CODEWITHSUNDEEP
tensorflowkerasauto-keras
Talha Anwar
Talha Anwar

Reputation: 2949

AutoKeras: TypeError: '<' not supported between instances of 'NoneType' and 'int'

Getting error in Autkeras model, where as same data work in keras model

Dataset example

image                  label
train/class0/3.jpg      0
train/class1/2.jpg      1
train/class1/6.jpg      1
train/class1/4.jpg      1
train/class0/7.jpg      0

load function

def load(image_path,label):
  img = tf.io.read_file(image_path)
  img = tf.image.decode_jpeg(img, channels=3)
  #img = tf.image.convert_image_dtype(img, tf.float32)
  img = tf.cast(img, tf.float32) / 255.0
  label = tf.cast(label, tf.int32)
  return img, label

load data

 bs=2
 train_ds = tf.data.Dataset.from_tensor_slices((train_df.image,train_df.label)).map(load).batch(bs)

keras model

model = tf.keras.Sequential([
  layers.InputLayer((224,224,3)),                         
  layers.Conv2D(32, 3, activation='relu'),
  layers.MaxPooling2D(),
  layers.Conv2D(32, 3, activation='relu'),
  layers.MaxPooling2D(),
  layers.Conv2D(32, 3, activation='relu'),
  layers.MaxPooling2D(),
  layers.Flatten(),
  layers.Dense(128, activation='relu'),
  layers.Dense(1,activation='sigmoid')
])

train keras model

model.compile(optimizer='adam',loss=tf.losses.BinaryCrossentropy(),metrics=['accuracy'])
model.fit(train_ds,epochs=1)

AutoKeras model

import autokeras as ak
clf = ak.ImageClassifier(overwrite=False, max_trials=1)
clf.fit(train_ds, epochs=1)

Error log in autokeras model

Traceback (most recent call last):
  File "/usr/local/lib/python3.7/dist-packages/kerastuner/engine/hypermodel.py", line 104, in build
    model = self.hypermodel.build(hp)
  File "/usr/local/lib/python3.7/dist-packages/kerastuner/engine/hypermodel.py", line 64, in _build_wrapper
    return self._build(hp, *args, **kwargs)
  File "/usr/local/lib/python3.7/dist-packages/autokeras/graph.py", line 250, in build
    outputs = block.build(hp, inputs=temp_inputs)
  File "/usr/local/lib/python3.7/dist-packages/autokeras/engine/block.py", line 38, in _build_wrapper
    return super()._build_wrapper(hp, *args, **kwargs)
  File "/usr/local/lib/python3.7/dist-packages/kerastuner/engine/hypermodel.py", line 64, in _build_wrapper
    return self._build(hp, *args, **kwargs)
  File "/usr/local/lib/python3.7/dist-packages/autokeras/blocks/wrapper.py", line 108, in build
    output_node = self._build_block(hp, output_node, block_type)
  File "/usr/local/lib/python3.7/dist-packages/autokeras/blocks/wrapper.py", line 77, in _build_block
    return basic.ResNetBlock().build(hp, output_node)
  File "/usr/local/lib/python3.7/dist-packages/autokeras/engine/block.py", line 38, in _build_wrapper
    return super()._build_wrapper(hp, *args, **kwargs)
  File "/usr/local/lib/python3.7/dist-packages/kerastuner/engine/hypermodel.py", line 64, in _build_wrapper
    return self._build(hp, *args, **kwargs)
  File "/usr/local/lib/python3.7/dist-packages/autokeras/blocks/basic.py", line 688, in build
    if input_node.shape[1] < min_size or input_node.shape[2] < min_size:
TypeError: '<' not supported between instances of 'NoneType' and 'int'

Upvotes: 1

Views: 779

Answers (1)

Frut Dzentready
Frut Dzentready

Reputation: 132

Too short to be real answer but. If you chek train_ds.element_spec you will see something like [None,None,3] or even [None,None,None]. In first case only thing you need to do is to resize image with img = tf.image.resize(img, [128, 128]) In second case you allso need to set channels on decode_<img> (i see that you allready set this)

Example used with loading from disk (tested to work with autokeras)

def process_path(filename):
    parts = tf.strings.split(filename, os.sep)
    label = int(parts[-2])

    image = tf.io.read_file(filename)
    image = tf.io.decode_png(image, channels=3)
    image = tf.image.convert_image_dtype(image, tf.float32)
    image = tf.image.resize(image, [256, 256])
    return image, label

import pathlib
load_from = pathlib.Path(load_from)
train_files_ds = tf.data.Dataset.list_files(str(load_from / 'train' / '*/*.png'))
train_ds: tf.data.Dataset = train_files_ds.map(process_path)

import autokeras as ak
clf = ak.ImageClassifier(overwrite=False, max_trials=1)
clf.fit(train_ds, epochs=1)

Upvotes: 0

Related Questions