MoA Inference 노트북 작성
Setting
Install and import libraries
# TabNet
!pip install --no-index --find-links /kaggle/input/pytorchtabnet/pytorch_tabnet-2.0.0-py3-none-any.whl pytorch-tabnet > /dev/null
# Iterative Stratification
!pip install /kaggle/input/iterative-stratification/iterative-stratification-master/ > /dev/nullimport os
import random
from shutil import copytree, ignore_patterns
import numpy as np
import pandas as pd
import pickle
import torch
import torch.optim as optim
import torch.nn as nn
import torch.nn.functional as F
from torch.nn.modules.loss import _WeightedLoss
from torch.utils.data import Dataset, DataLoader
from torch.optim.lr_scheduler import ReduceLROnPlateau, StepLR
from sklearn.metrics import roc_auc_score
from pytorch_tabnet.tab_network import TabNet
from pytorch_tabnet.metrics import Metric
from scipy.stats import kurtosis
from sklearn.svm import LinearSVC
from sklearn.feature_selection import SelectFromModel, VarianceThreshold, SelectKBest
from sklearn.preprocessing import StandardScaler, RobustScaler, QuantileTransformer
from sklearn.decomposition import PCA, NMF
from iterstrat.ml_stratifiers import MultilabelStratifiedKFold
import warnings
warnings.filterwarnings('ignore')# Util Functions
class Dict2Obj(object):
"""
Turns a dictionary into a class
"""
#----------------------------------------------------------------------
def __init__(self, dictionary):
"""Constructor"""
for key in dictionary:
setattr(self, key, dictionary[key])
class DatasetWithoutLabel(Dataset):
def __init__(self, X):
super().__init__()
self._X = X
def __getitem__(self, item):
return (self._X[item], )
def __len__(self):
return len(self._X)
def collate_fn_test(batch):
X = [x[0] for x in batch]
X = torch.FloatTensor(X)
return Xdef seed_everything(seed):
random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.backends.cudnn.benchmark = True
torch.backends.cudnn.enabled = True전처리 코드
cate2num: category feature → numerical feature make_folds: iterative stratified K Fold for multi-label classification task feature encoding: Rank Gauss, Feature statistics, PCA, Varaiance Threshold, LSVM feature selection
Variance Threshold의 경우 데이터가 달라질 경우 feature selection이 달라지기때문에 weight를 불러 inference를 하는 경우 private testset에서 에러가 날 수 있음. 이러한 문제를 해결하기위해 mask를 따로 만들어 전처리함.
def cate2num(df):
df['cp_time'] = df['cp_time'].map({24: 0, 48: 1, 72: 2})
df['cp_dose'] = df['cp_dose'].map({'D1': 0, 'D2': 1})
return df
def get_cols(test):
g_cols = [c for c in test.columns if 'g-' in c]
c_cols = [c for c in test.columns if 'c-' in c]
num_g = len(g_cols)
num_c = len(c_cols)
return g_cols, c_cols, num_g, num_c
def make_folds(train_features, train_targets_scored, n_folds, seed):
train = train_features.merge(train_targets_scored, on='sig_id')
target_cols = [c for c in train_targets_scored.columns if c not in ['sig_id']]
cols = target_cols + ['cp_type']
train_cp = train[train['cp_type'] != 'ctl_vehicle'].reset_index(drop=True)
mskf = MultilabelStratifiedKFold(n_splits=n_folds, shuffle=True, random_state=seed)
train_cp.loc[:, 'fold'] = 0
for n, (train_index, val_index) in enumerate(mskf.split(train_cp, train_cp[target_cols])):
train_cp.loc[val_index, 'fold'] = int(n)
train_cp['fold'] = train_cp['fold'].astype(int)
train_ctl = train[train['cp_type'] == 'ctl_vehicle'].reset_index(drop=True)
train_ctl.loc[:, 'fold'] = 100
train = pd.concat([train_cp, train_ctl])
train_features_with_fold = train_features.merge(train[['sig_id', 'fold']], on='sig_id')
return train_features_with_fold
# <https://www.kaggle.com/vbmokin/moa-pytorch-rankgauss-pca-nn-upgrade-3d-visual?scriptVersionId=46264255>
def create_rankgauss(train, test, cols, QT_n_quantile_min, QT_n_quantile_max, seed):
for col in (cols):
kurt = max(kurtosis(train[col]), kurtosis(test[col]))
QuantileTransformer_n_quantiles = n_quantile_for_kurt(kurt, calc_QT_par_kurt(QT_n_quantile_min,
QT_n_quantile_max))
transformer = QuantileTransformer(n_quantiles=QuantileTransformer_n_quantiles, random_state=seed,
output_distribution="normal")
vec_len = len(train[col].values)
vec_len_test = len(test[col].values)
raw_vec = train[col].values.reshape(vec_len, 1)
transformer.fit(raw_vec)
train[col] = transformer.transform(raw_vec).reshape(1, vec_len)[0]
test[col] = \\
transformer.transform(test[col].values.reshape(vec_len_test, 1)).reshape(1, vec_len_test)[0]
return train, test
def calc_QT_par_kurt(QT_n_quantile_min=10, QT_n_quantile_max=200):
# Calculation parameters of function: n_quantile(kurtosis) = k1*kurtosis + k0
# For Train & Test datasets (GENES + CELLS features): minimum kurtosis = 1.53655, maximum kurtosis = 30.4929
a = np.array([[1.53655, 1], [30.4929, 1]])
b = np.array([QT_n_quantile_min, QT_n_quantile_max])
return np.linalg.solve(a, b)
def n_quantile_for_kurt(kurt, calc_QT_par_kurt_transform):
# Calculation parameters of function: n_quantile(kurtosis) = calc_QT_par_kurt_transform[0]*kurtosis + calc_QT_par_kurt_transform[1]
return int(calc_QT_par_kurt_transform[0] * kurt + calc_QT_par_kurt_transform[1])
# Function to extract common stats features
def fe_stats(train, test, cols, kind='g'):
for df in [train, test]:
df[f'{kind}_sum'] = df[cols].sum(axis = 1)
df[f'{kind}_mean'] = df[cols].mean(axis = 1)
df[f'{kind}_std'] = df[cols].std(axis = 1)
df[f'{kind}_kurt'] = df[cols].kurtosis(axis = 1)
df[f'{kind}_skew'] = df[cols].skew(axis = 1)
return train, test
def create_pca(train, test, cols, seed, kind='g', n_components=50, drop=False):
train_ = train[cols].copy()
test_ = test[cols].copy()
data = pd.concat([train_, test_], axis = 0)
pca = PCA(n_components=n_components, random_state=seed)
data = pca.fit_transform(data)
columns = [f'{kind}-pca-{i}' for i in range(n_components)]
data = pd.DataFrame(data, columns = columns)
train_ = data.iloc[:train.shape[0]]
test_ = data.iloc[train.shape[0]:].reset_index(drop=True)
if drop:
train = train.drop(columns=cols)
test = test.drop(columns=cols)
train = pd.concat([train.reset_index(drop = True), train_], axis = 1)
test = pd.concat([test.reset_index(drop = True), test_], axis = 1)
return train, test
def fe_squared(train, test, cols):
for df in [train, test]:
for col in cols:
df[f'{col}_squared'] = df[col] ** 2
return train, test
# Function to scale our data
def scaling(train, test, cols):
scaler = RobustScaler()
scaler.fit(pd.concat([train[cols], test[cols]], axis = 0))
train[cols] = scaler.transform(train[cols])
test[cols] = scaler.transform(test[cols])
return train, test
def vt_selection(train, test, cols, VarianceThreshold_for_FS, kind='g', drop=True):
train_ = train[cols].copy()
test_ = test[cols].copy()
data = train_.append(test_)
if kind == 'g':
mask = [True, False, True, True, True, True, False, True, True, True, True, True, True, True, True, False, True, True, False, False, True, False, True, False, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, False, True, True, False, False, False, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, False, True, True, True, True, True, True, True, True, False, True, True, False, True, False, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, False, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, False, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, False, False, True, True, True, True, True, False, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, False, True, True, True, False, True, True, True, True, False, True, True, True, True, True, False, True, True, True, True, True, False, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, False, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, False, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, True, False, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, False, False, True, True, True, True, True, True, True, False, False, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, False, True, True, True, True, True, True, True, True, False, True, False, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, False, True, True, True, True, False, True, False, False, True, False, True, True, True, True, False, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, False, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, False, True, True, True, True, True, True, True, True, False, True, True, True, True, True, True, False, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, True, False, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False]
elif kind == 'c':
mask = [True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, True, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False]
data_transformed = data.loc[:, mask]
train_features_transformed = data_transformed[ : train_.shape[0]]
test_features_transformed = data_transformed[-test_.shape[0] : ]
columns = [f'{kind}-vt-{i}' for i in range(train_features_transformed.shape[1])]
train_features_transformed = pd.DataFrame(train_features_transformed)
train_features_transformed.columns = columns
test_features_transformed = pd.DataFrame(test_features_transformed)
test_features_transformed.columns = columns
if drop:
train = train.drop(columns=cols)
test = test.drop(columns=cols)
train = pd.concat([train.reset_index(drop=True), train_features_transformed], axis = 1)
test = pd.concat([test.reset_index(drop=True), test_features_transformed], axis = 1)
return train, test
def lsvc_selection(train, test, cols, score_cols, kind='g', model='lsvc', drop=True):
train_ = train[cols].copy()
test_ = test[cols].copy()
if model == 'lsvc':
lsvc = LinearSVC(C=0.01, penalty="l1", dual=False).fit(train_.values, [''.join(map(str, elem)) for elem in
train[score_cols].values])
else:
raise NotImplementedError
model = SelectFromModel(lsvc, prefit=True)
train_ = model.transform(train_[cols].values)
test_ = model.transform(test_[cols].values)
columns = [f'{kind}-lsvc-{i}' for i in range(train_.shape[1])]
train_ = pd.DataFrame(train_, columns=columns)
test_ = pd.DataFrame(test_, columns=columns)
if drop:
train = train.drop(columns=cols)
test = test.drop(columns=cols)
train = pd.concat([train.reset_index(drop=True), train_], axis=1)
test = pd.concat([test.reset_index(drop=True), test_], axis=1)
return train, test
def preprocess_fn(train_features, test_features, train_targets_scored, config):
score_cols = [c for c in train_targets_scored.columns if c not in ['sig_id']]
train_features = make_folds(train_features, train_targets_scored, config.n_folds, config.seed[0])
if config.rank_gauss:
g_cols, c_cols, num_g, num_c = get_cols(test_features)
train_features, test_features = create_rankgauss(train_features, test_features, g_cols + c_cols,
config.QT_n_quantile_min, config.QT_n_quantile_max,
seed=config.seed[0])
if config.feature_status:
g_cols, c_cols, num_g, num_c = get_cols(test_features)
train_features, test_features = fe_stats(train_features, test_features, g_cols, kind='g')
train_features, test_features = fe_stats(train_features, test_features, c_cols, kind='c')
if config.pca:
g_cols, c_cols, num_g, num_c = get_cols(test_features)
train_features, test_features = create_pca(train_features, test_features, g_cols,
seed=config.seed[0],
kind='g',
n_components=config.n_comp_GENES)
train_features, test_features = create_pca(train_features, test_features, c_cols,
seed=config.seed[0],
kind='c',
n_components=config.n_comp_CELLS)
if config.cell_squared:
g_cols, c_cols, num_g, num_c = get_cols(test_features)
train_features, test_features = fe_squared(train_features, test_features, c_cols)
if config.scaling:
g_cols, c_cols, num_g, num_c = get_cols(test_features)
train_features, test_features = scaling(train_features, test_features, g_cols)
train_features, test_features = scaling(train_features, test_features, c_cols)
if config.vt_selection:
g_cols, c_cols, num_g, num_c = get_cols(test_features)
train_features, test_features = vt_selection(train_features, test_features, g_cols,
config.VarianceThreshold_for_FS, kind='g', drop=True)
train_features, test_features = vt_selection(train_features, test_features, c_cols,
config.VarianceThreshold_for_FS, kind='c', drop=True)
if config.lsvc_selection:
g_cols, c_cols, num_g, num_c = get_cols(test_features)
train_features, test_features = lsvc_selection(train_features, test_features, g_cols, score_cols, kind='g')
train_features, test_features = lsvc_selection(train_features, test_features, c_cols, score_cols, kind='c')
return train_features, test_featuresInference
Configurations
미리 설정한 config 값을 dataframe으로 관리하여 ensemble할 때 사용할 수 있음.
class config:
name = '2020_11_22_0029'
description = 'Adam, nstep1'
seed = [0, 1, 2]
device = [1]
num_scored = 206
num_nonscored = 402
# preprocess
normalize = True
remove_vehicle = True
use_autoencoder = False
feature_status = False
rank_gauss = True
pca = True
cell_squared = False
scaling = False
vt_selection = True
lsvc_selection = False
n_comp_GENES = 463
n_comp_CELLS = 60
VarianceThreshold_for_FS = 0.84
QT_n_quantile_min = 10
QT_n_quantile_max = 200
# Model
n_d = 128
n_a = 32
n_independent = 1
n_shared = 1
cat_emb_dim = 4
n_steps = 1
gamma = 1.0
lambda_sparse = 0.
# Training
n_folds = 5
n_pretrain_epochs = 100
n_epochs = 100
batch_size = 1024
virtual_batch_size = 128
momentum = 0.02
lr = 2e-2
weight_decay = 1e-5
scheduler_step_size = 50
scheduler_gamma = 0.9
num_workers = 8commit_df = pd.DataFrame(columns = list(config.__dict__.keys()))
commit_df = commit_df.append(dict(config.__dict__), ignore_index=True)Predict testset
def get_tabnet(
input_dim, output_dim, n_d, n_a, n_steps, gamma,
remove_vehicle, cat_emb_dim, n_independent, n_shared,
virtual_batch_size, momentum, epsilon=1e-8, mask_type="sparsemax",
):
if remove_vehicle:
cat_idxs = [0, 1]
cat_dims = [3, 2] # Fixed
else:
cat_idxs = [0, 1, 2]
cat_dims = [2, 3, 2]
if cat_emb_dim == 0:
cat_idxs = []
cat_dims = []
network = TabNet(
input_dim,
output_dim,
n_d=n_d,
n_a=n_a,
n_steps=n_steps,
gamma=gamma,
cat_idxs=cat_idxs,
cat_dims=cat_dims,
cat_emb_dim=cat_emb_dim,
n_independent=n_independent,
n_shared=n_shared,
epsilon=epsilon,
virtual_batch_size=virtual_batch_size,
momentum=momentum,
mask_type=mask_type,
)
return network
def predict(model, loader, device):
model.eval()
preds = []
with torch.no_grad():
for feature in loader:
if device:
feature = feature.to(device)
logit, _ = model(feature)
pred = torch.sigmoid(logit)
preds.append(pred.detach().cpu().numpy())
preds = np.concatenate(preds, axis=0)
return predsdevice = 'cuda' if torch.cuda.is_available() else 'cpu'
data_dir = '../input/lish-moa'
log_dir = '../input/submission'test_cv_preds = []
for idx, config in commit_df.iterrows():
# Load Data
train_features = pd.read_csv(f'{data_dir}/train_features.csv')
train_targets_scored = pd.read_csv(f'{data_dir}/train_targets_scored.csv')
train_targets_nonscored = pd.read_csv(f'{data_dir}/train_targets_nonscored.csv')
test_features = pd.read_csv(f'{data_dir}/test_features.csv')
submission = pd.read_csv(f'{data_dir}/sample_submission.csv')
# Feature Engineering
train_features, test_features = preprocess_fn(train_features, test_features, train_targets_scored, config)
# Preprocess Data
if config.remove_vehicle:
train_features = train_features[train_features.cp_type != 'ctl_vehicle']
train_features = train_features.merge(train_targets_scored, on='sig_id')
train_features = train_features.merge(train_targets_nonscored, on='sig_id')
train_features = cate2num(train_features)
test_features = cate2num(test_features)
if config.normalize:
pass # TODO; add normalization
# Column Infos
score_cols = [c for c in train_targets_scored.columns if c not in ['sig_id']]
nonscore_cols = [c for c in train_targets_nonscored.columns if c not in ['sig_id']]
cate_cols = ['cp_time', 'cp_dose'] if config.remove_vehicle else ['cp_type', 'cp_time', 'cp_dose']
g_cols = [c for c in test_features.columns if 'g-' in c]
c_cols = [c for c in test_features.columns if 'c-' in c]
num_g = len(g_cols)
num_c = len(c_cols)
num_cate = len(cate_cols)
input_cols = g_cols + c_cols
if config.cat_emb_dim > 0:
input_cols = cate_cols + input_cols
X_test = test_features.loc[:, input_cols].values
for seed in config.seed:
for fold in range(config.n_folds):
fold_train = train_features[train_features.fold != fold].reset_index(drop=True)
fold_valid = train_features[train_features.fold == fold].reset_index(drop=True)
X_train, y_train = fold_train.loc[:, input_cols].values, fold_train.loc[:, score_cols].values
X_val, y_val = fold_valid.loc[:, input_cols].values, fold_valid.loc[:, score_cols].values
model = get_tabnet(
input_dim=X_train.shape[-1],
output_dim=y_train.shape[-1],
n_d=config.n_d,
n_a=config.n_a,
n_steps=config.n_steps,
gamma=config.gamma,
remove_vehicle=config.remove_vehicle,
cat_emb_dim=config.cat_emb_dim,
n_independent=config.n_independent,
n_shared=config.n_shared,
virtual_batch_size=config.virtual_batch_size,
momentum=config.momentum,
mask_type="sparsemax",
)
in_ft = model.tabnet.final_mapping.in_features
model.tabnet.final_mapping = nn.Linear(in_ft, y_train.shape[-1])
path = os.path.join(log_dir, f'{config["name"]}_seed{seed}_fold{fold}.pt')
checkpoint = torch.load(path, map_location=lambda storage, loc: storage)
model.load_state_dict(checkpoint['model'])
model = model.to(device)
test_loader = DataLoader(
DatasetWithoutLabel(X_test),
batch_size=config.batch_size,
collate_fn=collate_fn_test,
shuffle=False,
)
preds_test = predict(model, test_loader, device)
test_cv_preds.append(preds_test)
test_preds_all = np.stack(test_cv_preds)
all_feat = [col for col in submission.columns if col not in ["sig_id"]]
# To obtain the same length of test_preds_all and submission
sig_id = test_features[test_features["cp_type"] != "ctl_vehicle"].sig_id.reset_index(drop=True)
tmp = pd.DataFrame(test_preds_all.mean(axis=0), columns=all_feat)
tmp["sig_id"] = sig_idsub = pd.merge(test_features[["sig_id"]], tmp, on="sig_id", how="left")
sub.fillna(0., inplace=True)
sub.to_csv("submission.csv", index = False)
Last updated
Was this helpful?