Missed TensorFlow World? Check out the recap. Learn more

Load a pandas.DataFrame

View on TensorFlow.org View source on GitHub Download notebook

This tutorial provides an example of how to load pandas dataframes into a tf.data.Dataset.

This tutorials uses a small dataset provided by the Cleveland Clinic Foundation for Heart Disease. There are several hundred rows in the CSV. Each row describes a patient, and each column describes an attribute. We will use this information to predict whether a patient has heart disease, which in this dataset is a binary classification task.

Read data using pandas

from __future__ import absolute_import, division, print_function, unicode_literals

import pandas as pd
import tensorflow as tf

Download the csv file containing the heart dataset.

csv_file = tf.keras.utils.get_file('heart.csv', 'https://storage.googleapis.com/applied-dl/heart.csv')
Downloading data from https://storage.googleapis.com/applied-dl/heart.csv
16384/13273 [=====================================] - 0s 0us/step

Read the csv file using pandas.

df = pd.read_csv(csv_file)
df.head()
df.dtypes
age           int64
sex           int64
cp            int64
trestbps      int64
chol          int64
fbs           int64
restecg       int64
thalach       int64
exang         int64
oldpeak     float64
slope         int64
ca            int64
thal         object
target        int64
dtype: object

Convert thal column which is an object in the dataframe to a discrete numerical value.

df['thal'] = pd.Categorical(df['thal'])
df['thal'] = df.thal.cat.codes
df.head()

Load data using tf.data.Dataset

Use tf.data.Dataset.from_tensor_slices to read the values from a pandas dataframe.

One of the advantages of using tf.data.Dataset is it allows you to write simple, highly efficient data pipelines. Read the loading data guide to find out more.

target = df.pop('target')
dataset = tf.data.Dataset.from_tensor_slices((df.values, target.values))
for feat, targ in dataset.take(5):
  print ('Features: {}, Target: {}'.format(feat, targ))
Features: [ 63.    1.    1.  145.  233.    1.    2.  150.    0.    2.3   3.    0.
   2. ], Target: 0
Features: [ 67.    1.    4.  160.  286.    0.    2.  108.    1.    1.5   2.    3.
   3. ], Target: 1
Features: [ 67.    1.    4.  120.  229.    0.    2.  129.    1.    2.6   2.    2.
   4. ], Target: 0
Features: [ 37.    1.    3.  130.  250.    0.    0.  187.    0.    3.5   3.    0.
   3. ], Target: 0
Features: [ 41.    0.    2.  130.  204.    0.    2.  172.    0.    1.4   1.    0.
   3. ], Target: 0

Since a pd.Series implements the __array__ protocol it can be used transparently nearly anywhere you would use a np.array or a tf.Tensor.

tf.constant(df['thal'])
<tf.Tensor: id=21, shape=(303,), dtype=int32, numpy=
array([2, 3, 4, 3, 3, 3, 3, 3, 4, 4, 2, 3, 2, 4, 4, 3, 4, 3, 3, 3, 3, 3,
       3, 4, 4, 3, 3, 3, 3, 4, 3, 4, 3, 4, 3, 3, 4, 2, 4, 3, 4, 3, 4, 4,
       2, 3, 3, 4, 3, 3, 4, 3, 3, 3, 4, 3, 3, 3, 3, 3, 3, 4, 4, 3, 3, 4,
       4, 2, 3, 3, 4, 3, 4, 3, 3, 4, 4, 3, 3, 4, 4, 3, 3, 3, 3, 4, 4, 4,
       3, 3, 4, 3, 4, 4, 3, 4, 3, 3, 3, 4, 3, 4, 4, 3, 3, 4, 4, 4, 4, 4,
       3, 3, 3, 3, 4, 3, 4, 3, 4, 4, 3, 3, 2, 4, 4, 2, 3, 3, 4, 4, 3, 4,
       3, 3, 4, 2, 4, 4, 3, 4, 3, 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4,
       4, 3, 3, 3, 4, 3, 4, 3, 4, 3, 3, 3, 3, 3, 3, 3, 4, 3, 3, 3, 3, 3,
       3, 3, 3, 3, 3, 3, 3, 4, 4, 3, 3, 3, 3, 3, 3, 3, 3, 4, 3, 4, 3, 2,
       4, 4, 3, 3, 3, 3, 3, 3, 4, 3, 3, 3, 3, 3, 2, 2, 4, 3, 4, 2, 4, 3,
       3, 4, 3, 3, 3, 3, 4, 3, 4, 3, 4, 2, 2, 4, 3, 4, 3, 2, 4, 3, 3, 2,
       4, 4, 4, 4, 3, 0, 3, 3, 3, 3, 1, 4, 3, 3, 3, 4, 3, 4, 3, 3, 3, 4,
       3, 3, 4, 4, 4, 4, 3, 3, 4, 3, 4, 3, 4, 4, 3, 4, 4, 3, 4, 4, 3, 3,
       3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 3, 2, 4, 4, 4, 4], dtype=int32)>

Shuffle and batch the dataset.

train_dataset = dataset.shuffle(len(df)).batch(1)

Create and train a model

def get_compiled_model():
  model = tf.keras.Sequential([
    tf.keras.layers.Dense(10, activation='relu'),
    tf.keras.layers.Dense(10, activation='relu'),
    tf.keras.layers.Dense(1, activation='sigmoid')
  ])

  model.compile(optimizer='adam',
                loss='binary_crossentropy',
                metrics=['accuracy'])
  return model
model = get_compiled_model()
model.fit(train_dataset, epochs=15)
WARNING:tensorflow:Layer sequential is casting an input tensor from dtype float64 to the layer's dtype of float32, which is new behavior in TensorFlow 2.  The layer has dtype float32 because it's dtype defaults to floatx.

If you intended to run this layer in float32, you can safely ignore this warning. If in doubt, this warning is likely only an issue if you are porting a TensorFlow 1.X model to TensorFlow 2.

To change all layers to have dtype float64 by default, call `tf.keras.backend.set_floatx('float64')`. To change just this layer, pass dtype='float64' to the layer constructor. If you are the author of this layer, you can disable autocasting by passing autocast=False to the base Layer constructor.

Epoch 1/15
303/303 [==============================] - 2s 7ms/step - loss: 3.1701 - accuracy: 0.5809
Epoch 2/15
303/303 [==============================] - 1s 3ms/step - loss: 0.9615 - accuracy: 0.6733
Epoch 3/15
303/303 [==============================] - 1s 3ms/step - loss: 0.8341 - accuracy: 0.6535
Epoch 4/15
303/303 [==============================] - 1s 3ms/step - loss: 0.8440 - accuracy: 0.7030
Epoch 5/15
303/303 [==============================] - 1s 3ms/step - loss: 0.7243 - accuracy: 0.6898
Epoch 6/15
303/303 [==============================] - 1s 3ms/step - loss: 0.6807 - accuracy: 0.7030
Epoch 7/15
303/303 [==============================] - 1s 3ms/step - loss: 0.7206 - accuracy: 0.7030
Epoch 8/15
303/303 [==============================] - 1s 3ms/step - loss: 0.6427 - accuracy: 0.7162
Epoch 9/15
303/303 [==============================] - 1s 3ms/step - loss: 0.6158 - accuracy: 0.7162
Epoch 10/15
303/303 [==============================] - 1s 3ms/step - loss: 0.5526 - accuracy: 0.7294
Epoch 11/15
303/303 [==============================] - 1s 3ms/step - loss: 0.5217 - accuracy: 0.7492
Epoch 12/15
303/303 [==============================] - 1s 3ms/step - loss: 0.5208 - accuracy: 0.7558
Epoch 13/15
303/303 [==============================] - 1s 3ms/step - loss: 0.5160 - accuracy: 0.7393
Epoch 14/15
303/303 [==============================] - 1s 3ms/step - loss: 0.5308 - accuracy: 0.7558
Epoch 15/15
303/303 [==============================] - 1s 3ms/step - loss: 0.4587 - accuracy: 0.7822

<tensorflow.python.keras.callbacks.History at 0x7f06b0788fd0>

Alternative to feature columns

Passing a dictionary as an input to a model is as easy as creating a matching dictionary of tf.keras.layers.Input layers, applying any pre-processing and stacking them up using the functional api. You can use this as an alternative to feature columns.

inputs = {key: tf.keras.layers.Input(shape=(), name=key) for key in df.keys()}
x = tf.stack(list(inputs.values()), axis=-1)

x = tf.keras.layers.Dense(10, activation='relu')(x)
output = tf.keras.layers.Dense(1, activation='sigmoid')(x)

model_func = tf.keras.Model(inputs=inputs, outputs=output)

model_func.compile(optimizer='adam',
                   loss='binary_crossentropy',
                   metrics=['accuracy'])

The easiest way to preserve the column structure of a pd.DataFrame when used with tf.data is to convert the pd.DataFrame to a dict, and slice that dictionary.

dict_slices = tf.data.Dataset.from_tensor_slices((df.to_dict('list'), target.values)).batch(16)
for dict_slice in dict_slices.take(1):
  print (dict_slice)
({'thal': <tf.Tensor: id=14778, shape=(16,), dtype=int32, numpy=array([2, 3, 4, 3, 3, 3, 3, 3, 4, 4, 2, 3, 2, 4, 4, 3], dtype=int32)>, 'ca': <tf.Tensor: id=14769, shape=(16,), dtype=int32, numpy=array([0, 3, 2, 0, 0, 0, 2, 0, 1, 0, 0, 0, 1, 0, 0, 0], dtype=int32)>, 'trestbps': <tf.Tensor: id=14780, shape=(16,), dtype=int32, numpy=
array([145, 160, 120, 130, 130, 120, 140, 120, 130, 140, 140, 140, 130,
       120, 172, 150], dtype=int32)>, 'restecg': <tf.Tensor: id=14775, shape=(16,), dtype=int32, numpy=array([2, 2, 2, 0, 2, 0, 2, 0, 2, 2, 0, 2, 2, 0, 0, 0], dtype=int32)>, 'slope': <tf.Tensor: id=14777, shape=(16,), dtype=int32, numpy=array([3, 2, 2, 3, 1, 1, 3, 1, 2, 3, 2, 2, 2, 1, 1, 1], dtype=int32)>, 'cp': <tf.Tensor: id=14771, shape=(16,), dtype=int32, numpy=array([1, 4, 4, 3, 2, 2, 4, 4, 4, 4, 4, 2, 3, 2, 3, 3], dtype=int32)>, 'fbs': <tf.Tensor: id=14773, shape=(16,), dtype=int32, numpy=array([1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 1, 0], dtype=int32)>, 'oldpeak': <tf.Tensor: id=14774, shape=(16,), dtype=float32, numpy=
array([2.3, 1.5, 2.6, 3.5, 1.4, 0.8, 3.6, 0.6, 1.4, 3.1, 0.4, 1.3, 0.6,
       0. , 0.5, 1.6], dtype=float32)>, 'chol': <tf.Tensor: id=14770, shape=(16,), dtype=int32, numpy=
array([233, 286, 229, 250, 204, 236, 268, 354, 254, 203, 192, 294, 256,
       263, 199, 168], dtype=int32)>, 'sex': <tf.Tensor: id=14776, shape=(16,), dtype=int32, numpy=array([1, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 1], dtype=int32)>, 'exang': <tf.Tensor: id=14772, shape=(16,), dtype=int32, numpy=array([0, 1, 1, 0, 0, 0, 0, 1, 0, 1, 0, 0, 1, 0, 0, 0], dtype=int32)>, 'age': <tf.Tensor: id=14768, shape=(16,), dtype=int32, numpy=
array([63, 67, 67, 37, 41, 56, 62, 57, 63, 53, 57, 56, 56, 44, 52, 57],
      dtype=int32)>, 'thalach': <tf.Tensor: id=14779, shape=(16,), dtype=int32, numpy=
array([150, 108, 129, 187, 172, 178, 160, 163, 147, 155, 148, 153, 142,
       173, 162, 174], dtype=int32)>}, <tf.Tensor: id=14781, shape=(16,), dtype=int64, numpy=array([0, 1, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 0])>)
model_func.fit(dict_slices, epochs=15)
Epoch 1/15
19/19 [==============================] - 1s 28ms/step - loss: 1.8827 - accuracy: 0.5677
Epoch 2/15
19/19 [==============================] - 0s 5ms/step - loss: 0.9474 - accuracy: 0.6634
Epoch 3/15
19/19 [==============================] - 0s 5ms/step - loss: 0.8700 - accuracy: 0.6898
Epoch 4/15
19/19 [==============================] - 0s 5ms/step - loss: 0.8689 - accuracy: 0.6634
Epoch 5/15
19/19 [==============================] - 0s 5ms/step - loss: 0.8240 - accuracy: 0.6799
Epoch 6/15
19/19 [==============================] - 0s 5ms/step - loss: 0.7939 - accuracy: 0.6733
Epoch 7/15
19/19 [==============================] - 0s 5ms/step - loss: 0.7711 - accuracy: 0.6766
Epoch 8/15
19/19 [==============================] - 0s 5ms/step - loss: 0.7479 - accuracy: 0.6799
Epoch 9/15
19/19 [==============================] - 0s 5ms/step - loss: 0.7244 - accuracy: 0.6964
Epoch 10/15
19/19 [==============================] - 0s 5ms/step - loss: 0.7016 - accuracy: 0.6997
Epoch 11/15
19/19 [==============================] - 0s 5ms/step - loss: 0.6800 - accuracy: 0.7129
Epoch 12/15
19/19 [==============================] - 0s 5ms/step - loss: 0.6596 - accuracy: 0.7129
Epoch 13/15
19/19 [==============================] - 0s 5ms/step - loss: 0.6405 - accuracy: 0.7162
Epoch 14/15
19/19 [==============================] - 0s 4ms/step - loss: 0.6227 - accuracy: 0.7261
Epoch 15/15
19/19 [==============================] - 0s 5ms/step - loss: 0.6062 - accuracy: 0.7261

<tensorflow.python.keras.callbacks.History at 0x7f06a055ab00>