【发布时间】：2021-05-06 00:32:03
【问题描述】：
我有一个简单的 keras 模型，它由两个相同的预训练模型 (EfficientB2) 组成。
当我想保存整个模型（具有优化器状态的权重）时，会出现以下错误：
Found 964 validated image filenames belonging to 2 classes.
Found 964 validated image filenames belonging to 2 classes.
Epoch 1/2
120/120 [==============================] - ETA: 0s - loss: 13.4150 - accuracy: 0.5128Found 297 validated image filenames belonging to 2 classes.
Found 297 validated image filenames belonging to 2 classes.
120/120 [==============================] - 90s 466ms/step - loss: 13.4148 - accuracy: 0.5127 - val_loss: 13.7815 - val_accuracy: 0.4626
Epoch 00001: saving model to /content/drive/My Drive/web_crawling/weightst-01-0.4626-13.7815.hdf5
---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
<ipython-input-132-adc5fc37359d> in <module>()
1 # model.fit([x_train1,x_train2],y_train,batch_size=4,epochs=10,validation_split=0.1,shuffle=True,callbacks=callbacks_list)
----> 2 model.fit(train_generator,epochs=2,steps_per_epoch = tr_sample // batch_size, validation_data = validation_generator,validation_steps = val_sample // batch_size,callbacks=callbacks_list)#,class_weight=class_weight)
9 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_batch_size, validation_freq, max_queue_size, workers, use_multiprocessing)
1143 epoch_logs.update(val_logs)
1144
-> 1145 callbacks.on_epoch_end(epoch, epoch_logs)
1146 training_logs = epoch_logs
1147 if self.stop_training:
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/callbacks.py in on_epoch_end(self, epoch, logs)
426 for callback in self.callbacks:
427 if getattr(callback, '_supports_tf_logs', False):
--> 428 callback.on_epoch_end(epoch, logs)
429 else:
430 if numpy_logs is None: # Only convert once.
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/callbacks.py in on_epoch_end(self, epoch, logs)
1342 # pylint: disable=protected-access
1343 if self.save_freq == 'epoch':
-> 1344 self._save_model(epoch=epoch, logs=logs)
1345
1346 def _should_save_on_batch(self, batch):
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/callbacks.py in _save_model(self, epoch, logs)
1406 filepath, overwrite=True, options=self._options)
1407 else:
-> 1408 self.model.save(filepath, overwrite=True, options=self._options)
1409
1410 self._maybe_remove_file()
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py in save(self, filepath, overwrite, include_optimizer, save_format, signatures, options, save_traces)
2000 # pylint: enable=line-too-long
2001 save.save_model(self, filepath, overwrite, include_optimizer, save_format,
-> 2002 signatures, options, save_traces)
2003
2004 def save_weights(self,
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/saving/save.py in save_model(model, filepath, overwrite, include_optimizer, save_format, signatures, options, save_traces)
152 'or using `save_weights`.')
153 hdf5_format.save_model_to_hdf5(
--> 154 model, filepath, overwrite, include_optimizer)
155 else:
156 saved_model_save.save(model, filepath, overwrite, include_optimizer,
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/saving/hdf5_format.py in save_model_to_hdf5(model, filepath, overwrite, include_optimizer)
129 if (include_optimizer and model.optimizer and
130 not isinstance(model.optimizer, optimizer_v1.TFOptimizer)):
--> 131 save_optimizer_weights_to_hdf5_group(f, model.optimizer)
132
133 f.flush()
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/saving/hdf5_format.py in save_optimizer_weights_to_hdf5_group(hdf5_group, optimizer)
594 for name, val in zip(weight_names, weight_values):
595 param_dset = weights_group.create_dataset(
--> 596 name, val.shape, dtype=val.dtype)
597 if not val.shape:
598 # scalar
/usr/local/lib/python3.7/dist-packages/h5py/_hl/group.py in create_dataset(self, name, shape, dtype, data, **kwds)
137 dset = dataset.Dataset(dsid)
138 if name is not None:
--> 139 self[name] = dset
140 return dset
141
/usr/local/lib/python3.7/dist-packages/h5py/_hl/group.py in __setitem__(self, name, obj)
371
372 if isinstance(obj, HLObject):
--> 373 h5o.link(obj.id, self.id, name, lcpl=lcpl, lapl=self._lapl)
374
375 elif isinstance(obj, SoftLink):
h5py/_objects.pyx in h5py._objects.with_phil.wrapper()
h5py/_objects.pyx in h5py._objects.with_phil.wrapper()
h5py/h5o.pyx in h5py.h5o.link()
RuntimeError: Unable to create link (name already exists)
这是我的代码：
new_input1 = Input(shape=(224, 224, 3))
new_input2 = Input(shape=(224, 224, 3))
effB3_1 = EfficientNetB2(include_top=False,weights='imagenet')
effB3_1._name="bf1"
effB3_11=effB3_1(new_input1)
gap_1=GlobalMaxPooling2D(name="gap_1")(effB3_11)
effB3_2 = EfficientNetB2(include_top=False,weights='imagenet')
effB3_2._name="bf2"
effB3_22=effB3_2(new_input2)
gap_2=GlobalMaxPooling2D(name="gap_2")(effB3_22)
merge=concatenate([gap_1,gap_2])
dense1=Dense(64, activation='relu', name="fc2", kernel_regularizer=l1(0.001),bias_regularizer=l1(0.001))(merge)
output=Dense(2, activation='softmax', name="fc_out")(dense1)
model=Model([new_input1,new_input2],output)
model.compile(loss='categorical_crossentropy',
optimizer=sc,
metrics=["accuracy"])
filepath="path"
checkpoint=ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, mode='auto',save_weights_only=False)
callbacks_list = [checkpoint]
model.fit(train_generator,epochs=20,steps_per_epoch = tr_sample // batch_size, validation_data = validation_generator,validation_steps = val_sample // batch_size,callbacks=callbacks_list)
如果我将 save_weights_only 更改为 True 一切都很好。我知道这个问题与保存优化器参数有关，但我不知道如何修复这个错误并保存整个模型？
P.S：我还在编译之前为完整模型和模型之一（effb3_1）编写了以下代码，但它没有解决问题。
for i in range(len(model.weights)):
model.weights[i]._handle_name = model.weights[i].name + "_" + str(i)
谷歌协作 TF 2.4
【问题讨论】：
标签： python tensorflow keras h5py