google colaboratoryでTPUが使えないエラーについて

Question

このサイト通りにgoogle colaboratoryでTPUを使おうとしましたが、下のようなエラーがでます。

原因がわからないので教えていただけないでしょうか？

# エラー
AttributeError: 'Model' object has no attribute 'target_tensors'

# 実行コード
import tensorflow as tf
import tensorflow.keras.backend as K
import numpy as np
from tensorflow.contrib.tpu.python.tpu import keras_support
from tensorflow.keras.layers import Input, Conv2D, BatchNormalization, Activation, AveragePooling2D, Dense, Dropout, Flatten
from tensorflow.keras.datasets import cifar10
from tensorflow.keras.models import Model, Sequential
from tensorflow.keras.utils import to_categorical
import os

def basic_mlp_module(input, units):
    x = Dense(units)(input)
    x = BatchNormalization()(x)
    x = Activation("relu")(x)
    x = Dropout(0.5)(x)
    return x

def create_mlp_model():
    input = Input((32*32*3,))
    x = basic_mlp_module(input, 2048)
    x = basic_mlp_module(x, 1024)
    x = basic_mlp_module(x, 512)
    x = basic_mlp_module(x, 256)
    x = basic_mlp_module(x, 128)
    x = basic_mlp_module(x, 64)
    x = basic_mlp_module(x, 32)
    x = basic_mlp_module(x, 16)
    x = Dense(10, activation="softmax")(x)
    return Model(input, x)


K.clear_session()

# CIFAR
(X_train, y_train), (_, _) = cifar10.load_data()
X_train = (X_train / 255.0).reshape(50000, -1)
y_train = to_categorical(y_train)

# モデルを作成
model = create_mlp_model()
model.compile(tf.train.AdamOptimizer(learning_rate=1e-3), loss="categorical_crossentropy", metrics=["acc"])

# TPU
tpu_grpc_url = "grpc://"+os.environ["COLAB_TPU_ADDR"]
tpu_cluster_resolver = tf.contrib.cluster_resolver.TPUClusterResolver(tpu_grpc_url)
strategy = keras_support.TPUDistributionStrategy(tpu_cluster_resolver)
model = tf.contrib.tpu.keras_to_tpu_model(model, strategy=strategy)

# エラー全文
W0719 06:46:19.781610 140396333524864 keras_support.py:217] Keras support is now deprecated in support of TPU Strategy. Please follow the distribution strategy guide on tensorflow.org to migrate to the 2.0 supported version.
W0719 06:46:29.678592 140396333524864 keras_support.py:1394] Keras support is now deprecated in support of TPU Strategy. Please follow the distribution strategy guide on tensorflow.org to migrate to the 2.0 supported version.
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-36-4751bcfa9298> in <module>()
     45 tpu_cluster_resolver = tf.contrib.cluster_resolver.TPUClusterResolver(tpu_grpc_url)
     46 strategy = keras_support.TPUDistributionStrategy(tpu_cluster_resolver)
---> 47 model = tf.contrib.tpu.keras_to_tpu_model(model, strategy=strategy)

2 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/contrib/tpu/python/tpu/keras_support.py in __init__(self, cpu_model, strategy)
   1426           self._cpu_model.sample_weight_mode,
   1427           self._cpu_model._compile_weighted_metrics,
-> 1428           self._cpu_model.target_tensors,
   1429       )
   1430 

AttributeError: 'Model' object has no attribute 'target_tensors'

Answer 1

最新バージョンではこちらのように書き方が変更されたみたいです。

その書き方にしてみるか、!pip install tensorflow==1.13.1でバージョンを下げてみるのはどうでしょうか。

Comments

[trafalbad]

解決しました、ありがとうございます