easy-tensorflow 1.3.0

An interface containing easy tensorflow model building blocks and feature pipelines

Easy Tensorflow:

A Keras and Tensorflow native implementation that mimics the functionality of sklearn's Pipeline objects but implemented using Tensorflow. This interface contains easy feature preprocessing pipelines.

Often when we build ML training pipeline we might end up using an pandas, sklearn Pipeline or FeatureUnion composer for preprocessing and feature encoding for example one-hot-encoding our data. Previous implementation was for Tensroflow feature columns, but these will be depracted in future releases. If we end up using a Keras model after preprocessing we end up with multiple artifacts, one for preprocessing and feature engineering from sklearn and the other a Keras saved model. For this case the preprocessing is not part of Keras model which can cause training-serving skew. Recently Keras implemented these same preprocesing transforms as preprocessing layers. Using these layers the Data scientist/Machine learning engineer can implement the preprocessing layers as part neural net architecture which will prevent training-serving skew. One missing component is a Pipeline type object for Keras preprocessing and Tensorflow feature columns. The EasyFlow project implements these feature preprocessing Pipelines for easier model building with interfaces for both Keras preprocessing layers and Tensroflow feature columns. We also do not use for example pandas dataframes in the Pipeline and use Tensorflow Datasets from the tf.data.Dataset module for faster execution.

Below we will showcase two implementations that can be achieved with easyflow module

• Preprocessing Pipeline using Keras preprocessing layers with easyflow preprocessing module
• Preprocessing Pipeline using Tensorflow feature columns with easyflow feature_encoders module

To install package:

pip install easy-tensorflow

Example 1: Preprocessing Pipeline and FeaturePreprocessorUnion example

The easyflow.preprocessing module contains functionality similar to what sklearn does with its Pipeline, FeatureUnion and ColumnTransformer does. Full example also in notebooks folder

import pandas as pd
import tensorflow as tf
from tensorflow.keras.layers import Normalization, StringLookup, IntegerLookup

# local imports
from easyflow.data import TensorflowDataMapper
from easyflow.preprocessing import FeaturePreprocessorUnion
from easyflow.preprocessing.custom import (
    FeatureInputLayer,
    SequentialPreprocessingChainer,
)

Read in data and map as tf.data.Dataset

Use the TensorflowDataMapper class to map pandas data frame to a tf.data.Dataset type.

file_url = "http://storage.googleapis.com/download.tensorflow.org/data/heart.csv"
dataframe = pd.read_csv(file_url)
labels = dataframe.pop("target")

batch_size = 32
dataset_mapper = TensorflowDataMapper() 
dataset = dataset_mapper.map(dataframe, labels)
train_data_set, val_data_set = dataset_mapper.split_data_set(dataset)
train_data_set = train_data_set.batch(batch_size)
val_data_set = val_data_set.batch(batch_size)

Set constants

NUMERICAL_FEATURES = ['age', 'trestbps', 'chol', 'thalach', 'oldpeak', 'slope']
CATEGORICAL_FEATURES = ['sex', 'cp', 'fbs', 'restecg', 'exang', 'ca']
# thal is represented as a string
STRING_CATEGORICAL_FEATURES = ['thal']

dtype_mapper = {
    "age": tf.float32,
    "sex": tf.float32,
    "cp": tf.float32,
    "trestbps": tf.float32,
    "chol": tf.float32,
    "fbs": tf.float32,
    "restecg": tf.float32,
    "thalach": tf.float32,
    "exang": tf.float32,
    "oldpeak": tf.float32,
    "slope": tf.float32,
    "ca": tf.float32,
    "thal": tf.string,
}

Setup Preprocessing layer using FeatureUnion

feature_preprocessor_list = [
    ('numeric_encoder', Normalization(), NUMERICAL_FEATURES),
    ('categorical_encoder', IntegerLookup(output_mode='binary'), CATEGORICAL_FEATURES),
    # For feature thal we first need to run StringLookup followed by a IntegerLookup layer
    ('string_encoder', 
     SequentialPreprocessingChainer([StringLookup(), IntegerLookup(output_mode='binary')]),
     STRING_CATEGORICAL_FEATURES)
]

preprocessor = FeaturePreprocessorUnion(feature_preprocessor_list)
preprocessor.adapt(train_data_set)

feature_layer_inputs = FeatureInputLayer(dtype_mapper)
preprocessing_layer = preprocessor(feature_layer_inputs)

Set up network

# setup simple network
x = tf.keras.layers.Dense(128, activation="relu")(preprocessing_layer)
x = tf.keras.layers.Dropout(0.5)(x)
outputs = tf.keras.layers.Dense(1, activation='sigmoid')(x)
model = tf.keras.Model(inputs=feature_layer_inputs, outputs=outputs)
model.compile(
    optimizer=tf.keras.optimizers.Adam(),
    loss=tf.keras.losses.BinaryCrossentropy(),
    metrics=[tf.keras.metrics.BinaryAccuracy(name='accuracy'), tf.keras.metrics.AUC(name='auc')])

Fit model

history=model.fit(train_data_set, validation_data=val_data_set, epochs=10)

Example 2: Model building Pipeline using easyflow feature encoders module

This module is a fusion between keras layers and tensorflow feature columns.

FeatureColumnTransformer and FeatureUnionTransformer are the main interfaces and serves as feature transformation pipelines.

Wrapper classes exists for the following feature_columns

• CategoricalFeatureEncoder
• EmbeddingFeatureEncoder
• EmbeddingCrossingFeatureEncoder
• CategoryCrossingFeatureEncoder
• NumericalFeatureEncoder
• BucketizedFeatureEncoder

To create a custom encoder or one where wrapper class does not exist, there are two base interfaces to use:

• BaseFeatureColumnEncoder
• BaseCategoricalFeatureColumnEncoder

import pandas as pd
import tensorflow as tf

# local imports
from easyflow.data import TensorflowDataMapper
from easyflow.feature_encoders import FeatureColumnTransformer, FeatureUnionTransformer
from easyflow.feature_encoders import NumericalFeatureEncoder, EmbeddingFeatureEncoder, CategoricalFeatureEncoder

Load data

CSV_HEADER = [
    "age",
    "workclass",
    "fnlwgt",
    "education",
    "education_num",
    "marital_status",
    "occupation",
    "relationship",
    "race",
    "gender",
    "capital_gain",
    "capital_loss",
    "hours_per_week",
    "native_country",
    "income_bracket",
]

data_url = "https://archive.ics.uci.edu/ml/machine-learning-databases/adult/adult.data"


data_frame = pd.read_csv(data_url, header=None, names=CSV_HEADER)
labels = data_frame.pop("income_bracket")
labels_binary = 1.0 * (labels == " >50K")
data_frame.to_csv('adult_features.csv', index=False)
labels_binary.to_csv('adult_labels.csv', index=False)

Map data frame to tf.data.Dataset

batch_size = 256
dataset_mapper = TensorflowDataMapper() 
dataset = dataset_mapper.map(data_frame, labels_binary)

train_data_set, val_data_set = dataset_mapper.split_data_set(dataset)
train_data_set = train_data_set.batch(batch_size)
val_data_set = val_data_set.batch(batch_size)

Set up the feature encoding list

NUMERIC_FEATURE_NAMES = [
    "age",
    "education_num",
    "capital_gain",
    "capital_loss",
    "hours_per_week",
]

CATEGORICAL_FEATURES_NAMES = [
    "workclass",
    "marital_status",
    "relationship",
    "race",
    "gender"]

EMBEDDING_FEATURES_NAMES = ['education',
                            'occupation',
                            'native_country']

feature_encoder_list = [('numerical_features', NumericalFeatureEncoder(), NUMERIC_FEATURE_NAMES),
                        ('categorical_features', CategoricalFeatureEncoder(), CATEGORICAL_FEATURES_NAMES),
                        ('embedding_features_deep', EmbeddingFeatureEncoder(dimension=10), EMBEDDING_FEATURES_NAMES),
                        ('embedding_features_wide', CategoricalFeatureEncoder(), EMBEDDING_FEATURES_NAMES)]

Setting up feature layer and feature encoders

There are two main column transformer classes namely FeatureColumnTransformer and FeatureUnionTransformer. For this example we are going to build a Wide and Deep model architecture. So we will be using the FeatureColumnTransformer since it gives us more flexibility. FeatureUnionTransformer concatenates all the features in the input layer

feature_layer_inputs, feature_layer =  FeatureColumnTransformer(feature_encoder_list).transform(train_data_set)

deep = tf.keras.layers.concatenate([feature_layer['numerical_features'],
                                    feature_layer['categorical_features'],
                                    feature_layer['embedding_features_deep']])

wide = feature_layer['embedding_features_wide']

Set up Wide and Deep model architecture

deep = tf.keras.layers.BatchNormalization()(deep)

for nodes in [128, 64, 32]:
    deep = tf.keras.layers.Dense(nodes, activation='relu')(deep)
    deep = tf.keras.layers.Dropout(0.5)(deep)

# combine wide and deep layers
wide_and_deep = tf.keras.layers.concatenate([deep, wide])
output = tf.keras.layers.Dense(1, activation='sigmoid')(wide_and_deep)
model = tf.keras.Model(inputs=[v for v in feature_layer_inputs.values()], outputs=output)
model.compile(loss=tf.keras.losses.BinaryCrossentropy(label_smoothing=0.0),
              optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),
              metrics=[tf.keras.metrics.BinaryAccuracy(name='accuracy'), tf.keras.metrics.AUC(name='auc')])

Fit model

model.fit(train_data_set, validation_data=val_data_set, epochs=10)