def test_catboost():
try:
import catboost
from catboost.datasets import amazon
except:
print("Skipping test_catboost!")
return
import shap
import numpy as np
# train catboost model
X, y = shap.datasets.boston()
X["RAD"] = X["RAD"].astype(np.int)
model = catboost.CatBoostRegressor(iterations=300,
learning_rate=0.1,
random_seed=123)
p = catboost.Pool(X, y, cat_features=["RAD"])
model.fit(p, verbose=False, plot=False)
# explain the model's predictions using SHAP values
ex = shap.TreeExplainer(model)
shap_values = ex.shap_values(p)
predicted = model.predict(X)
assert np.abs(shap_values.sum(1) + ex.expected_value - predicted).max() < 1e-4, \
"SHAP values don't sum to model output!"
train_df, _ = amazon()
ix = 100
X_train = train_df.drop('ACTION', axis=1)[:ix]
y_train = train_df.ACTION[:ix]
X_val = train_df.drop('ACTION', axis=1)[ix:ix + 20]
y_val = train_df.ACTION[ix:ix + 20]
model = catboost.CatBoostClassifier(iterations=100,
learning_rate=0.5,
random_seed=12)
model.fit(X_train,
y_train,
eval_set=(X_val, y_val),
verbose=False,
plot=False)
shap.TreeExplainer(model)
# In[21]:
import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)
import gc
import os
# # Exploratory Data Analysis
# In[22]:
# Loading data directly from CatBoost
from catboost.datasets import amazon
train, test = amazon()
# In[23]:
print("Train shape: {}, Test shape: {}".format(train.shape, test.shape))
# In[24]:
train.head(5)
# In[25]:
test.head(5)
# dataset has 9 columns, plus target (`ACTION`) for train and `id` for test.
# All these columns are categorical encoded as integers.
import catboost
from catboost import CatBoostClassifier
from catboost import Pool
from catboost.utils import create_cd
from sklearn.model_selection import train_test_split
#######################
# DATA EXPLORATION
#######################
#Set the seed
random_seed = 12345
# Read in the dataset
from catboost.datasets import amazon
(train_df, test_df) = amazon()
# Extract labels
y = train_df.ACTION
X = train_df.drop('ACTION', axis=1)
# Define categorical variables
cat_features = list(range(0, X.shape[1]))
print(cat_features)
#######################
# BASIC MODEL
#######################
# Create a basic model with 200 trees
model = CatBoostClassifier(iterations=200)
@author: gandhi
"""
import itertools
import numpy as np
import pandas as pd
import os
from catboost.datasets import amazon
import seaborn as sns
import matplotlib.pyplot as plt
import sklearn.model_selection
from sklearn.model_selection import StratifiedKFold
from sklearn.model_selection import cross_validate
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report, confusion_matrix, precision_recall_curve, roc_curve, roc_auc_score, log_loss
##Extracting the data
train_data, test_data = amazon()
train_data.head()
df = train_data
## Analysing the data
plt.figure()
plt.figure(figsize=(30, 20))
for i in range(1, 10):
plt.subplot(5, 2, i)
plt.hist(df[df.columns[i]])
plt.xlabel(df.columns[i])
plt.ylabel("Frequency")
plt.show()
plt.figure(figsize=(30, 20))
sns.heatmap(df.corr(), annot=True, cmap='viridis', linewidth=1)
plt.figure(figsize=(10, 30))
def main():
# fill NaN's with <UNK> token
unk_token = '<UNK>'
# retrieve dataset
start = time.time()
if os.path.exists('raw'):
train = pd.read_csv('raw/train.csv')
test = pd.read_csv('raw/test.csv')
else:
data, _ = amazon() # train is the only one with labels
os.makedirs('raw', exist_ok=True)
train = data[:26215] # same split they use in Scharchilev et al.
test = data[26215:]
train.to_csv('raw/train.csv', index=None)
test.to_csv('raw/test.csv', index=None)
print('time to load amazon: {}'.format(time.time() - start))
# define columns
label_col = 'ACTION'
feature_col = list(train.columns)
feature_col.remove(label_col)
# nan rows
train_nan_rows = train[train.isnull().any(axis=1)]
test_nan_rows = test[test.isnull().any(axis=1)]
print('train nan rows: {}'.format(len(train_nan_rows)))
print('test nan rows: {}'.format(len(test_nan_rows)))
# fit encoders and fill in NaNs with unknown token or mean value
encoders = {}
for col in feature_col:
if str(train[col].dtype) == 'object':
train[col] = train[col].fillna(unk_token)
test[col] = test[col].fillna(unk_token)
encoders[col] = OrdinalEncoder().fit(train[col].to_numpy().reshape(
-1, 1))
else:
train[col] = train[col].fillna(int(train[col].mean()))
test[col] = test[col].fillna(int(test[col].mean()))
label_encoder = LabelEncoder().fit(train[label_col])
# transform train dataframe
new_train = train.copy()
for col in feature_col:
if col in encoders:
new_train[col] = encoders[col].transform(
new_train[col].to_numpy().reshape(-1, 1))
new_train[label_col] = label_encoder.transform(new_train[label_col])
# transform test dataframe
new_test = test.copy()
for col in feature_col:
if col in encoders:
new_test[col] = encoders[col].transform(
new_test[col].to_numpy().reshape(-1, 1))
new_test[label_col] = label_encoder.transform(new_test[label_col])
# show difference
print('train')
print(train.head(5))
print(new_train.head(5))
print('test')
print(test.head(5))
print(new_test.head(5))
# save to numpy format
print('saving to train.npy...')
np.save('train.npy', new_train.to_numpy())
print('saving to test.npy...')
np.save('test.npy', new_test.to_numpy())
np.save('feature.npy', feature_col)