 Ver em TensorFlow.org
|
 Executar em Google Colab
|
 Ver código fonte no GitHub
|
 Baixar notebook
|
Este tutorial fornece um exemplo de como carregar dados CSV de um arquivo em um tf.data.Dataset.
Os dados usados neste tutorial foram retirados da lista de passageiros do Titanic. O modelo preverá a probabilidade de sobrevivência de um passageiro com base em características como idade, sexo, classe de passagem e se a pessoa estava viajando sozinha.
Setup
try:
# %tensorflow_version only exists in Colab.
%tensorflow_version 2.x
except Exception:
pass
from __future__ import absolute_import, division, print_function, unicode_literals
import functools
import numpy as np
import tensorflow as tf
TRAIN_DATA_URL = "https://storage.googleapis.com/tf-datasets/titanic/train.csv"
TEST_DATA_URL = "https://storage.googleapis.com/tf-datasets/titanic/eval.csv"
train_file_path = tf.keras.utils.get_file("train.csv", TRAIN_DATA_URL)
test_file_path = tf.keras.utils.get_file("eval.csv", TEST_DATA_URL)
Downloading data from https://storage.googleapis.com/tf-datasets/titanic/train.csv 8192/30874 [======>.......................] - ETA: 0s 32768/30874 [===============================] - 0s 0us/step Downloading data from https://storage.googleapis.com/tf-datasets/titanic/eval.csv 8192/13049 [=================>............] - ETA: 0s 16384/13049 [=====================================] - 0s 0us/step
# Facilitar a leitura de valores numpy.
np.set_printoptions(precision=3, suppress=True)
Carregar dados
Para começar, vejamos a parte superior do arquivo CSV para ver como ele está formatado.
head {train_file_path}survived,sex,age,n_siblings_spouses,parch,fare,class,deck,embark_town,alone 0,male,22.0,1,0,7.25,Third,unknown,Southampton,n 1,female,38.0,1,0,71.2833,First,C,Cherbourg,n 1,female,26.0,0,0,7.925,Third,unknown,Southampton,y 1,female,35.0,1,0,53.1,First,C,Southampton,n 0,male,28.0,0,0,8.4583,Third,unknown,Queenstown,y 0,male,2.0,3,1,21.075,Third,unknown,Southampton,n 1,female,27.0,0,2,11.1333,Third,unknown,Southampton,n 1,female,14.0,1,0,30.0708,Second,unknown,Cherbourg,n 1,female,4.0,1,1,16.7,Third,G,Southampton,n
Você pode carregar isso usando pandas e passar as matrizes NumPy para o TensorFlow. Se você precisar escalar até um grande conjunto de arquivos ou precisar de um carregador que se integre ao TensorFlow e tf.data, use o tf.data.experimental. função make_csv_dataset:
A única coluna que você precisa identificar explicitamente é aquela com o valor que o modelo pretende prever.
LABEL_COLUMN = 'survived'
LABELS = [0, 1]
Now read the CSV data from the file and create a dataset.
(For the full documentation, see tf.data.experimental.make_csv_dataset)
def get_dataset(file_path, **kwargs):
dataset = tf.data.experimental.make_csv_dataset(
file_path,
batch_size=5, # Artificialmente pequeno para facilitar a exibição de exemplos
label_name=LABEL_COLUMN,
na_value="?",
num_epochs=1,
ignore_errors=True,
**kwargs)
return dataset
raw_train_data = get_dataset(train_file_path)
raw_test_data = get_dataset(test_file_path)
def show_batch(dataset):
for batch, label in dataset.take(1):
for key, value in batch.items():
print("{:20s}: {}".format(key,value.numpy()))
Cada item do conjunto de dados é um lote, representado como uma tupla de (* muitos exemplos *, * muitos rótulos *). Os dados dos exemplos são organizados em tensores baseados em colunas (em vez de tensores baseados em linhas), cada um com tantos elementos quanto o tamanho do lote (5 neste caso).
Pode ajudar a ver isso por si mesmo.
show_batch(raw_train_data)
sex : [b'male' b'male' b'female' b'male' b'female'] age : [18. 47. 28. 33. 29.] n_siblings_spouses : [0 0 8 1 0] parch : [0 0 2 1 4] fare : [ 8.3 25.587 69.55 20.525 21.075] class : [b'Third' b'First' b'Third' b'Third' b'Third'] deck : [b'unknown' b'E' b'unknown' b'unknown' b'unknown'] embark_town : [b'Southampton' b'Southampton' b'Southampton' b'Southampton' b'Southampton'] alone : [b'y' b'y' b'n' b'n' b'n']
Como você pode ver, as colunas no CSV são nomeadas. O construtor do conjunto de dados selecionará esses nomes automaticamente. Se o arquivo com o qual você está trabalhando não contém os nomes das colunas na primeira linha, passe-os em uma lista de strings para o argumento column_names na função make_csv_dataset.
CSV_COLUMNS = ['survived', 'sex', 'age', 'n_siblings_spouses', 'parch', 'fare', 'class', 'deck', 'embark_town', 'alone']
temp_dataset = get_dataset(train_file_path, column_names=CSV_COLUMNS)
show_batch(temp_dataset)
sex : [b'female' b'male' b'male' b'male' b'female'] age : [44. 28. 32.5 21. 33. ] n_siblings_spouses : [0 0 1 0 0] parch : [0 0 0 0 2] fare : [27.721 8.05 30.071 7.733 26. ] class : [b'First' b'Third' b'Second' b'Third' b'Second'] deck : [b'B' b'unknown' b'unknown' b'unknown' b'unknown'] embark_town : [b'Cherbourg' b'Southampton' b'Cherbourg' b'Queenstown' b'Southampton'] alone : [b'y' b'y' b'n' b'y' b'n']
Este exemplo vai usar todas as colunas disponíveis. Se você precisar omitir algumas colunas do conjunto de dados, crie uma lista apenas das colunas que planeja usar e passe-a para o argumento (opcional) select_columns do construtor.
SELECT_COLUMNS = ['survived', 'age', 'n_siblings_spouses', 'class', 'deck', 'alone']
temp_dataset = get_dataset(train_file_path, select_columns=SELECT_COLUMNS)
show_batch(temp_dataset)
age : [28. 23. 44. 23. 28.] n_siblings_spouses : [0 0 1 3 0] class : [b'Third' b'Third' b'Second' b'First' b'Third'] deck : [b'unknown' b'unknown' b'unknown' b'C' b'unknown'] alone : [b'y' b'y' b'n' b'n' b'y']
Pré-processamento dos Dados
Um arquivo CSV pode conter uma variedade de tipos de dados. Normalmente, você deseja converter desses tipos mistos em um vetor de comprimento fixo antes de alimentar os dados em seu modelo.
O TensorFlow possui um sistema interno para descrever conversões de entrada comuns: tf.feature_column, consulte este tutorial para detalhes.
Você pode pré-processar seus dados usando qualquer ferramenta que desejar (como nltk ou sklearn) e apenas passar a saída processada para o TensorFlow.
A principal vantagem de fazer o pré-processamento dentro do seu modelo é que, quando você exporta o modelo, ele inclui o pré-processamento. Dessa forma, você pode passar os dados brutos diretamente para o seu modelo.
Dados contínuos
Se seus dados já estiverem em um formato numérico apropriado, você poderá compactá-los em um vetor antes de transmiti-los ao modelo:
SELECT_COLUMNS = ['survived', 'age', 'n_siblings_spouses', 'parch', 'fare']
DEFAULTS = [0, 0.0, 0.0, 0.0, 0.0]
temp_dataset = get_dataset(train_file_path,
select_columns=SELECT_COLUMNS,
column_defaults = DEFAULTS)
show_batch(temp_dataset)
age : [ 4. 28. 40. 25. 35.] n_siblings_spouses : [1. 0. 1. 0. 0.] parch : [1. 0. 1. 0. 0.] fare : [ 16.7 7.55 134.5 0. 512.329]
example_batch, labels_batch = next(iter(temp_dataset))
Aqui está uma função simples que agrupará todas as colunas:
def pack(features, label):
return tf.stack(list(features.values()), axis=-1), label
Aplique isso a cada elemento do conjunto de dados:
packed_dataset = temp_dataset.map(pack)
for features, labels in packed_dataset.take(1):
print(features.numpy())
print()
print(labels.numpy())
[[22. 0. 0. 7.25] [24. 2. 3. 18.75] [28. 0. 0. 13. ] [24. 2. 0. 24.15] [46. 0. 0. 79.2 ]] [0 1 1 0 0]
Se você tiver tipos de dados mistos, poderá separar esses campos numéricos simples. A API tf.feature_column pode lidar com eles, mas isso gera alguma sobrecarga e deve ser evitado, a menos que seja realmente necessário. Volte para o conjunto de dados misto:
show_batch(raw_train_data)
sex : [b'female' b'male' b'male' b'male' b'female'] age : [22. 45. 28. 29. 14.] n_siblings_spouses : [0 0 8 0 0] parch : [0 0 2 0 0] fare : [10.517 8.05 69.55 30. 7.854] class : [b'Third' b'Third' b'Third' b'First' b'Third'] deck : [b'unknown' b'unknown' b'unknown' b'D' b'unknown'] embark_town : [b'Southampton' b'Southampton' b'Southampton' b'Southampton' b'Southampton'] alone : [b'y' b'y' b'n' b'y' b'y']
example_batch, labels_batch = next(iter(temp_dataset))
Portanto, defina um pré-processador mais geral que selecione uma lista de recursos numéricos e os agrupe em uma única coluna:
class PackNumericFeatures(object):
def __init__(self, names):
self.names = names
def __call__(self, features, labels):
numeric_features = [features.pop(name) for name in self.names]
numeric_features = [tf.cast(feat, tf.float32) for feat in numeric_features]
numeric_features = tf.stack(numeric_features, axis=-1)
features['numeric'] = numeric_features
return features, labels
NUMERIC_FEATURES = ['age','n_siblings_spouses','parch', 'fare']
packed_train_data = raw_train_data.map(
PackNumericFeatures(NUMERIC_FEATURES))
packed_test_data = raw_test_data.map(
PackNumericFeatures(NUMERIC_FEATURES))
show_batch(packed_train_data)
sex : [b'female' b'female' b'female' b'male' b'male'] class : [b'Third' b'Second' b'First' b'Third' b'Third'] deck : [b'unknown' b'unknown' b'unknown' b'unknown' b'unknown'] embark_town : [b'Southampton' b'Southampton' b'Cherbourg' b'Southampton' b'Cherbourg'] alone : [b'n' b'n' b'y' b'y' b'y'] numeric : [[ 36. 1. 0. 17.4 ] [ 29. 1. 0. 26. ] [ 30. 0. 0. 106.425] [ 55.5 0. 0. 8.05 ] [ 28. 0. 0. 7.225]]
example_batch, labels_batch = next(iter(packed_train_data))
Normalização dos dados
Dados contínuos sempre devem ser normalizados.
import pandas as pd
desc = pd.read_csv(train_file_path)[NUMERIC_FEATURES].describe()
desc
MEAN = np.array(desc.T['mean'])
STD = np.array(desc.T['std'])
def normalize_numeric_data(data, mean, std):
# Center the data
return (data-mean)/std
Agora crie uma coluna numérica. A API tf.feature_columns.numeric_column aceita um argumento normalizer_fn, que será executado em cada lote.
Ligue o MEAN e oSTD ao normalizador fn usando functools.partial
# Veja o que você acabou de criar.
normalizer = functools.partial(normalize_numeric_data, mean=MEAN, std=STD)
numeric_column = tf.feature_column.numeric_column('numeric', normalizer_fn=normalizer, shape=[len(NUMERIC_FEATURES)])
numeric_columns = [numeric_column]
numeric_column
NumericColumn(key='numeric', shape=(4,), default_value=None, dtype=tf.float32, normalizer_fn=functools.partial(<function normalize_numeric_data at 0x7f20001947b8>, mean=array([29.631, 0.545, 0.38 , 34.385]), std=array([12.512, 1.151, 0.793, 54.598])))
Ao treinar o modelo, inclua esta coluna de característica para selecionar e centralizar este bloco de dados numéricos:
example_batch['numeric']
<tf.Tensor: shape=(5, 4), dtype=float32, numpy=
array([[17. , 0. , 0. , 7.125],
[ 3. , 1. , 1. , 26. ],
[28. , 0. , 0. , 7.896],
[25. , 0. , 0. , 7.05 ],
[42. , 1. , 0. , 52. ]], dtype=float32)>
numeric_layer = tf.keras.layers.DenseFeatures(numeric_columns)
numeric_layer(example_batch).numpy()
array([[-1.01 , -0.474, -0.479, -0.499],
[-2.128, 0.395, 0.782, -0.154],
[-0.13 , -0.474, -0.479, -0.485],
[-0.37 , -0.474, -0.479, -0.501],
[ 0.989, 0.395, -0.479, 0.323]], dtype=float32)
A normalização baseada em média usada aqui requer conhecer os meios de cada coluna antes do tempo.
Dados categóricos
Algumas das colunas nos dados CSV são colunas categóricas. Ou seja, o conteúdo deve ser um dentre um conjunto limitado de opções.
Use a API tf.feature_column para criar uma coleção com uma tf.feature_column.indicator_column para cada coluna categórica.
CATEGORIES = {
'sex': ['male', 'female'],
'class' : ['First', 'Second', 'Third'],
'deck' : ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J'],
'embark_town' : ['Cherbourg', 'Southhampton', 'Queenstown'],
'alone' : ['y', 'n']
}
categorical_columns = []
for feature, vocab in CATEGORIES.items():
cat_col = tf.feature_column.categorical_column_with_vocabulary_list(
key=feature, vocabulary_list=vocab)
categorical_columns.append(tf.feature_column.indicator_column(cat_col))
# Veja o que você acabou de criar.
categorical_columns
[IndicatorColumn(categorical_column=VocabularyListCategoricalColumn(key='sex', vocabulary_list=('male', 'female'), dtype=tf.string, default_value=-1, num_oov_buckets=0)),
IndicatorColumn(categorical_column=VocabularyListCategoricalColumn(key='class', vocabulary_list=('First', 'Second', 'Third'), dtype=tf.string, default_value=-1, num_oov_buckets=0)),
IndicatorColumn(categorical_column=VocabularyListCategoricalColumn(key='deck', vocabulary_list=('A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J'), dtype=tf.string, default_value=-1, num_oov_buckets=0)),
IndicatorColumn(categorical_column=VocabularyListCategoricalColumn(key='embark_town', vocabulary_list=('Cherbourg', 'Southhampton', 'Queenstown'), dtype=tf.string, default_value=-1, num_oov_buckets=0)),
IndicatorColumn(categorical_column=VocabularyListCategoricalColumn(key='alone', vocabulary_list=('y', 'n'), dtype=tf.string, default_value=-1, num_oov_buckets=0))]
categorical_layer = tf.keras.layers.DenseFeatures(categorical_columns)
print(categorical_layer(example_batch).numpy()[0])
[1. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0.]
Isso fará parte de uma entrada de processamento de dados posteriormente, quando você construir o modelo.
Camada combinada de pré-processamento
Adicione as duas coleções de colunas de recursos e passe-as para um tf.keras.layers.DenseFeatures para criar uma camada de entrada que extrairá e pré-processará os dois tipos de entrada:
preprocessing_layer = tf.keras.layers.DenseFeatures(categorical_columns+numeric_columns)
print(preprocessing_layer(example_batch).numpy()[0])
[ 1. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. -1.01 -0.474 -0.479 -0.499 1. 0. ]
Construir o modelo
Crie um tf.keras.Sequential, começando com o preprocessing_layer.
model = tf.keras.Sequential([
preprocessing_layer,
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(1),
])
model.compile(
loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
optimizer='adam',
metrics=['accuracy'])
Treinar, avaliar, e prever
Agora o modelo pode ser instanciado e treinado.
train_data = packed_train_data.shuffle(500)
test_data = packed_test_data
model.fit(train_data, epochs=20)
Epoch 1/20
WARNING:tensorflow:Layers in a Sequential model should only have a single input tensor, but we receive a <class 'collections.OrderedDict'> input: OrderedDict([('sex', <tf.Tensor 'ExpandDims_4:0' shape=(None, 1) dtype=string>), ('class', <tf.Tensor 'ExpandDims_1:0' shape=(None, 1) dtype=string>), ('deck', <tf.Tensor 'ExpandDims_2:0' shape=(None, 1) dtype=string>), ('embark_town', <tf.Tensor 'ExpandDims_3:0' shape=(None, 1) dtype=string>), ('alone', <tf.Tensor 'ExpandDims:0' shape=(None, 1) dtype=string>), ('numeric', <tf.Tensor 'IteratorGetNext:4' shape=(None, 4) dtype=float32>)])
Consider rewriting this model with the Functional API.
WARNING:tensorflow:Layers in a Sequential model should only have a single input tensor, but we receive a <class 'collections.OrderedDict'> input: OrderedDict([('sex', <tf.Tensor 'ExpandDims_4:0' shape=(None, 1) dtype=string>), ('class', <tf.Tensor 'ExpandDims_1:0' shape=(None, 1) dtype=string>), ('deck', <tf.Tensor 'ExpandDims_2:0' shape=(None, 1) dtype=string>), ('embark_town', <tf.Tensor 'ExpandDims_3:0' shape=(None, 1) dtype=string>), ('alone', <tf.Tensor 'ExpandDims:0' shape=(None, 1) dtype=string>), ('numeric', <tf.Tensor 'IteratorGetNext:4' shape=(None, 4) dtype=float32>)])
Consider rewriting this model with the Functional API.
1/Unknown - 0s 455us/step - loss: 0.7093 - accuracy: 0.6000
18/Unknown - 0s 3ms/step - loss: 0.6065 - accuracy: 0.6444
36/Unknown - 0s 3ms/step - loss: 0.6114 - accuracy: 0.6444
53/Unknown - 0s 3ms/step - loss: 0.5786 - accuracy: 0.6679
70/Unknown - 0s 3ms/step - loss: 0.5628 - accuracy: 0.6974
88/Unknown - 0s 3ms/step - loss: 0.5413 - accuracy: 0.7094
106/Unknown - 0s 3ms/step - loss: 0.5194 - accuracy: 0.7287
123/Unknown - 0s 3ms/step - loss: 0.5166 - accuracy: 0.7402
126/126 [==============================] - 0s 3ms/step - loss: 0.5113 - accuracy: 0.7400
Epoch 2/20
1/126 [..............................] - ETA: 0s - loss: 0.4266 - accuracy: 0.6000
19/126 [===>..........................] - ETA: 0s - loss: 0.4363 - accuracy: 0.7895
37/126 [=======>......................] - ETA: 0s - loss: 0.4753 - accuracy: 0.7730
55/126 [============>.................] - ETA: 0s - loss: 0.4560 - accuracy: 0.7927
73/126 [================>.............] - ETA: 0s - loss: 0.4480 - accuracy: 0.7983
91/126 [====================>.........] - ETA: 0s - loss: 0.4241 - accuracy: 0.8031
109/126 [========================>.....] - ETA: 0s - loss: 0.4244 - accuracy: 0.8044
126/126 [==============================] - 0s 3ms/step - loss: 0.4204 - accuracy: 0.8102
Epoch 3/20
1/126 [..............................] - ETA: 0s - loss: 0.1026 - accuracy: 1.0000
19/126 [===>..........................] - ETA: 0s - loss: 0.3695 - accuracy: 0.8526
37/126 [=======>......................] - ETA: 0s - loss: 0.4410 - accuracy: 0.8187
55/126 [============>.................] - ETA: 0s - loss: 0.4198 - accuracy: 0.8346
72/126 [================>.............] - ETA: 0s - loss: 0.4083 - accuracy: 0.8347
90/126 [====================>.........] - ETA: 0s - loss: 0.3929 - accuracy: 0.8345
108/126 [========================>.....] - ETA: 0s - loss: 0.4214 - accuracy: 0.8194
126/126 [==============================] - ETA: 0s - loss: 0.4088 - accuracy: 0.8246
126/126 [==============================] - 0s 3ms/step - loss: 0.4088 - accuracy: 0.8246
Epoch 4/20
1/126 [..............................] - ETA: 0s - loss: 0.6042 - accuracy: 0.8000
19/126 [===>..........................] - ETA: 0s - loss: 0.3795 - accuracy: 0.8421
37/126 [=======>......................] - ETA: 0s - loss: 0.3670 - accuracy: 0.8324
55/126 [============>.................] - ETA: 0s - loss: 0.3764 - accuracy: 0.8327
74/126 [================>.............] - ETA: 0s - loss: 0.3706 - accuracy: 0.8447
92/126 [====================>.........] - ETA: 0s - loss: 0.3911 - accuracy: 0.8293
110/126 [=========================>....] - ETA: 0s - loss: 0.3898 - accuracy: 0.8282
126/126 [==============================] - 0s 3ms/step - loss: 0.3902 - accuracy: 0.8309
Epoch 5/20
1/126 [..............................] - ETA: 0s - loss: 0.4116 - accuracy: 0.8000
19/126 [===>..........................] - ETA: 0s - loss: 0.4069 - accuracy: 0.7895
37/126 [=======>......................] - ETA: 0s - loss: 0.4074 - accuracy: 0.7892
55/126 [============>.................] - ETA: 0s - loss: 0.3873 - accuracy: 0.8182
73/126 [================>.............] - ETA: 0s - loss: 0.3943 - accuracy: 0.8149
91/126 [====================>.........] - ETA: 0s - loss: 0.3767 - accuracy: 0.8296
109/126 [========================>.....] - ETA: 0s - loss: 0.3832 - accuracy: 0.8321
126/126 [==============================] - 0s 3ms/step - loss: 0.3828 - accuracy: 0.8357
Epoch 6/20
1/126 [..............................] - ETA: 0s - loss: 0.2257 - accuracy: 0.8000
19/126 [===>..........................] - ETA: 0s - loss: 0.3763 - accuracy: 0.8211
37/126 [=======>......................] - ETA: 0s - loss: 0.3512 - accuracy: 0.8462
55/126 [============>.................] - ETA: 0s - loss: 0.3359 - accuracy: 0.8566
72/126 [================>.............] - ETA: 0s - loss: 0.3552 - accuracy: 0.8459
90/126 [====================>.........] - ETA: 0s - loss: 0.3533 - accuracy: 0.8389
108/126 [========================>.....] - ETA: 0s - loss: 0.3682 - accuracy: 0.8399
126/126 [==============================] - ETA: 0s - loss: 0.3705 - accuracy: 0.8405
126/126 [==============================] - 0s 3ms/step - loss: 0.3705 - accuracy: 0.8405
Epoch 7/20
1/126 [..............................] - ETA: 0s - loss: 0.0615 - accuracy: 1.0000
19/126 [===>..........................] - ETA: 0s - loss: 0.3637 - accuracy: 0.8632
37/126 [=======>......................] - ETA: 0s - loss: 0.3762 - accuracy: 0.8432
55/126 [============>.................] - ETA: 0s - loss: 0.3724 - accuracy: 0.8364
73/126 [================>.............] - ETA: 0s - loss: 0.3619 - accuracy: 0.8425
91/126 [====================>.........] - ETA: 0s - loss: 0.3739 - accuracy: 0.8429
109/126 [========================>.....] - ETA: 0s - loss: 0.3630 - accuracy: 0.8561
126/126 [==============================] - 0s 3ms/step - loss: 0.3695 - accuracy: 0.8517
Epoch 8/20
1/126 [..............................] - ETA: 0s - loss: 0.2391 - accuracy: 1.0000
18/126 [===>..........................] - ETA: 0s - loss: 0.3660 - accuracy: 0.8444
36/126 [=======>......................] - ETA: 0s - loss: 0.3438 - accuracy: 0.8556
54/126 [===========>..................] - ETA: 0s - loss: 0.3869 - accuracy: 0.8333
72/126 [================>.............] - ETA: 0s - loss: 0.3717 - accuracy: 0.8333
89/126 [====================>.........] - ETA: 0s - loss: 0.3813 - accuracy: 0.8315
107/126 [========================>.....] - ETA: 0s - loss: 0.3610 - accuracy: 0.8374
125/126 [============================>.] - ETA: 0s - loss: 0.3557 - accuracy: 0.8441
126/126 [==============================] - 0s 3ms/step - loss: 0.3575 - accuracy: 0.8437
Epoch 9/20
1/126 [..............................] - ETA: 0s - loss: 0.3238 - accuracy: 0.8000
19/126 [===>..........................] - ETA: 0s - loss: 0.3500 - accuracy: 0.8211
37/126 [=======>......................] - ETA: 0s - loss: 0.3031 - accuracy: 0.8681
55/126 [============>.................] - ETA: 0s - loss: 0.3029 - accuracy: 0.8640
73/126 [================>.............] - ETA: 0s - loss: 0.3231 - accuracy: 0.8591
91/126 [====================>.........] - ETA: 0s - loss: 0.3480 - accuracy: 0.8451
109/126 [========================>.....] - ETA: 0s - loss: 0.3603 - accuracy: 0.8432
126/126 [==============================] - 0s 3ms/step - loss: 0.3656 - accuracy: 0.8437
Epoch 10/20
1/126 [..............................] - ETA: 0s - loss: 0.4949 - accuracy: 0.8000
19/126 [===>..........................] - ETA: 0s - loss: 0.4774 - accuracy: 0.7579
38/126 [========>.....................] - ETA: 0s - loss: 0.3723 - accuracy: 0.8211
56/126 [============>.................] - ETA: 0s - loss: 0.3826 - accuracy: 0.8214
74/126 [================>.............] - ETA: 0s - loss: 0.3744 - accuracy: 0.8338
92/126 [====================>.........] - ETA: 0s - loss: 0.3426 - accuracy: 0.8556
110/126 [=========================>....] - ETA: 0s - loss: 0.3286 - accuracy: 0.8592
126/126 [==============================] - 0s 3ms/step - loss: 0.3425 - accuracy: 0.8517
Epoch 11/20
1/126 [..............................] - ETA: 0s - loss: 0.1793 - accuracy: 1.0000
19/126 [===>..........................] - ETA: 0s - loss: 0.2845 - accuracy: 0.8947
37/126 [=======>......................] - ETA: 0s - loss: 0.3418 - accuracy: 0.8595
55/126 [============>.................] - ETA: 0s - loss: 0.3706 - accuracy: 0.8400
73/126 [================>.............] - ETA: 0s - loss: 0.3822 - accuracy: 0.8370
91/126 [====================>.........] - ETA: 0s - loss: 0.3716 - accuracy: 0.8429
110/126 [=========================>....] - ETA: 0s - loss: 0.3545 - accuracy: 0.8501
126/126 [==============================] - 0s 3ms/step - loss: 0.3444 - accuracy: 0.8501
Epoch 12/20
1/126 [..............................] - ETA: 0s - loss: 0.7250 - accuracy: 0.8000
19/126 [===>..........................] - ETA: 0s - loss: 0.3498 - accuracy: 0.8842
37/126 [=======>......................] - ETA: 0s - loss: 0.3590 - accuracy: 0.8811
55/126 [============>.................] - ETA: 0s - loss: 0.3372 - accuracy: 0.8824
73/126 [================>.............] - ETA: 0s - loss: 0.3357 - accuracy: 0.8702
91/126 [====================>.........] - ETA: 0s - loss: 0.3272 - accuracy: 0.8695
109/126 [========================>.....] - ETA: 0s - loss: 0.3448 - accuracy: 0.8524
126/126 [==============================] - 0s 3ms/step - loss: 0.3461 - accuracy: 0.8517
Epoch 13/20
1/126 [..............................] - ETA: 0s - loss: 0.0474 - accuracy: 1.0000
18/126 [===>..........................] - ETA: 0s - loss: 0.3032 - accuracy: 0.8667
36/126 [=======>......................] - ETA: 0s - loss: 0.3609 - accuracy: 0.8222
53/126 [===========>..................] - ETA: 0s - loss: 0.3759 - accuracy: 0.8151
71/126 [===============>..............] - ETA: 0s - loss: 0.3589 - accuracy: 0.8338
89/126 [====================>.........] - ETA: 0s - loss: 0.3444 - accuracy: 0.8462
107/126 [========================>.....] - ETA: 0s - loss: 0.3361 - accuracy: 0.8496
125/126 [============================>.] - ETA: 0s - loss: 0.3289 - accuracy: 0.8569
126/126 [==============================] - 0s 3ms/step - loss: 0.3269 - accuracy: 0.8581
Epoch 14/20
1/126 [..............................] - ETA: 0s - loss: 0.1820 - accuracy: 1.0000
19/126 [===>..........................] - ETA: 0s - loss: 0.3688 - accuracy: 0.8211
37/126 [=======>......................] - ETA: 0s - loss: 0.3120 - accuracy: 0.8649
55/126 [============>.................] - ETA: 0s - loss: 0.3140 - accuracy: 0.8655
73/126 [================>.............] - ETA: 0s - loss: 0.2854 - accuracy: 0.8712
91/126 [====================>.........] - ETA: 0s - loss: 0.3195 - accuracy: 0.8593
109/126 [========================>.....] - ETA: 0s - loss: 0.3266 - accuracy: 0.8606
126/126 [==============================] - 0s 3ms/step - loss: 0.3354 - accuracy: 0.8501
Epoch 15/20
1/126 [..............................] - ETA: 0s - loss: 0.2959 - accuracy: 1.0000
18/126 [===>..........................] - ETA: 0s - loss: 0.2620 - accuracy: 0.8889
36/126 [=======>......................] - ETA: 0s - loss: 0.2754 - accuracy: 0.8701
53/126 [===========>..................] - ETA: 0s - loss: 0.3063 - accuracy: 0.8664
71/126 [===============>..............] - ETA: 0s - loss: 0.3223 - accuracy: 0.8665
89/126 [====================>.........] - ETA: 0s - loss: 0.3228 - accuracy: 0.8620
107/126 [========================>.....] - ETA: 0s - loss: 0.3273 - accuracy: 0.8571
125/126 [============================>.] - ETA: 0s - loss: 0.3211 - accuracy: 0.8666
126/126 [==============================] - 0s 3ms/step - loss: 0.3225 - accuracy: 0.8660
Epoch 16/20
1/126 [..............................] - ETA: 0s - loss: 0.0785 - accuracy: 1.0000
19/126 [===>..........................] - ETA: 0s - loss: 0.2118 - accuracy: 0.8842
37/126 [=======>......................] - ETA: 0s - loss: 0.2635 - accuracy: 0.8791
55/126 [============>.................] - ETA: 0s - loss: 0.2943 - accuracy: 0.8603
73/126 [================>.............] - ETA: 0s - loss: 0.2800 - accuracy: 0.8729
91/126 [====================>.........] - ETA: 0s - loss: 0.2857 - accuracy: 0.8695
109/126 [========================>.....] - ETA: 0s - loss: 0.3004 - accuracy: 0.8635
126/126 [==============================] - 0s 3ms/step - loss: 0.3213 - accuracy: 0.8533
Epoch 17/20
1/126 [..............................] - ETA: 0s - loss: 0.1402 - accuracy: 1.0000
19/126 [===>..........................] - ETA: 0s - loss: 0.2376 - accuracy: 0.9158
37/126 [=======>......................] - ETA: 0s - loss: 0.2356 - accuracy: 0.9121
55/126 [============>.................] - ETA: 0s - loss: 0.2688 - accuracy: 0.8897
73/126 [================>.............] - ETA: 0s - loss: 0.2915 - accuracy: 0.8840
90/126 [====================>.........] - ETA: 0s - loss: 0.3129 - accuracy: 0.8770
107/126 [========================>.....] - ETA: 0s - loss: 0.3185 - accuracy: 0.8684
124/126 [============================>.] - ETA: 0s - loss: 0.3185 - accuracy: 0.8671
126/126 [==============================] - 0s 3ms/step - loss: 0.3193 - accuracy: 0.8628
Epoch 18/20
1/126 [..............................] - ETA: 0s - loss: 0.0973 - accuracy: 1.0000
18/126 [===>..........................] - ETA: 0s - loss: 0.3765 - accuracy: 0.8391
36/126 [=======>......................] - ETA: 0s - loss: 0.3184 - accuracy: 0.8531
53/126 [===========>..................] - ETA: 0s - loss: 0.3120 - accuracy: 0.8702
70/126 [===============>..............] - ETA: 0s - loss: 0.3058 - accuracy: 0.8732
87/126 [===================>..........] - ETA: 0s - loss: 0.3237 - accuracy: 0.8588
105/126 [========================>.....] - ETA: 0s - loss: 0.3102 - accuracy: 0.8678
122/126 [============================>.] - ETA: 0s - loss: 0.3132 - accuracy: 0.8633
126/126 [==============================] - 0s 3ms/step - loss: 0.3155 - accuracy: 0.8612
Epoch 19/20
1/126 [..............................] - ETA: 0s - loss: 0.0096 - accuracy: 1.0000
19/126 [===>..........................] - ETA: 0s - loss: 0.2066 - accuracy: 0.8804
36/126 [=======>......................] - ETA: 0s - loss: 0.2172 - accuracy: 0.8870
54/126 [===========>..................] - ETA: 0s - loss: 0.2529 - accuracy: 0.8839
71/126 [===============>..............] - ETA: 0s - loss: 0.2536 - accuracy: 0.8920
88/126 [===================>..........] - ETA: 0s - loss: 0.2970 - accuracy: 0.8696
106/126 [========================>.....] - ETA: 0s - loss: 0.3064 - accuracy: 0.8672
123/126 [============================>.] - ETA: 0s - loss: 0.3122 - accuracy: 0.8644
126/126 [==============================] - 0s 3ms/step - loss: 0.3075 - accuracy: 0.8676
Epoch 20/20
1/126 [..............................] - ETA: 0s - loss: 0.1408 - accuracy: 1.0000
19/126 [===>..........................] - ETA: 0s - loss: 0.2732 - accuracy: 0.8421
36/126 [=======>......................] - ETA: 0s - loss: 0.2767 - accuracy: 0.8667
53/126 [===========>..................] - ETA: 0s - loss: 0.2784 - accuracy: 0.8792
70/126 [===============>..............] - ETA: 0s - loss: 0.2585 - accuracy: 0.8886
88/126 [===================>..........] - ETA: 0s - loss: 0.2745 - accuracy: 0.8750
105/126 [========================>.....] - ETA: 0s - loss: 0.3010 - accuracy: 0.8602
122/126 [============================>.] - ETA: 0s - loss: 0.3041 - accuracy: 0.8616
126/126 [==============================] - 0s 3ms/step - loss: 0.3061 - accuracy: 0.8596
<tensorflow.python.keras.callbacks.History at 0x7f20001bdef0>
Depois que o modelo é treinado, você pode verificar sua acurácia no conjunto test_data.
test_loss, test_accuracy = model.evaluate(test_data)
print('\n\nTest Loss {}, Test Accuracy {}'.format(test_loss, test_accuracy))
WARNING:tensorflow:Layers in a Sequential model should only have a single input tensor, but we receive a <class 'collections.OrderedDict'> input: OrderedDict([('sex', <tf.Tensor 'ExpandDims_4:0' shape=(None, 1) dtype=string>), ('class', <tf.Tensor 'ExpandDims_1:0' shape=(None, 1) dtype=string>), ('deck', <tf.Tensor 'ExpandDims_2:0' shape=(None, 1) dtype=string>), ('embark_town', <tf.Tensor 'ExpandDims_3:0' shape=(None, 1) dtype=string>), ('alone', <tf.Tensor 'ExpandDims:0' shape=(None, 1) dtype=string>), ('numeric', <tf.Tensor 'IteratorGetNext:4' shape=(None, 4) dtype=float32>)])
Consider rewriting this model with the Functional API.
1/Unknown - 0s 325us/step - loss: 0.1994 - accuracy: 1.0000
21/Unknown - 0s 2ms/step - loss: 0.4007 - accuracy: 0.8762
41/Unknown - 0s 2ms/step - loss: 0.4847 - accuracy: 0.8390
53/53 [==============================] - 0s 3ms/step - loss: 0.4557 - accuracy: 0.8447
Test Loss 0.45572778582572937, Test Accuracy 0.8446969985961914
Use tf.keras.Model.predict para inferir rótulos em um lote ou em um conjunto de dados de lotes.
predictions = model.predict(test_data)
# Mostrar alguns resultados
for prediction, survived in zip(predictions[:10], list(test_data)[0][1][:10]):
print("Predicted survival: {:.2%}".format(prediction[0]),
" | Actual outcome: ",
("SURVIVED" if bool(survived) else "DIED"))
WARNING:tensorflow:Layers in a Sequential model should only have a single input tensor, but we receive a <class 'collections.OrderedDict'> input: OrderedDict([('sex', <tf.Tensor 'ExpandDims_4:0' shape=(None, 1) dtype=string>), ('class', <tf.Tensor 'ExpandDims_1:0' shape=(None, 1) dtype=string>), ('deck', <tf.Tensor 'ExpandDims_2:0' shape=(None, 1) dtype=string>), ('embark_town', <tf.Tensor 'ExpandDims_3:0' shape=(None, 1) dtype=string>), ('alone', <tf.Tensor 'ExpandDims:0' shape=(None, 1) dtype=string>), ('numeric', <tf.Tensor 'IteratorGetNext:4' shape=(None, 4) dtype=float32>)])
Consider rewriting this model with the Functional API.
Predicted survival: 598.16% | Actual outcome: SURVIVED
Predicted survival: 223.96% | Actual outcome: SURVIVED
Predicted survival: 295.53% | Actual outcome: DIED
Predicted survival: 493.27% | Actual outcome: DIED
Predicted survival: -175.62% | Actual outcome: SURVIVED