比肩xgboost和lightgbm表格数据的最强神器tabnet，而2020年退出的tabnet神经网络专门针对表格数据而诞生的神器。实战，单任务和多任务回归！！！

---

前言

比肩xgboost和lightgbm表格数据的最强神器tabnet，而2020年退出的tabnet神经网络专门针对表格数据而诞生的神器。

上图是官网的论文给出的效果，我们可以看出来，catnet效果还是不错的，而且数据量越大效果比机器学习的更好。

---

`

一、官网的实战案例

1.pytorch版的回归任务

from pytorch_tabnet.tab_model import TabNetRegressor
import torch
from sklearn.preprocessing import LabelEncoder
from sklearn.metrics import mean_squared_error
import pandas as pd
import numpy as np
import os
import wget
from pathlib import Path
np.random.seed(0)

"导入数据"
url = "https://archive.ics.uci.edu/ml/machine-learning-databases/adult/adult.data"
dataset_name = 'census-income'
out = Path(os.getcwd()+'/data/'+dataset_name+'.csv')

out.parent.mkdir(parents=True, exist_ok=True)
if out.exists():
    print("File already exists.")
else:
    print("Downloading file...")
    wget.download(url, out.as_posix())

train = pd.read_csv(out)
target = ' <=50K'
if "Set" not in train.columns:
    train["Set"] = np.random.choice(["train", "valid", "test"], p =[.8, .1, .1], size=(train.shape[0],))

train_indices = train[train.Set=="train"].index
valid_indices = train[train.Set=="valid"].index
test_indices = train[train.Set=="test"].index

categorical_columns = []
categorical_dims =  {}
for col in train.columns[train.dtypes == object]:
    print(col, train[col].nunique())
    l_enc = LabelEncoder()
    train[col] = train[col].fillna("VV_likely")
    train[col] = l_enc.fit_transform(train[col].values)
    categorical_columns.append(col)
    categorical_dims[col] = len(l_enc.classes_)

for col in train.columns[train.dtypes == 'float64']:
    train.fillna(train.loc[train_indices, col].mean(), inplace=True)

unused_feat = ['Set']

features = [ col for col in train.columns if col not in unused_feat+[target]] 

cat_idxs = [ i for i, f in enumerate(features) if f in categorical_columns]

cat_dims = [ categorical_dims[f] for i, f in enumerate(features) if f in categorical_columns]

# define your embedding sizes : here just a random choice
cat_emb_dim = [5, 4, 3, 6, 2, 2, 1, 10]

clf = TabNetRegressor(cat_dims=cat_dims, cat_emb_dim=cat_emb_dim, cat_idxs=cat_idxs)

X_train = train[features].values[train_indices]
y_train = train[target].values[train_indices].reshape(-1, 1)

X_valid = train[features].values[valid_indices]
y_valid = train[target].values[valid_indices].reshape(-1, 1)

X_test = train[features].values[test_indices]
y_test = train[target].values[test_indices].reshape(-1, 1)

max_epochs = 100 if not os.getenv("CI", False) else 2

clf.fit(
    X_train=X_train, y_train=y_train,
    eval_set=[(X_train, y_train), (X_valid, y_valid)],
    eval_name=['train', 'valid'],
    eval_metric=['rmsle', 'mae', 'rmse', 'mse'],
    max_epochs=max_epochs,
    patience=50,
    batch_size=1024, virtual_batch_size=128,
    num_workers=0,
    drop_last=False,
    augmentations=aug, #aug
) 

preds = clf.predict(X_test)

y_true = y_test

test_score = mean_squared_error(y_pred=preds, y_true=y_true)

print(f"BEST VALID SCORE FOR {dataset_name} : {clf.best_cost}")
print(f"FINAL TEST SCORE FOR {dataset_name} : {test_score}")

# save tabnet model
saving_path_name = "./tabnet_model_test_1"
saved_filepath = clf.save_model(saving_path_name)

# define new model with basic parameters and load state dict weights
loaded_clf = TabNetRegressor()
loaded_clf.load_model(saved_filepath)

loaded_preds = loaded_clf.predict(X_test)
loaded_test_mse = mean_squared_error(loaded_preds, y_test)

print(f"FINAL TEST SCORE FOR {dataset_name} : {loaded_test_mse}")

代码已调通，此任务为单任务回归
下面为多任务回归，区别 在划分数据集那里通过np.tile()将目标字段横着复制八份，np.transpose()将目标字段转置就由行变成列了就是八列了

from pytorch_tabnet.tab_model import TabNetRegressor
import torch
from sklearn.preprocessing import LabelEncoder
from sklearn.metrics import mean_squared_error
import pandas as pd
import numpy as np
import os
import wget
from pathlib import Path
np.random.seed(0)

"导入数据"
url = "https://archive.ics.uci.edu/ml/machine-learning-databases/adult/adult.data"
dataset_name = 'census-income'
out = Path(os.getcwd()+'/data/'+dataset_name+'.csv')

out.parent.mkdir(parents=True, exist_ok=True)
if out.exists():
    print("File already exists.")
else:
    print("Downloading file...")
    wget.download(url, out.as_posix())

train = pd.read_csv(out)
target = ' <=50K'
if "Set" not in train.columns:
    train["Set"] = np.random.choice(["train", "valid", "test"], p =[.8, .1, .1], size=(train.shape[0],))

train_indices = train[train.Set=="train"].index
valid_indices = train[train.Set=="valid"].index
test_indices = train[train.Set=="test"].index

categorical_columns = []
categorical_dims =  {}
for col in train.columns[train.dtypes == object]:
    print(col, train[col].nunique())
    l_enc = LabelEncoder()
    train[col] = train[col].fillna("VV_likely")
    train[col] = l_enc.fit_transform(train[col].values)
    categorical_columns.append(col)
    categorical_dims[col] = len(l_enc.classes_)

for col in train.columns[train.dtypes == 'float64']:
    train.fillna(train.loc[train_indices, col].mean(), inplace=True)

unused_feat = ['Set']

features = [ col for col in train.columns if col not in unused_feat+[target]] 

cat_idxs = [ i for i, f in enumerate(features) if f in categorical_columns]

cat_dims = [ categorical_dims[f] for i, f in enumerate(features) if f in categorical_columns]

# define your embedding sizes : here just a random choice
cat_emb_dim = [5, 4, 3, 6, 2, 2, 1, 10]

clf = TabNetRegressor(cat_dims=cat_dims, cat_emb_dim=cat_emb_dim, cat_idxs=cat_idxs)


**"""**这里做了一点区别，是多任务回归，意思就是目标列有8个****
n_targets = 8

X_train = train[features].values[train_indices]
y_train = train[target].values[train_indices]
y_train = np.transpose(np.tile(y_train, (n_targets,1)))

X_valid = train[features].values[valid_indices]
y_valid = train[target].values[valid_indices]
y_valid = np.transpose(np.tile(y_valid, (n_targets,1)))

X_test = train[features].values[test_indices]
y_test = train[target].values[test_indices]
y_test = np.transpose(np.tile(y_test, (n_targets,1)))

max_epochs = 100 if not os.getenv("CI", False) else 2

clf.fit(
    X_train=X_train, y_train=y_train,
    eval_set=[(X_train, y_train), (X_valid, y_valid)],
    eval_name=['train', 'valid'],
    eval_metric=['rmsle', 'mae', 'rmse', 'mse'],
    max_epochs=max_epochs,
    patience=50,
    batch_size=1024, virtual_batch_size=128,
    num_workers=0,
    drop_last=False,
    augmentations=aug, #aug
) 

preds = clf.predict(X_test)

y_true = y_test

test_score = mean_squared_error(y_pred=preds, y_true=y_true)

print(f"BEST VALID SCORE FOR {dataset_name} : {clf.best_cost}")
print(f"FINAL TEST SCORE FOR {dataset_name} : {test_score}")

# save tabnet model
saving_path_name = "./tabnet_model_test_1"
saved_filepath = clf.save_model(saving_path_name)

# define new model with basic parameters and load state dict weights
loaded_clf = TabNetRegressor()
loaded_clf.load_model(saved_filepath)

loaded_preds = loaded_clf.predict(X_test)
loaded_test_mse = mean_squared_error(loaded_preds, y_test)

print(f"FINAL TEST SCORE FOR {dataset_name} : {loaded_test_mse}")

总结

多看代码，多理解就能掌握该算法啦。加油，家人们！