tbp_lv_eccentricity + bp_lv_norm_color + tbp_lv_symm_2axis / 4
'color_variance_ratio', # tbp_lv_color_std_mean / tbp_lv_stdLExt
'border_color_interaction', # tbp_lv_norm_border * tbp_lv_norm_color
'border_color_interaction_2',
'size_color_contrast_ratio', # clin_size_long_diam_mm / tbp_lv_deltaLBnorm
'age_normalized_nevi_confidence',# tbp_lv_nevi_confidence / age_approx
'age_normalized_nevi_confidence_2',
'color_asymmetry_index', # tbp_lv_symm_2axis * tbp_lv_radial_color_std_max
'volume_approximation_3d', # tbp_lv_areaMM2 * sqrt(tbp_lv_x**2 + tbp_lv_y**2 + tbp_lv_z**2)
'color_range', # abs(tbp_lv_L - tbp_lv_Lext) + abs(tbp_lv_A - tbp_lv_Aext) + abs(tbp_lv_B - tbp_lv_Bext)
'shape_color_consistency', # tbp_lv_eccentricity * tbp_lv_color_std_mean
'border_length_ratio', # tbp_lv_perimeterMM / pi * sqrt(tbp_lv_areaMM2 / pi)
'age_size_symmetry_index', # age_approx * clin_size_long_diam_mm * tbp_lv_symm_2axis
'index_age_size_symmetry', # age_approx * tbp_lv_areaMM2 * tbp_lv_symm_2axis
]
cat_cols = ['sex', 'anatom_site_general', 'tbp_tile_type', 'tbp_lv_location', 'tbp_lv_location_simple', 'attribution']
norm_cols = [f'{col}_patient_norm' for col in num_cols + new_num_cols]
special_cols = ['count_per_patient']
image_cols = ["target_3","target_ser02","target_den02","target_effnetv1b0"]
#image_cols = ["target_3","target_effnetv1b0"]
#norm_cols += image_cols
feature_cols = num_cols + new_num_cols + cat_cols + norm_cols + special_cols
def read_data(path):
return (
pl.read_csv(path)
.with_columns(
pl.col('age_approx').cast(pl.String).replace('NA', np.nan).cast(pl.Float64),
)
.with_columns(
pl.col(pl.Float64).fill_nan(pl.col(pl.Float64).median()), # You may want to impute test data with train
)
.with_columns(
lesion_size_ratio = pl.col('tbp_lv_minorAxisMM') / pl.col('clin_size_long_diam_mm'),
lesion_shape_index = pl.col('tbp_lv_areaMM2') / (pl.col('tbp_lv_perimeterMM') ** 2),
hue_contrast = (pl.col('tbp_lv_H') - pl.col('tbp_lv_Hext')).abs(),
luminance_contrast = (pl.col('tbp_lv_L') - pl.col('tbp_lv_Lext')).abs(),
lesion_color_difference = (pl.col('tbp_lv_deltaA') ** 2 + pl.col('tbp_lv_deltaB') ** 2 + pl.col('tbp_lv_deltaL') ** 2).sqrt(),
border_complexity = pl.col('tbp_lv_norm_border') + pl.col('tbp_lv_symm_2axis'),
color_uniformity = pl.col('tbp_lv_color_std_mean') / (pl.col('tbp_lv_radial_color_std_max') + err),
)
.with_columns(
position_distance_3d = (pl.col('tbp_lv_x') ** 2 + pl.col('tbp_lv_y') ** 2 + pl.col('tbp_lv_z') ** 2).sqrt(),
perimeter_to_area_ratio = pl.col('tbp_lv_perimeterMM') / pl.col('tbp_lv_areaMM2'),
area_to_perimeter_ratio = pl.col('tbp_lv_areaMM2') / pl.col('tbp_lv_perimeterMM'),
lesion_visibility_score = pl.col('tbp_lv_deltaLBnorm') + pl.col('tbp_lv_norm_color'),
combined_anatomical_site = pl.col('anatom_site_general') + '_' + pl.col('tbp_lv_location'),
symmetry_border_consistency = pl.col('tbp_lv_symm_2axis') * pl.col('tbp_lv_norm_border'),
consistency_symmetry_border = pl.col('tbp_lv_symm_2axis') * pl.col('tbp_lv_norm_border') / (pl.col('tbp_lv_symm_2axis') + pl.col('tbp_lv_norm_border')),
)
.with_columns(
color_consistency = pl.col('tbp_lv_stdL') / pl.col('tbp_lv_Lext'),
consistency_color = pl.col('tbp_lv_stdL') * pl.col('tbp_lv_Lext') / (pl.col('tbp_lv_stdL') + pl.col('tbp_lv_Lext')),
size_age_interaction = pl.col('clin_size_long_diam_mm') * pl.col('age_approx'),
hue_color_std_interaction = pl.col('tbp_lv_H') * pl.col('tbp_lv_color_std_mean'),
lesion_severity_index = (pl.col('tbp_lv_norm_border') + pl.col('tbp_lv_norm_color') + pl.col('tbp_lv_eccentricity')) / 3,
shape_complexity_index = pl.col('border_complexity') + pl.col('lesion_shape_index'),
color_contrast_index = pl.col('tbp_lv_deltaA') + pl.col('tbp_lv_deltaB') + pl.col('tbp_lv_deltaL') + pl.col('tbp_lv_deltaLBnorm'),
)
.with_columns(
log_lesion_area = (pl.col('tbp_lv_areaMM2') + 1).log(),
normalized_lesion_size = pl.col('clin_size_long_diam_mm') / pl.col('age_approx'),
mean_hue_difference = (pl.col('tbp_lv_H') + pl.col('tbp_lv_Hext')) / 2,
std_dev_contrast = ((pl.col('tbp_lv_deltaA') ** 2 + pl.col('tbp_lv_deltaB') ** 2 + pl.col('tbp_lv_deltaL') ** 2) / 3).sqrt(),
color_shape_composite_index = (pl.col('tbp_lv_color_std_mean') + pl.col('tbp_lv_area_perim_ratio') + pl.col('tbp_lv_symm_2axis')) / 3,
lesion_orientation_3d = pl.arctan2(pl.col('tbp_lv_y'), pl.col('tbp_lv_x')),
overall_color_difference = (pl.col('tbp_lv_deltaA') + pl.col('tbp_lv_deltaB') + pl.col('tbp_lv_deltaL')) / 3,
)
.with_columns(
symmetry_perimeter_interaction = pl.col('tbp_lv_symm_2axis') * pl.col('tbp_lv_perimeterMM'),
comprehensive_lesion_index = (pl.col('tbp_lv_area_perim_ratio') + pl.col('tbp_lv_eccentricity') + pl.col('tbp_lv_norm_color') + pl.col('tbp_lv_symm_2axis')) / 4,
color_variance_ratio = pl.col('tbp_lv_color_std_mean') / pl.col('tbp_lv_stdLExt'),
border_color_interaction = pl.col('tbp_lv_norm_border') * pl.col('tbp_lv_norm_color'),
border_color_interaction_2 = pl.col('tbp_lv_norm_border') * pl.col('tbp_lv_norm_color') / (pl.col('tbp_lv_norm_border') + pl.col('tbp_lv_norm_color')),
size_color_contrast_ratio = pl.col('clin_size_long_diam_mm') / pl.col('tbp_lv_deltaLBnorm'),
age_normalized_nevi_confidence = pl.col('tbp_lv_nevi_confidence') / pl.col('age_approx'),
age_normalized_nevi_confidence_2 = (pl.col('clin_size_long_diam_mm')**2 + pl.col('age_approx')**2).sqrt(),
color_asymmetry_index = pl.col('tbp_lv_radial_color_std_max') * pl.col('tbp_lv_symm_2axis'),
)
.with_columns(
volume_approximation_3d = pl.col('tbp_lv_areaMM2') * (pl.col('tbp_lv_x')**2 + pl.col('tbp_lv_y')**2 + pl.col('tbp_lv_z')**2).sqrt(),
color_range = (pl.col('tbp_lv_L') - pl.col('tbp_lv_Lext')).abs() + (pl.col('tbp_lv_A') - pl.col('tbp_lv_Aext')).abs() + (pl.col('tbp_lv_B') - pl.col('tbp_lv_Bext')).abs(),
shape_color_consistency = pl.col('tbp_lv_eccentricity') * pl.col('tbp_lv_color_std_mean'),
border_length_ratio = pl.col('tbp_lv_perimeterMM') / (2 * np.pi * (pl.col('tbp_lv_areaMM2') / np.pi).sqrt()),
age_size_symmetry_index = pl.col('age_approx') * pl.col('clin_size_long_diam_mm') * pl.col('tbp_lv_symm_2axis'),
index_age_size_symmetry = pl.col('age_approx') * pl.col('tbp_lv_areaMM2') * pl.col('tbp_lv_symm_2axis'),
)
.with_columns(
((pl.col(col) - pl.col(col).mean().over('patient_id')) / (pl.col(col).std().over('patient_id') + err)).alias(f'{col}_patient_norm') for col in (num_cols + new_num_cols)
)
.with_columns(
count_per_patient = pl.col('isic_id').count().over('patient_id'),
)
.with_columns(
pl.col(cat_cols).cast(pl.Categorical),
)
.to_pandas()
.set_index(id_col)
)
def preprocess(df_train, df_test):
global cat_cols
encoder = OneHotEncoder(sparse_output=False, dtype=np.int32, handle_unknown='ignore')
encoder.fit(df_train[cat_cols])
new_cat_cols = [f'onehot_{i}' for i in range(len(encoder.get_feature_names_out()))]
df_train[new_cat_cols] = encoder.transform(df_train[cat_cols])
df_train[new_cat_cols] = df_train[new_cat_cols].astype('category')
df_test[new_cat_cols] = encoder.transform(df_test[cat_cols])
df_test[new_cat_cols] = df_test[new_cat_cols].astype('category')
# effnetv1b0
df_eff = pd.read_csv("/kaggle/input/isic-inference-effnetv1b0-for-training-data/train_effnetv1b0.csv")
df_train = df_train.reset_index(drop=True)
df_eff = df_eff.reset_index(drop=True)
#upload effnetv1b0 train predictions values
df_train["target_effnetv1b0"] = df_eff["target_effnetv1b0"]
df_eff = pd.read_csv("submission_effnetv1b0.csv")
df_test = df_test.reset_index(drop=True)
df_eff = df_eff.reset_index(drop=True)
#upload effnetv1b0 test predictions values
df_test["target_effnetv1b0"] = df_eff["target"]
# effnetv1b0
df_den = pd.read_csv("/kaggle/input/dense-train-cat/train_eva02.csv")
df_train = df_train.reset_index(drop=True)
df_den = df_den.reset_index(drop=True)
#upload effnetv1b0 train predictions values
df_train["target_den02"] = df_den["target_eva02"]
df_den = pd.read_csv("submission_den.csv")
df_test = df_test.reset_index(drop=True)
df_den = df_den.reset_index(drop=True)
#upload effnetv1b0 test predictions values
df_test["target_den02"] = df_den["target"]
# target 3
df_image_3 = pd.read_csv("/kaggle/input/isic-2024-pl-submission-script-and-preds/train_preds.csv")
df_train = df_train.reset_index(drop=True)
df_image_3 = df_image_3.reset_index(drop=True)
df_train["target_3"] = df_image_3["pred"]
df_3 = pd.read_csv("submission_image3.csv")
df_test = df_test.reset_index(drop=True)
df_image_3 = df_image_3.reset_index(drop=True)
df_test["target_3"] = df_3["target"]
#eva02
df_ser = pd.read_csv("/kaggle/input/ser-train/train_eva02.csv")
df_train = df_train.reset_index(drop=True)
df_ser = df_ser.reset_index(drop=True)
df_ser = df_ser[["target_eva02"]]
df_train["target_ser02"] = df_ser["target_eva02"]
df_ser = pd.read_csv("submission_ser.csv")
df_test = df_test.reset_index(drop=True)
df_ser = df_ser.reset_index(drop=True)
df_test["target_ser02"] = df_ser["target"]
for col in cat_cols:
feature_cols.remove(col)
feature_cols.extend(new_cat_cols)
cat_cols = new_cat_cols
return df_train, df_test
def custom_metric(estimator, X, y_true):
y_hat = estimator.predict_proba(X)[:, 1]
min_tpr = 0.80
max_fpr = abs(1 - min_tpr)
v_gt = abs(y_true - 1)
v_pred = np.array([1.0 - x for x in y_hat])
partial_auc_scaled = roc_auc_score(v_gt, v_pred, max_fpr=max_fpr)
partial_auc = 0.5 * max_fpr**2 + (max_fpr - 0.5 * max_fpr**2) / (1.0 - 0.5) * (partial_auc_scaled - 0.5)
return partial_auc
df_train = read_data(train_path)
df_test = read_data(test_path)
df_subm = pd.read_csv(subm_path, index_col=id_col)
df_train, df_test = preprocess(df_train, df_test)
least_important_features = ['onehot_32', 'onehot_6', 'onehot_33', 'onehot_30', 'onehot_26', 'onehot_22', 'onehot_36', 'onehot_4']
#they are detected after the least_important_features are removed and it has increased cv score also so I add it
#least_important_features_2 = ['onehot_17', 'onehot_42', 'onehot_29', 'onehot_13', 'onehot_25']
#least_important_features += least_important_features_2
df_train.drop(columns =least_important_features,inplace = True)
for feature in least_important_features:
cat_cols.remove(feature)
feature_cols.remove(feature)
from sklearn.base import BaseEstimator, TransformerMixin
import copy
feature_cols_without_image_cols = copy.copy(feature_cols)
feature_cols += image_cols
class SelectColumns(BaseEstimator, TransformerMixin):
def __init__(self, columns):
self.columns = columns
def fit(self, X, y=None):
return self
def transform(self, X):
return X[self.columns]
lgb_params = {
'objective': 'binary',
'verbosity': -1,
'n_iter': 250,
'boosting_type': 'gbdt',
'random_state': seed,
'lambda_l1': 0.08758718919397321,
'lambda_l2': 0.0039689175176025465,
'learning_rate': 0.03231007103195577,
'max_depth': 4,
'num_leaves': 103,
'colsample_bytree': 0.8329551585827726,
'colsample_bynode': 0.4025961355653304,
'bagging_fraction': 0.7738954452473223,
'bagging_freq': 4,
'min_data_in_leaf': 85,
'scale_pos_weight': 2.7984184778875543,
}
sampling_ratio = 0.01
seed =42
lgb_model = Pipeline([
('sampler_1', RandomOverSampler(sampling_strategy= 0.003 , random_state=seed)),
('sampler_2', RandomUnderSampler(sampling_strategy=sampling_ratio, random_state=seed)),
('filter', SelectColumns(feature_cols_without_image_cols)),
('classifier', lgb.LGBMClassifier(**lgb_params)),
])
cb_params = {
'loss_function': 'Logloss',
'iterations': 250,
'verbose': False,
'random_state': seed,
'max_depth': 7,
'learning_rate': 0.06936242010150652,
'scale_pos_weight': 2.6149345838209532,
'l2_leaf_reg': 6.216113851699493,
'subsample': 0.6249261779711819,
'min_data_in_leaf': 24,
'cat_features': cat_cols,
}
cb_model = Pipeline([
('sampler_1', RandomOverSampler(sampling_strategy= 0.003 , random_state=seed)),
('sampler_2', RandomUnderSampler(sampling_strategy=sampling_ratio, random_state=seed)),
('classifier', cb.CatBoostClassifier(**cb_params)),
])
xgb_params = {
'enable_categorical': True,
'tree_method': 'hist',
'random_state': seed,
'learning_rate': 0.08501257473292347,
'lambda': 8.879624125465703,
'alpha': 0.6779926606782505,
'max_depth': 6,
'subsample': 0.6012681388711075,
'colsample_bytree': 0.8437772277074493,
'colsample_bylevel': 0.5476090898823716,
'colsample_bynode': 0.9928601203635129,
'scale_pos_weight': 3.29440313334688,
}
xgb_model = Pipeline([
('sampler_1', RandomOverSampler(sampling_strategy= 0.003 , random_state=seed)),
('sampler_2', RandomUnderSampler(sampling_strategy=sampling_ratio, random_state=seed)),
('classifier', xgb.XGBClassifier(**xgb_params)),
])
estimator = VotingClassifier([
('lgb', lgb_model), ('cb', cb_model), ('xgb', xgb_model),
], voting='soft')
X = df_train[feature_cols]
y = df_train[target_col]
groups = df_train[group_col]
cv = StratifiedGroupKFold(5, shuffle=True, random_state=seed)
val_score = cross_val_score(
estimator=estimator,
X=X, y=y,
cv=cv,
groups=groups,
scoring=custom_metric,
)
np.mean(val_score), val_score
X, y = df_train[feature_cols], df_train[target_col]
estimator.fit(X, y)
df_subm['target'] = estimator.predict_proba(df_test[feature_cols])[:, 1]
df_subm.to_csv('submission_0.csv')第三个notebook
root = Path('/kaggle/input/isic-2024-challenge')
train_path = root / 'train-metadata.csv'
test_path = root / 'test-metadata.csv'
subm_path = root / 'sample_submission.csv'
id_col = 'isic_id'
target_col = 'target'
group_col = 'patient_id'
err = 1e-5
sampling_ratio = 0.01
seed = 42
num_cols = [
'age_approx', # Approximate age of patient at time of imaging.
'clin_size_long_diam_mm', # Maximum diameter of the lesion (mm).+
'tbp_lv_A', # A inside lesion.+
'tbp_lv_Aext', # A outside lesion.+
'tbp_lv_B', # B inside lesion.+
'tbp_lv_Bext', # B outside lesion.+
'tbp_lv_C', # Chroma inside lesion.+
'tbp_lv_Cext', # Chroma outside lesion.+
'tbp_lv_H', # Hue inside the lesion; calculated as the angle of A* and B* in LAB* color space. Typical values range from 25 (red) to 75 (brown).+
'tbp_lv_Hext', # Hue outside lesion.+
'tbp_lv_L', # L inside lesion.+
'tbp_lv_Lext', # L outside lesion.+
'tbp_lv_areaMM2', # Area of lesion (mm^2).+
'tbp_lv_area_perim_ratio', # Border jaggedness, the ratio between lesions perimeter and area. Circular lesions will have low values; irregular shaped lesions will have higher values. Values range 0-10.+
'tbp_lv_color_std_mean', # Color irregularity, calculated as the variance of colors within the lesion's boundary.
'tbp_lv_deltaA', # Average A contrast (inside vs. outside lesion).+
'tbp_lv_deltaB', # Average B contrast (inside vs. outside lesion).+
'tbp_lv_deltaL', # Average L contrast (inside vs. outside lesion).+
'tbp_lv_deltaLB', #
'tbp_lv_deltaLBnorm', # Contrast between the lesion and its immediate surrounding skin. Low contrast lesions tend to be faintly visible such as freckles; high contrast lesions tend to be those with darker pigment. Calculated as the average delta LB of the lesion relative to its immediate background in LAB* color space. Typical values range from 5.5 to 25.+
'tbp_lv_eccentricity', # Eccentricity.+
'tbp_lv_minorAxisMM', # Smallest lesion diameter (mm).+
'tbp_lv_nevi_confidence', # Nevus confidence score (0-100 scale) is a convolutional neural network classifier estimated probability that the lesion is a nevus. The neural network was trained on approximately 57,000 lesions that were classified and labeled by a dermatologist.+,++
'tbp_lv_norm_border', # Border irregularity (0-10 scale); the normalized average of border jaggedness and asymmetry.+
'tbp_lv_norm_color', # Color variation (0-10 scale); the normalized average of color asymmetry and color irregularity.+
'tbp_lv_perimeterMM', # Perimeter of lesion (mm).+
'tbp_lv_radial_color_std_max', # Color asymmetry, a measure of asymmetry of the spatial distribution of color within the lesion. This score is calculated by looking at the average standard deviation in LAB* color space within concentric rings originating from the lesion center. Values range 0-10.+
'tbp_lv_stdL', # Standard deviation of L inside lesion.+
'tbp_lv_stdLExt', # Standard deviation of L outside lesion.+
'tbp_lv_symm_2axis', # Border asymmetry; a measure of asymmetry of the lesion's contour about an axis perpendicular to the lesion's most symmetric axis. Lesions with two axes of symmetry will therefore have low scores (more symmetric), while lesions with only one or zero axes of symmetry will have higher scores (less symmetric). This score is calculated by comparing opposite halves of the lesion contour over many degrees of rotation. The angle where the halves are most similar identifies the principal axis of symmetry, while the second axis of symmetry is perpendicular to the principal axis. Border asymmetry is reported as the asymmetry value about this second axis. Values range 0-10.+
'tbp_lv_symm_2axis_angle', # Lesion border asymmetry angle.+
'tbp_lv_x', # X-coordinate of the lesion on 3D TBP.+
'tbp_lv_y', # Y-coordinate of the lesion on 3D TBP.+
'tbp_lv_z', # Z-coordinate of the lesion on 3D TBP.+
]
new_num_cols = [
'lesion_size_ratio', # tbp_lv_minorAxisMM / clin_size_long_diam_mm
'lesion_shape_index', # tbp_lv_areaMM2 / tbp_lv_perimeterMM **2
'hue_contrast', # tbp_lv_H - tbp_lv_Hext abs
'luminance_contrast', # tbp_lv_L - tbp_lv_Lext abs
'lesion_color_difference', # tbp_lv_deltaA **2 + tbp_lv_deltaB **2 + tbp_lv_deltaL **2 sqrt
'border_complexity', # tbp_lv_norm_border + tbp_lv_symm_2axis
'color_uniformity', # tbp_lv_color_std_mean / tbp_lv_radial_color_std_max
'position_distance_3d', # tbp_lv_x **2 + tbp_lv_y **2 + tbp_lv_z **2 sqrt
'perimeter_to_area_ratio', # tbp_lv_perimeterMM / tbp_lv_areaMM2
'area_to_perimeter_ratio', # tbp_lv_areaMM2 / tbp_lv_perimeterMM
'lesion_visibility_score', # tbp_lv_deltaLBnorm + tbp_lv_norm_color
'symmetry_border_consistency', # tbp_lv_symm_2axis * tbp_lv_norm_border
'consistency_symmetry_border', # tbp_lv_symm_2axis * tbp_lv_norm_border / (tbp_lv_symm_2axis + tbp_lv_norm_border)
'color_consistency', # tbp_lv_stdL / tbp_lv_Lext
'consistency_color', # tbp_lv_stdL*tbp_lv_Lext / tbp_lv_stdL + tbp_lv_Lext
'size_age_interaction', # clin_size_long_diam_mm * age_approx
'hue_color_std_interaction', # tbp_lv_H * tbp_lv_color_std_mean
'lesion_severity_index', # tbp_lv_norm_border + tbp_lv_norm_color + tbp_lv_eccentricity / 3
'shape_complexity_index', # border_complexity + lesion_shape_index
'color_contrast_index', # tbp_lv_deltaA + tbp_lv_deltaB + tbp_lv_deltaL + tbp_lv_deltaLBnorm
'log_lesion_area', # tbp_lv_areaMM2 + 1 np.log
'normalized_lesion_size', # clin_size_long_diam_mm / age_approx
'mean_hue_difference', # tbp_lv_H + tbp_lv_Hext / 2
'std_dev_contrast', # tbp_lv_deltaA **2 + tbp_lv_deltaB **2 + tbp_lv_deltaL **2 / 3 np.sqrt
'color_shape_composite_index', # tbp_lv_color_std_mean + bp_lv_area_perim_ratio + tbp_lv_symm_2axis / 3
'lesion_orientation_3d', # tbp_lv_y , tbp_lv_x np.arctan2
'overall_color_difference', # tbp_lv_deltaA + tbp_lv_deltaB + tbp_lv_deltaL / 3
'symmetry_perimeter_interaction',# tbp_lv_symm_2axis * tbp_lv_perimeterMM
'comprehensive_lesion_index', # tbp_lv_area_perim_ratio + tbp_lv_eccentricity + bp_lv_norm_color + tbp_lv_symm_2axis / 4
'color_variance_ratio', # tbp_lv_color_std_mean / tbp_lv_stdLExt
'border_color_interaction', # tbp_lv_norm_border * tbp_lv_norm_color
'border_color_interaction_2',
'size_color_contrast_ratio', # clin_size_long_diam_mm / tbp_lv_deltaLBnorm
'age_normalized_nevi_confidence',# tbp_lv_nevi_confidence / age_approx
'age_normalized_nevi_confidence_2',
'color_asymmetry_index', # tbp_lv_symm_2axis * tbp_lv_radial_color_std_max
'volume_approximation_3d', # tbp_lv_areaMM2 * sqrt(tbp_lv_x**2 + tbp_lv_y**2 + tbp_lv_z**2)
'color_range', # abs(tbp_lv_L - tbp_lv_Lext) + abs(tbp_lv_A - tbp_lv_Aext) + abs(tbp_lv_B - tbp_lv_Bext)
'shape_color_consistency', # tbp_lv_eccentricity * tbp_lv_color_std_mean
'border_length_ratio', # tbp_lv_perimeterMM / pi * sqrt(tbp_lv_areaMM2 / pi)
'age_size_symmetry_index', # age_approx * clin_size_long_diam_mm * tbp_lv_symm_2axis
'index_age_size_symmetry', # age_approx * tbp_lv_areaMM2 * tbp_lv_symm_2axis
]
cat_cols = ['sex', 'anatom_site_general', 'tbp_tile_type', 'tbp_lv_location', 'tbp_lv_location_simple', 'attribution']
norm_cols = [f'{col}_patient_norm' for col in num_cols + new_num_cols]
special_cols = ['count_per_patient']
image_cols = ["target_3","target_ser02","target_fb02","target_effnetv1b0"]
#image_cols = ["target_3","target_effnetv1b0"]
#norm_cols += image_cols
feature_cols = num_cols + new_num_cols + cat_cols + norm_cols + special_cols
def read_data(path):
return (
pl.read_csv(path)
.with_columns(
pl.col('age_approx').cast(pl.String).replace('NA', np.nan).cast(pl.Float64),
)
.with_columns(
pl.col(pl.Float64).fill_nan(pl.col(pl.Float64).median()), # You may want to impute test data with train
)
.with_columns(
lesion_size_ratio = pl.col('tbp_lv_minorAxisMM') / pl.col('clin_size_long_diam_mm'),
lesion_shape_index = pl.col('tbp_lv_areaMM2') / (pl.col('tbp_lv_perimeterMM') ** 2),
hue_contrast = (pl.col('tbp_lv_H') - pl.col('tbp_lv_Hext')).abs(),
luminance_contrast = (pl.col('tbp_lv_L') - pl.col('tbp_lv_Lext')).abs(),
lesion_color_difference = (pl.col('tbp_lv_deltaA') ** 2 + pl.col('tbp_lv_deltaB') ** 2 + pl.col('tbp_lv_deltaL') ** 2).sqrt(),
border_complexity = pl.col('tbp_lv_norm_border') + pl.col('tbp_lv_symm_2axis'),
color_uniformity = pl.col('tbp_lv_color_std_mean') / (pl.col('tbp_lv_radial_color_std_max') + err),
)
.with_columns(
position_distance_3d = (pl.col('tbp_lv_x') ** 2 + pl.col('tbp_lv_y') ** 2 + pl.col('tbp_lv_z') ** 2).sqrt(),
perimeter_to_area_ratio = pl.col('tbp_lv_perimeterMM') / pl.col('tbp_lv_areaMM2'),
area_to_perimeter_ratio = pl.col('tbp_lv_areaMM2') / pl.col('tbp_lv_perimeterMM'),
lesion_visibility_score = pl.col('tbp_lv_deltaLBnorm') + pl.col('tbp_lv_norm_color'),
combined_anatomical_site = pl.col('anatom_site_general') + '_' + pl.col('tbp_lv_location'),
symmetry_border_consistency = pl.col('tbp_lv_symm_2axis') * pl.col('tbp_lv_norm_border'),
consistency_symmetry_border = pl.col('tbp_lv_symm_2axis') * pl.col('tbp_lv_norm_border') / (pl.col('tbp_lv_symm_2axis') + pl.col('tbp_lv_norm_border')),
)
.with_columns(
color_consistency = pl.col('tbp_lv_stdL') / pl.col('tbp_lv_Lext'),
consistency_color = pl.col('tbp_lv_stdL') * pl.col('tbp_lv_Lext') / (pl.col('tbp_lv_stdL') + pl.col('tbp_lv_Lext')),
size_age_interaction = pl.col('clin_size_long_diam_mm') * pl.col('age_approx'),
hue_color_std_interaction = pl.col('tbp_lv_H') * pl.col('tbp_lv_color_std_mean'),
lesion_severity_index = (pl.col('tbp_lv_norm_border') + pl.col('tbp_lv_norm_color') + pl.col('tbp_lv_eccentricity')) / 3,
shape_complexity_index = pl.col('border_complexity') + pl.col('lesion_shape_index'),
color_contrast_index = pl.col('tbp_lv_deltaA') + pl.col('tbp_lv_deltaB') + pl.col('tbp_lv_deltaL') + pl.col('tbp_lv_deltaLBnorm'),
)
.with_columns(
log_lesion_area = (pl.col('tbp_lv_areaMM2') + 1).log(),
normalized_lesion_size = pl.col('clin_size_long_diam_mm') / pl.col('age_approx'),
mean_hue_difference = (pl.col('tbp_lv_H') + pl.col('tbp_lv_Hext')) / 2,
std_dev_contrast = ((pl.col('tbp_lv_deltaA') ** 2 + pl.col('tbp_lv_deltaB') ** 2 + pl.col('tbp_lv_deltaL') ** 2) / 3).sqrt(),
color_shape_composite_index = (pl.col('tbp_lv_color_std_mean') + pl.col('tbp_lv_area_perim_ratio') + pl.col('tbp_lv_symm_2axis')) / 3,
lesion_orientation_3d = pl.arctan2(pl.col('tbp_lv_y'), pl.col('tbp_lv_x')),
overall_color_difference = (pl.col('tbp_lv_deltaA') + pl.col('tbp_lv_deltaB') + pl.col('tbp_lv_deltaL')) / 3,
)
.with_columns(
symmetry_perimeter_interaction = pl.col('tbp_lv_symm_2axis') * pl.col('tbp_lv_perimeterMM'),
comprehensive_lesion_index = (pl.col('tbp_lv_area_perim_ratio') + pl.col('tbp_lv_eccentricity') + pl.col('tbp_lv_norm_color') + pl.col('tbp_lv_symm_2axis')) / 4,
color_variance_ratio = pl.col('tbp_lv_color_std_mean') / pl.col('tbp_lv_stdLExt'),
border_color_interaction = pl.col('tbp_lv_norm_border') * pl.col('tbp_lv_norm_color'),
border_color_interaction_2 = pl.col('tbp_lv_norm_border') * pl.col('tbp_lv_norm_color') / (pl.col('tbp_lv_norm_border') + pl.col('tbp_lv_norm_color')),
size_color_contrast_ratio = pl.col('clin_size_long_diam_mm') / pl.col('tbp_lv_deltaLBnorm'),
age_normalized_nevi_confidence = pl.col('tbp_lv_nevi_confidence') / pl.col('age_approx'),
age_normalized_nevi_confidence_2 = (pl.col('clin_size_long_diam_mm')**2 + pl.col('age_approx')**2).sqrt(),
color_asymmetry_index = pl.col('tbp_lv_radial_color_std_max') * pl.col('tbp_lv_symm_2axis'),
)
.with_columns(
volume_approximation_3d = pl.col('tbp_lv_areaMM2') * (pl.col('tbp_lv_x')**2 + pl.col('tbp_lv_y')**2 + pl.col('tbp_lv_z')**2).sqrt(),
color_range = (pl.col('tbp_lv_L') - pl.col('tbp_lv_Lext')).abs() + (pl.col('tbp_lv_A') - pl.col('tbp_lv_Aext')).abs() + (pl.col('tbp_lv_B') - pl.col('tbp_lv_Bext')).abs(),
shape_color_consistency = pl.col('tbp_lv_eccentricity') * pl.col('tbp_lv_color_std_mean'),
border_length_ratio = pl.col('tbp_lv_perimeterMM') / (2 * np.pi * (pl.col('tbp_lv_areaMM2') / np.pi).sqrt()),
age_size_symmetry_index = pl.col('age_approx') * pl.col('clin_size_long_diam_mm') * pl.col('tbp_lv_symm_2axis'),
index_age_size_symmetry = pl.col('age_approx') * pl.col('tbp_lv_areaMM2') * pl.col('tbp_lv_symm_2axis'),
)
.with_columns(
((pl.col(col) - pl.col(col).mean().over('patient_id')) / (pl.col(col).std().over('patient_id') + err)).alias(f'{col}_patient_norm') for col in (num_cols + new_num_cols)
)
.with_columns(
count_per_patient = pl.col('isic_id').count().over('patient_id'),
)
.with_columns(
pl.col(cat_cols).cast(pl.Categorical),
)
.to_pandas()
.set_index(id_col)
)
def preprocess(df_train, df_test):
global cat_cols
encoder = OneHotEncoder(sparse_output=False, dtype=np.int32, handle_unknown='ignore')
encoder.fit(df_train[cat_cols])
new_cat_cols = [f'onehot_{i}' for i in range(len(encoder.get_feature_names_out()))]
df_train[new_cat_cols] = encoder.transform(df_train[cat_cols])
df_train[new_cat_cols] = df_train[new_cat_cols].astype('category')
df_test[new_cat_cols] = encoder.transform(df_test[cat_cols])
df_test[new_cat_cols] = df_test[new_cat_cols].astype('category')
# effnetv1b0
df_eff = pd.read_csv("/kaggle/input/isic-inference-effnetv1b0-for-training-data/train_effnetv1b0.csv")
df_train = df_train.reset_index(drop=True)
df_eff = df_eff.reset_index(drop=True)
#upload effnetv1b0 train predictions values
df_train["target_effnetv1b0"] = df_eff["target_effnetv1b0"]
df_eff = pd.read_csv("submission_effnetv1b0.csv")
df_test = df_test.reset_index(drop=True)
df_eff = df_eff.reset_index(drop=True)
#upload effnetv1b0 test predictions values
df_test["target_effnetv1b0"] = df_eff["target"]
# target 3
df_image_3 = pd.read_csv("/kaggle/input/isic-2024-pl-submission-script-and-preds/train_preds.csv")
df_train = df_train.reset_index(drop=True)
df_image_3 = df_image_3.reset_index(drop=True)
df_train["target_3"] = df_image_3["pred"]
df_3 = pd.read_csv("submission_image3.csv")
df_test = df_test.reset_index(drop=True)
df_image_3 = df_image_3.reset_index(drop=True)
df_test["target_3"] = df_3["target"]
#eva02
df_ser = pd.read_csv("/kaggle/input/ser-train/train_eva02.csv")
df_train = df_train.reset_index(drop=True)
df_ser = df_ser.reset_index(drop=True)
df_ser = df_ser[["target_eva02"]]
df_train["target_ser02"] = df_ser["target_eva02"]
df_ser = pd.read_csv("submission_ser.csv")
df_test = df_test.reset_index(drop=True)
df_ser = df_ser.reset_index(drop=True)
df_test["target_ser02"] = df_ser["target"]
#eva02
df_fb = pd.read_csv("/kaggle/input/fb-train/train_eva02.csv")
df_train = df_train.reset_index(drop=True)
df_fb = df_fb.reset_index(drop=True)
df_fb = df_fb[["target_eva02"]]
df_train["target_fb02"] = df_fb["target_eva02"]
df_fb = pd.read_csv("submission_fb.csv")
df_test = df_test.reset_index(drop=True)
df_fb = df_fb.reset_index(drop=True)
df_test["target_fb02"] = df_ser["target"]
for col in cat_cols:
feature_cols.remove(col)
feature_cols.extend(new_cat_cols)
cat_cols = new_cat_cols
return df_train, df_test
def custom_metric(estimator, X, y_true):
y_hat = estimator.predict_proba(X)[:, 1]
min_tpr = 0.80
max_fpr = abs(1 - min_tpr)
v_gt = abs(y_true - 1)
v_pred = np.array([1.0 - x for x in y_hat])
partial_auc_scaled = roc_auc_score(v_gt, v_pred, max_fpr=max_fpr)
partial_auc = 0.5 * max_fpr**2 + (max_fpr - 0.5 * max_fpr**2) / (1.0 - 0.5) * (partial_auc_scaled - 0.5)
return partial_auc
df_train = read_data(train_path)
df_test = read_data(test_path)
df_subm = pd.read_csv(subm_path, index_col=id_col)
df_train, df_test = preprocess(df_train, df_test)
least_important_features = ['onehot_32', 'onehot_6', 'onehot_33', 'onehot_30', 'onehot_26', 'onehot_22', 'onehot_36', 'onehot_4']
#they are detected after the least_important_features are removed and it has increased cv score also so I add it
#least_important_features_2 = ['onehot_17', 'onehot_42', 'onehot_29', 'onehot_13', 'onehot_25']
#least_important_features += least_important_features_2
df_train.drop(columns =least_important_features,inplace = True)
for feature in least_important_features:
cat_cols.remove(feature)
feature_cols.remove(feature)
from sklearn.base import BaseEstimator, TransformerMixin
import copy
feature_cols_without_image_cols = copy.copy(feature_cols)
feature_cols += image_cols
class SelectColumns(BaseEstimator, TransformerMixin):
def __init__(self, columns):
self.columns = columns
def fit(self, X, y=None):
return self
def transform(self, X):
return X[self.columns]
lgb_params = {
'objective': 'binary',
'verbosity': -1,
'n_iter': 250,
'boosting_type': 'gbdt',
'random_state': seed,
'lambda_l1': 0.08758718919397321,
'lambda_l2': 0.0039689175176025465,
'learning_rate': 0.03231007103195577,
'max_depth': 4,
'num_leaves': 103,
'colsample_bytree': 0.8329551585827726,
'colsample_bynode': 0.4025961355653304,
'bagging_fraction': 0.7738954452473223,
'bagging_freq': 4,
'min_data_in_leaf': 85,
'scale_pos_weight': 2.7984184778875543,
}
sampling_ratio = 0.01
seed =42
lgb_model = Pipeline([
('sampler_1', RandomOverSampler(sampling_strategy= 0.003 , random_state=seed)),
('sampler_2', RandomUnderSampler(sampling_strategy=sampling_ratio, random_state=seed)),
('filter', SelectColumns(feature_cols_without_image_cols)),
('classifier', lgb.LGBMClassifier(**lgb_params)),
])
cb_params = {
'loss_function': 'Logloss',
'iterations': 250,
'verbose': False,
'random_state': seed,
'max_depth': 7,
'learning_rate': 0.06936242010150652,
'scale_pos_weight': 2.6149345838209532,
'l2_leaf_reg': 6.216113851699493,
'subsample': 0.6249261779711819,
'min_data_in_leaf': 24,
'cat_features': cat_cols,
}
cb_model = Pipeline([
('sampler_1', RandomOverSampler(sampling_strategy= 0.003 , random_state=seed)),
('sampler_2', RandomUnderSampler(sampling_strategy=sampling_ratio, random_state=seed)),
('classifier', cb.CatBoostClassifier(**cb_params)),
])
xgb_params = {
'enable_categorical': True,
'tree_method': 'hist',
'random_state': seed,
'learning_rate': 0.08501257473292347,
'lambda': 8.879624125465703,
'alpha': 0.6779926606782505,
'max_depth': 6,
'subsample': 0.6012681388711075,
'colsample_bytree': 0.8437772277074493,
'colsample_bylevel': 0.5476090898823716,
'colsample_bynode': 0.9928601203635129,
'scale_pos_weight': 3.29440313334688,
}
xgb_model = Pipeline([
('sampler_1', RandomOverSampler(sampling_strategy= 0.003 , random_state=seed)),
('sampler_2', RandomUnderSampler(sampling_strategy=sampling_ratio, random_state=seed)),
('classifier', xgb.XGBClassifier(**xgb_params)),
])
estimator = VotingClassifier([
('lgb', lgb_model), ('cb', cb_model), ('xgb', xgb_model),
], voting='soft',weights=[0.2,0.4,0.4])
X = df_train[feature_cols]
y = df_train[target_col]
groups = df_train[group_col]
cv = StratifiedGroupKFold(5, shuffle=True, random_state=seed)
val_score = cross_val_score(
estimator=estimator,
X=X, y=y,
cv=cv,
groups=groups,
scoring=custom_metric,
)
np.mean(val_score), val_score
X, y = df_train[feature_cols], df_train[target_col]
estimator.fit(X, y)
df_subm['target'] = estimator.predict_proba(df_test[feature_cols])[:, 1]
df_subm.to_csv('submission_1.csv')进行合成
id_col = 'isic_id'
su0=pd.read_csv('submission_0.csv',index_col=id_col)
su1=pd.read_csv('submission_1.csv',index_col=id_col)
su2=pd.read_csv('submission_2.csv',index_col=id_col)
df_subm = pd.read_csv("/kaggle/input/isic-2024-challenge/sample_submission.csv", index_col=id_col)
df_subm['target']=0.05*su0['target']+0.05*su1['target']+0.9*su2['target']
df_subm.to_csv('submission.csv')
df_subm.head()需要的额外数据，已经公开
扫一扫
733
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
