使用tf.estimator训练Boosted Trees

1，数据集下载

!pip install -q tf-nightly  # Requires tf 1.13
from __future__ import absolute_import, division, print_function

import numpy as np
import pandas as pd
import tensorflow as tf

tf.enable_eager_execution()

tf.logging.set_verbosity(tf.logging.ERROR)
tf.set_random_seed(123)

# Load dataset.
dftrain = pd.read_csv('https://storage.googleapis.com/tfbt/titanic_train.csv')
dfeval = pd.read_csv('https://storage.googleapis.com/tfbt/titanic_eval.csv')
y_train = dftrain.pop('survived')
y_eval = dfeval.pop('survived')

2，一系列数据检查

这一条特别

dftrain.sex.value_counts().plot(kind='barh');
(dftrain['class']
  .value_counts()
  .plot(kind='barh'));

3，Create feature columns and input functions，特征列和输入函数

3.1，one-hot-encoding, normalization, and bucketization

3.2，数字型和分类型

fc = tf.feature_column
CATEGORICAL_COLUMNS = ['sex', 'n_siblings_spouses', 'parch', 'class', 'deck',
                       'embark_town', 'alone']
NUMERIC_COLUMNS = ['age', 'fare']
 
def one_hot_cat_column(feature_name, vocab):
  return fc.indicator_column(
      fc.categorical_column_with_vocabulary_list(feature_name,
                                                 vocab))
feature_columns = []
for feature_name in CATEGORICAL_COLUMNS:
  # Need to one-hot encode categorical features.
  vocabulary = dftrain[feature_name].unique()
  print(feature_name,vocabulary)
  feature_columns.append(one_hot_cat_column(feature_name, vocabulary))
  print(feature_columns," ")
 
for feature_name in NUMERIC_COLUMNS:
  feature_columns.append(fc.numeric_column(feature_name,
                                           dtype=tf.float32))

3.3，view all of the feature column transformations//第0条example

fc.input_layer(dict(example), feature_columns).numpy()

3.4，create the input functions创建输入函数

# Use entire batch since this is such a small dataset.
NUM_EXAMPLES = len(y_train)

def make_input_fn(X, y, n_epochs=None, shuffle=True):
  def input_fn():
    dataset = tf.data.Dataset.from_tensor_slices((dict(X), y))
    if shuffle:
      dataset = dataset.shuffle(NUM_EXAMPLES)
    # For training, cycle thru dataset as many times as need (n_epochs=None).    
    dataset = dataset.repeat(n_epochs)  
    # In memory training doesn't use batching.
    dataset = dataset.batch(NUM_EXAMPLES)
    return dataset
  return input_fn

# Training and evaluation input functions.
train_input_fn = make_input_fn(dftrain, y_train)
eval_input_fn = make_input_fn(dfeval, y_eval, shuffle=False, n_epochs=1)

4，Train and evaluate the model训练和评估

4.1，first train a linear classifier (logistic regression model)，训练一个二分类模型

linear_est = tf.estimator.LinearClassifier(feature_columns)

# Train model.
linear_est.train(train_input_fn, max_steps=100)

# Evaluation.
results = linear_est.evaluate(eval_input_fn)
print('Accuracy : ', results['accuracy'])
print('Dummy model: ', results['accuracy_baseline'])

4.2，train a Boosted Trees model

# Since data fits into memory, use entire dataset per layer. It will be faster.
# Above one batch is defined as the entire dataset.
n_batches = 1
est = tf.estimator.BoostedTreesClassifier(feature_columns,n_batches_per_layer=n_batches)

//est = tf.estimator.BoostedTreesClassifier(feature_columns,n_batches_per_layer=n_batches,n_trees=300)

# The model will stop training once the specified number of trees is built, not
# based on the number of steps.
est.train(train_input_fn, max_steps=100)

# Eval.
results = est.evaluate(eval_input_fn)
print('Accuracy : ', results['accuracy'])
print('Dummy model: ', results['accuracy_baseline'])

4.3，内存模式

def make_inmemory_train_input_fn(X, y):
  def input_fn():
    return dict(X), y
  return input_fn


train_input_fn = make_inmemory_train_input_fn(dftrain, y_train)
eval_input_fn = make_input_fn(dfeval, y_eval, shuffle=False, n_epochs=1)
est = tf.contrib.estimator.boosted_trees_classifier_train_in_memory(
    train_input_fn,
    feature_columns)
print(est.evaluate(eval_input_fn)['accuracy'])

5，预测

pred_dicts = list(est.predict(eval_input_fn))
probs = pd.Series([pred['probabilities'][1] for pred in pred_dicts])

probs.plot(kind='hist', bins=20, title='predicted probabilities');

6，看看roc曲线

from sklearn.metrics import roc_curve
from matplotlib import pyplot as plt

fpr, tpr, _ = roc_curve(y_eval, probs)
plt.plot(fpr, tpr)
plt.title('ROC curve')
plt.xlabel('false positive rate')
plt.ylabel('true positive rate')
plt.xlim(0,)
plt.ylim(0,);