tensorflow (Python 3.7) h5 model not loading

[voqueric]

Package name: tensorflow
Link to piwheels page: https://www.piwheels.org/project/tensorflow
Version: 1.13.0
Python version: 3.7.3
More information: Downloaded, compiled, installed separate from OS package manager on Raspbian Stretch using RPI0W

Prior to recent tensorflow 1.13.0 package update the dependencies were:

h5py==2.9.0
Keras-Applications==1.0.7
Keras-Preprocessing==1.0.9
tensorboard==1.13.1
tensorflow==1.13.1
tensorflow-estimator==1.13.0

After the update the dependencies are:

h5py==2.9.0
Keras-Applications==1.0.8
Keras-Preprocessing==1.1.0
tensorboard==1.13.1
tensorflow==1.13.1
tensorflow-estimator==1.14.0rc1

I build an h5 model file using tensorflow 1.13.0 on a PC running AMD64 Kubuntu 18.04 and a separately downloaded, compiled, installed Python 3.7.3, dependencies are same as second list (above):

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.callbacks import Callback, ModelCheckpoint
from tensorflow.keras import optimizers as ko

    class EarlyStoppingByVal(Callback):
        def __init__(self, monitor='loss', value=0.0001, verbose=0):
            super(Callback, self).__init__()
            self.monitor = monitor
            self.value = value
            self.verbose = verbose

        def on_epoch_end(self, epoch, logs={}):
            current = logs.get(self.monitor)
            if current is None:
                print("Early stopping requires %s available!" % self.monitor, RuntimeWarning)

            if self.monitor == 'loss':
                if current < self.value:
                    if self.verbose > 0:
                        print("Epoch %05d: early stopping THR" % epoch)
                    self.model.stop_training = True

            if self.monitor == 'acc':
                if current >= self.value:
                    if self.verbose > 0:
                        print("Epoch %05d: early stopping THR" % epoch)
                    self.model.stop_training = True

    model = keras.Sequential([
        keras.layers.Flatten(
            input_shape=(train_data.shape[1], train_data.shape[2])
        ),
        keras.layers.Dense(
            250,
            activation=tf.nn.relu
        ),
        keras.layers.Dense(
            500,
            activation=tf.nn.relu
        ),
        keras.layers.Dense(
            1000,
            activation=tf.nn.relu
        ),
        keras.layers.Dense(
            500,
            activation=tf.nn.relu
        ),
        keras.layers.Dense(
            250,
            activation=tf.nn.relu
        ),
        keras.layers.Dense(
            250,
            activation=tf.nn.softmax
        )
    ])

    model.compile(
        optimizer=ko.Adamax(),
        loss='sparse_categorical_crossentropy',
        metrics=['accuracy']
    )

    model.fit(
        train_data,
        train_labels,
        epochs=150,
        shuffle=True,
        verbose=1,
        callbacks=callbacks
    )
    model.summary()
    model.save(train_mdl_file)

At this point the model is transferred to RPI0W and the following code is executed with tensorflow 1.13.0 also using the dependencies in the second list:

import tensorflow as tf

    tf_mdl = tf.keras.models.load_model(data_url_dict['mdl'])
    tf_preds = tf_mdl.predict(x=img_data_np)

The RPI0W returns the following warnings/errors:

WARNING: Logging before flag parsing goes to stderr.
W0618 13:47:20.288138 3069616128 deprecation_wrapper.py:118] From /usr/local/opt/python-3.7.3/lib/python3.7/site-packages/tensorflow/__init__.py:98: The name tf.AUTO_REUSE is deprecated. Please use tf.compat.v1.AUTO_REUSE instead.

W0618 13:47:20.295286 3069616128 deprecation_wrapper.py:118] From /usr/local/opt/python-3.7.3/lib/python3.7/site-packages/tensorflow/__init__.py:98: The name tf.AttrValue is deprecated. Please use tf.compat.v1.AttrValue instead.

W0618 13:47:20.300013 3069616128 deprecation_wrapper.py:118] From /usr/local/opt/python-3.7.3/lib/python3.7/site-packages/tensorflow/__init__.py:98: The name tf.COMPILER_VERSION is deprecated. Please use tf.version.COMPILER_VERSION instead.

W0618 13:47:20.304398 3069616128 deprecation_wrapper.py:118] From /usr/local/opt/python-3.7.3/lib/python3.7/site-packages/tensorflow/__init__.py:98: The name tf.CXX11_ABI_FLAG is deprecated. Please use tf.sysconfig.CXX11_ABI_FLAG instead.

W0618 13:47:20.309751 3069616128 deprecation_wrapper.py:118] From /usr/local/opt/python-3.7.3/lib/python3.7/site-packages/tensorflow/__init__.py:98: The name tf.ConditionalAccumulator is deprecated. Please use tf.compat.v1.ConditionalAccumulator instead.

SavedModel file does not exist at: /path/to/model/model.h5/{saved_model.pbtxt|saved_model.pb}

When I attempt to load the model on the PC using the same code, no errors are returned and predictions take place. Based on this outcome I am assuming that there is something different in the cp37 packages built for AMD64 and ARM, but unsure if it is the tensorflow package itself or one of its dependencies.

[bennuttall]

This is an upstream issue. Ping @petewarden

[leematthewshome]

I have just built a new image on Raspbian (RPI3B+) and installed tensorflow from tensorflow-1.13.1-cp37-none-linux-armv71.whl.

I am getting exactly the same errors about deprecated names when I import tensor flow into python3.7.

[leematthewshome]

Errr, sorry but what do you mean by "This is an upstream issue"?

[bennuttall]

"upstream" means the maintainer. File an issue with them: https://github.com/tensorflow/tensorflow/issues

[leematthewshome]

got it. thanks. I am finding that my model loads but python regularly locks up while processing.

thanks

[bennuttall]

Ok, it's just not something we can fix here. It's an issue with the source code of the package itself, not the piwheels project.