Tensorflow 错误“TypeError：传递给 'Pack' Op 的 'values' 的列表中的张量具有不完全匹配的类型 [int32, int64, int32, int32, int32]。”

Question

I am new to Tensorflow and Keras. I have loaded a dataset from CSV and created a train_dataset as such:

column_names = ['a', 'date', 'c', 'd', 'e', 'f']
label_name = column_names[0]
feature_names = column_names[1:]
class_names = ['good', 'bad']

train_dataset = tf.data.experimental.make_csv_dataset(
    train_dataset_fp,
    batch_size,
    column_names=column_names,
    label_name=label_name,
    num_epochs=1)

features, labels = next(iter(train_dataset))
print(features)

My features are an OrderedDict and print as:

OrderedDict([('b', <tf.Tensor: shape=(32,), dtype=int32, numpy= array([1, 1, 0, 0, 0, 1, 0, 1, 0, 2, 1, 1, 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 1], dtype=int32)>), ('date', <tf.Tensor: shape=(32,), dtype=int64, numpy= array([-9223372036855, 1262478794000, 1262426153000, 1262431717000, 1262425334000, 1262588520000, 1262425515000, 1262418072000, 1262420797000, 1262428601000, 1262590037000, 1262421322000, 1262433023000, 1262390762000, 1262590200000, 1262432769000, 1262427397000, -9223372036855, 1262425996000, 1262430050000, 1262431867000, 1262424427000, 1262420906000, 1262391208000, 1262590114000, -9223372036855, 1262589645000, 1262424306000, 1262428178000, 1262421300000, 1262423456000, 1262515569000])>), ('d', <tf.Tensor: shape=(32,), dtype=int32, numpy= array([357, 313, 557, 691, 292, 557, 605, 605, 48, 295, 81, 656, 321, 734, 584, 652, 575, 465, 71, 453, 196, 48, 689, 591, 676, 271, 67, 229, 740, 713, 230, 664], dtype=int32)>), ('e', <tf.Tensor: shape=(32,), dtype=i nt32, numpy= array([519, 537, 610, 178, 552, 610, 240, 240, 343, 643, 481, 340, 362, 143, 511, 167, 5, 685, 436, 105, 659, 343, 427, 242, 30, 717, 531, 492, 433, 452, 645, 303], dtype=int32)>), ('f', <tf.Tensor: shape=(32,), dtype=int32, numpy= array([ 345, 545, 1663, 1426, 2065, 1017, 1655, 47, 2070, -1, 1191, 191, 1569, 547, 1295, 1776, 1620, 680, 1990, 1642, 1930, 1465, 1887, 2128, 999, 447, 844, 1851, 1586, 1742, 2079, 729], dtype=int32)>)])

As you can see one of them has dtype=int64. I then use the following function to pack the features into an array:

def pack_features_vector(features, labels):
  features = tf.stack(list(features.values()), axis=1)
  return features, labels

However when I run it:

train_dataset = train_dataset.map(pack_features_vector)

I get the following error:

"TypeError: Tensors in list passed to 'values' of 'Pack' Op have types [int32, int64, int32, int32, int32] that don't all match."

I understand that the issue is the stack function. I have an epoch format date as my second feature which was read in as int64. I think it may be easiest to convert all tensors to the same dType but I am not sure how. I can see that features collection is an OrderedDict of Numpy arrays but I do not know how to change dType of the items. I tried the following, it did not yeild a traceback but when I printed my features again all dtypes were still the same:

for k,v in train_dataset:
  tf.dtypes.cast(v, tf.int64)

I would greatly appreciate any help. Thank you.

Answer 1

I think I have figured it out. I needed to add column_defaults:

column_names = ['a', 'b', 'date', 'd', 'e', 'f']
column_defaults=[tf.int64, tf.int64, tf.int64, tf.int64, tf.int64, tf.int64]
label_name = column_names[0]
feature_names = column_names[1:]
class_names = ['good', 'bad']

batch_size = 32

train_dataset = tf.data.experimental.make_csv_dataset(
    train_dataset_fp,
    batch_size,
    column_names=column_names,
    column_defaults=column_defaults,
    label_name=label_name,
    num_epochs=1)

features, labels = next(iter(train_dataset))

def pack_features_vector(features, labels):
  features = tf.stack(list(features.values()), axis=1)
  return features, labels

train_dataset = train_dataset.map(pack_features_vector)

But I am now trying to create a model and predictions:

model = tf.keras.Sequential([
  tf.keras.layers.Dense(128, activation=tf.nn.relu, input_shape=(5,)),
  tf.keras.layers.Dense(128, activation=tf.nn.relu),
  tf.keras.layers.Dense(2)
])

predictions = model(features)

And get "KeyError: 'dense_input'", not sure why.