我正在使用 tf.keras 构建一个简单的深度学习模型，结构如下:
def build_simple_model(timestep, dim1, show_model_structure=False):
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.LSTM(units=64, input_shape=(timestep, dim1), activation="relu", return_sequences=True, dtype="float64"))
model.add(tf.keras.layers.LSTM(units=64, input_shape=(timestep, dim1), activation="relu", return_sequences=True, dtype="float64"))
model.add(tf.keras.layers.LSTM(units=64, input_shape=(timestep, dim1), activation="relu", dtype="float64"))
model.add(tf.keras.layers.Dense(32, activation="relu",dtype="float64"))
model.add(tf.keras.layers.Dense(1,dtype="float64"))
# adam = Adam(lr=0.01, beta_1=0.9, beta_2=0.999, amsgrad=False)
model.compile(loss='mse',
optimizer='adam',
metrics=['mape'])
if show_model_structure:
model.summary()
draw_model_graph(model)
return model
并在 Colab 上使用了 TPU:
resolver = tf.distribute.cluster_resolver.TPUClusterResolver()
tf.config.experimental_connect_to_cluster(resolver)
tf.tpu.experimental.initialize_tpu_system(resolver)
strategy = tf.distribute.experimental.TPUStrategy(resolver)
使用新的 TPU API 构建和拟合模型:
with strategy.scope():
test_model = build_simple_model(23, x_t.shape[2])
test_model.fit(x_t.astype('float64'), y_train.astype('float64'), epochs=100, batch_size=64)#, callbacks=[callback])
但是，奇怪的错误仍然存在，并提示错误的数据类型。
完整跟踪堆栈:
---------------------------------------------------------------------------
InvalidArgumentError Traceback (most recent call last)
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py in _do_call(self, fn, *args)
1364 try:
-> 1365 return fn(*args)
1366 except errors.OpError as e:
11 frames
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py in _run_fn(feed_dict, fetch_list, target_list, options, run_metadata)
1347 # Ensure any changes to the graph are reflected in the runtime.
-> 1348 self._extend_graph()
1349 return self._call_tf_sessionrun(options, feed_dict, fetch_list,
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py in _extend_graph(self)
1387 with self._graph._session_run_lock(): # pylint: disable=protected-access
-> 1388 tf_session.ExtendSession(self._session)
1389
InvalidArgumentError: Unsupported data type for TPU: double, caused by output IteratorGetNext:0
During handling of the above exception, another exception occurred:
InvalidArgumentError Traceback (most recent call last)
<ipython-input-150-8f5f7911c182> in <module>()
1 with strategy.scope():
2 test_model = build_simple_model(23, x_t.shape[2])
----> 3 test_model.fit(x_t.astype('float64'), y_train.astype('float64'), epochs=100, batch_size=64)#, callbacks=[callback])
/usr/local/lib/python3.6/dist-packages/tensorflow_core/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_freq, max_queue_size, workers, use_multiprocessing, **kwargs)
725 max_queue_size=max_queue_size,
726 workers=workers,
--> 727 use_multiprocessing=use_multiprocessing)
728
729 def evaluate(self,
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/engine/training_distributed.py in fit(self, model, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_freq, **kwargs)
617 validation_split=validation_split,
618 shuffle=shuffle,
--> 619 epochs=epochs)
620 if not dist_utils.is_distributing_by_cloning(model):
621 with model._distribution_strategy.scope():
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/engine/training.py in _distribution_standardize_user_data(self, x, y, sample_weight, class_weight, batch_size, validation_split, shuffle, epochs, allow_partial_batch)
2270 session = None
2271 else:
-> 2272 session = K.get_session()
2273
2274 first_x_value = nest.flatten(x)[0]
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/backend.py in get_session(op_input_list)
484 if not _MANUAL_VAR_INIT:
485 with session.graph.as_default():
--> 486 _initialize_variables(session)
487 return session
488
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/backend.py in _initialize_variables(session)
901 # marked as initialized.
902 is_initialized = session.run(
--> 903 [variables_module.is_variable_initialized(v) for v in candidate_vars])
904 uninitialized_vars = []
905 for flag, v in zip(is_initialized, candidate_vars):
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py in run(self, fetches, feed_dict, options, run_metadata)
954 try:
955 result = self._run(None, fetches, feed_dict, options_ptr,
--> 956 run_metadata_ptr)
957 if run_metadata:
958 proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py in _run(self, handle, fetches, feed_dict, options, run_metadata)
1178 if final_fetches or final_targets or (handle and feed_dict_tensor):
1179 results = self._do_run(handle, final_targets, final_fetches,
-> 1180 feed_dict_tensor, options, run_metadata)
1181 else:
1182 results = []
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py in _do_run(self, handle, target_list, fetch_list, feed_dict, options, run_metadata)
1357 if handle is None:
1358 return self._do_call(_run_fn, feeds, fetches, targets, options,
-> 1359 run_metadata)
1360 else:
1361 return self._do_call(_prun_fn, handle, feeds, fetches)
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py in _do_call(self, fn, *args)
1382 '\nsession_config.graph_options.rewrite_options.'
1383 'disable_meta_optimizer = True')
-> 1384 raise type(e)(node_def, op, message)
1385
1386 def _extend_graph(self):
InvalidArgumentError: Unsupported data type for TPU: double, caused by output IteratorGetNext:0
有人知道这是怎么回事吗？
提前致谢。
最佳答案
我在使用 TPU 运行 keras 模型时遇到了同样的问题。
我发现它与提供给 .fit 的 y 标签(第二个)参数的数据类型有关。它顺利工作的唯一方法是使用 dtype int64。任何 float 类型都会收到相同的异常。
在我的例子中，虽然我计划使用需要浮点标签的二进制交叉熵来运行模型，但我可以使用带有 int 标签的分类交叉熵来解决它。
在你的情况下，我看到你的损失是 mse，所以不确定你是否可以通过使用 int 标签来解决。
另一种成功解决该异常但随后出现其他异常的方法如下:
定义y标签向量为tf.Variable:
sess = K.get_session()
...
y = tf.Variable(my_baseline_y_vector, dtype='bfloat16') # any float dtype EXCEPT float64
然后在调用 .fit 之前:
train_y = y.eval(session=sess)
将其转换为可提供给 .fit 的数组
model.fit(train_x, train_y, ...)
如前所述，这通过了讨论的 dtype 异常，但随后收到了其他异常 :(。我放弃了，正在使用带有 int 标签的分类。如果您没有这样的选择，那么我很乐意听到你是如何处理其他异常(exception)的:)。
关于python-3.x - TPU : double, 不支持的数据类型由输出 IteratorGetNext:0 引起，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/58535392/