CODEWITHSUNDEEP
pythontensorflowfeature-engineering
AlSub
AlSub

Reputation: 1155

tensorflow- how to keep column label name when one-hot-encoding?

When trying to get the label of a column in order to one hot encode it by using tensorflow:

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

# some data
d={'column1':['a', 'b', 'c', 'd'], 'column2':['e', 'f', 'g', 'h'], 'column3':[1, 2, 3, 4]}


# convert from pandas df to TensorSliceDataset
df=pd.DataFrame(d)
ds = tf.data.Dataset.from_tensor_slices(dict(df))

# convert to specific feature_column
just_types = tf.feature_column.categorical_column_with_vocabulary_list(
      'column1', ds.column1.unique())

# apply one hot encoding
type_one_hot = feature_column.indicator_column(just_types)
type_one_hot

The next error arises:

AttributeError: 'TensorSliceDataset' object has no attribute 'column1'

I know this is possible with pandas but is it possible to get a dataframe in tensorflow and then change it to pandas again in a way that kinda looks like this by using tensorflow? :

#   column1_a  column1_b  column_c  column_d
#       1           0          0        0
#       0           1          0        0
#       0           0          1        0
#       0           0          0        1

Upvotes: 1

Views: 559

Answers (1)

Damir Devetak
Damir Devetak

Reputation: 762

Using function from_tensor_slices in my opinion is an overkill in this example. Just to point out two functions

import tensorflow_datasets as tfds

df=pd.DataFrame(data)
ds = tf.data.Dataset.from_tensor_slices(dict(df))

# going back to pandas DataFrame

df_reversed = tfds.as_dataframe(ds)

And CategoryEncoding option

layer = tf.keras.layers.CategoryEncoding(num_tokens=4, output_mode="one_hot")
layer([1,2,0, 3])

Upvotes: 2

Related Questions