In [0]:
#@title Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
|
|
|
|
In [0]:
!pip install sklearn
!pip install -U tf-estimator-nightly
!pip install -U tf-nightly
In [0]:
import os
import sys
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from IPython.display import clear_output
from six.moves import urllib
In [0]:
import tensorflow.compat.v2.feature_column as fc
import tensorflow as tf
In [0]:
# Load dataset.
dftrain = pd.read_csv('https://storage.googleapis.com/tf-datasets/titanic/train.csv')
dfeval = pd.read_csv('https://storage.googleapis.com/tf-datasets/titanic/eval.csv')
y_train = dftrain.pop('survived')
y_eval = dfeval.pop('survived')
The dataset contains the following features
In [0]:
dftrain.head()
In [0]:
dftrain.describe()
There are 627 and 264 examples in the training and evaluation sets, respectively.
In [0]:
dftrain.shape[0], dfeval.shape[0]
The majority of passengers are in their 20's and 30's.
In [0]:
dftrain.age.hist(bins=20)
There are approximately twice as many male passengers as female passengers aboard.
In [0]:
dftrain.sex.value_counts().plot(kind='barh')
The majority of passengers were in the "third" class.
In [0]:
dftrain['class'].value_counts().plot(kind='barh')
Females have a much higher chance of surviving versus males. This is clearly a predictive feature for the model.
In [0]:
pd.concat([dftrain, y_train], axis=1).groupby('sex').survived.mean().plot(kind='barh').set_xlabel('% survive')
Estimators use a system called feature columns to describe how the model should interpret each of the raw input features. An Estimator expects a vector of numeric inputs, and feature columns describe how the model should convert each feature.
Selecting and crafting the right set of feature columns is key to learning an effective model. A feature column can be either one of the raw inputs in the original features dict (a base feature column), or any new columns created using transformations defined over one or multiple base columns (a derived feature columns).
The linear estimator uses both numeric and categorical features. Feature columns work with all TensorFlow estimators and their purpose is to define the features used for modeling. Additionally, they provide some feature engineering capabilities like one-hot-encoding, normalization, and bucketization.
In [0]:
CATEGORICAL_COLUMNS = ['sex', 'n_siblings_spouses', 'parch', 'class', 'deck',
'embark_town', 'alone']
NUMERIC_COLUMNS = ['age', 'fare']
feature_columns = []
for feature_name in CATEGORICAL_COLUMNS:
vocabulary = dftrain[feature_name].unique()
feature_columns.append(tf.feature_column.categorical_column_with_vocabulary_list(feature_name, vocabulary))
for feature_name in NUMERIC_COLUMNS:
feature_columns.append(tf.feature_column.numeric_column(feature_name, dtype=tf.float32))
The input_function specifies how data is converted to a tf.data.Dataset that feeds the input pipeline in a streaming fashion. tf.data.Dataset can take in multiple sources such as a dataframe, a csv-formatted file, and more.
In [0]:
def make_input_fn(data_df, label_df, num_epochs=10, shuffle=True, batch_size=32):
def input_function():
ds = tf.data.Dataset.from_tensor_slices((dict(data_df), label_df))
if shuffle:
ds = ds.shuffle(1000)
ds = ds.batch(batch_size).repeat(num_epochs)
return ds
return input_function
train_input_fn = make_input_fn(dftrain, y_train)
eval_input_fn = make_input_fn(dfeval, y_eval, num_epochs=1, shuffle=False)
You can inspect the dataset:
In [0]:
ds = make_input_fn(dftrain, y_train, batch_size=10)()
for feature_batch, label_batch in ds.take(1):
print('Some feature keys:', list(feature_batch.keys()))
print()
print('A batch of class:', feature_batch['class'].numpy())
print()
print('A batch of Labels:', label_batch.numpy())
You can also inspect the result of a specific feature column using the tf.keras.layers.DenseFeatures layer:
In [0]:
age_column = feature_columns[7]
tf.keras.layers.DenseFeatures([age_column])(feature_batch).numpy()
DenseFeatures only accepts dense tensors, to inspect a categorical column you need to transform that to a indicator column first:
In [0]:
gender_column = feature_columns[0]
tf.keras.layers.DenseFeatures([tf.feature_column.indicator_column(gender_column)])(feature_batch).numpy()
After adding all the base features to the model, let's train the model. Training a model is just a single command using the tf.estimator API:
In [0]:
linear_est = tf.estimator.LinearClassifier(feature_columns=feature_columns)
linear_est.train(train_input_fn)
result = linear_est.evaluate(eval_input_fn)
clear_output()
print(result)
Now you reached an accuracy of 75%. Using each base feature column separately may not be enough to explain the data. For example, the correlation between gender and the label may be different for different gender. Therefore, if you only learn a single model weight for gender="Male" and gender="Female", you won't capture every age-gender combination (e.g. distinguishing between gender="Male" AND age="30" AND gender="Male" AND age="40").
To learn the differences between different feature combinations, you can add crossed feature columns to the model (you can also bucketize age column before the cross column):
In [0]:
age_x_gender = tf.feature_column.crossed_column(['age', 'sex'], hash_bucket_size=100)
After adding the combination feature to the model, let's train the model again:
In [0]:
derived_feature_columns = [age_x_gender]
linear_est = tf.estimator.LinearClassifier(feature_columns=feature_columns+derived_feature_columns)
linear_est.train(train_input_fn)
result = linear_est.evaluate(eval_input_fn)
clear_output()
print(result)
It now achieves an accuracy of 77.6%, which is slightly better than only trained in base features. You can try using more features and transformations to see if you can do better!
Now you can use the train model to make predictions on a passenger from the evaluation set. TensorFlow models are optimized to make predictions on a batch, or collection, of examples at once. Earlier, the eval_input_fn was defined using the entire evaluation set.
In [0]:
pred_dicts = list(linear_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')
Finally, look at the receiver operating characteristic (ROC) of the results, which will give us a better idea of the tradeoff between the true positive rate and false positive rate.
In [0]:
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,)