TypeError: Expected float32, got 'auto' of type 'str' instead

Question

I'm trying to use Keras' functional API to handle multiple inputs, and with a custom loss function RMSLE. Below is my code:

import tensorflow as tf
from tensorflow.keras.layers import *
from tensorflow.keras.models import Sequential, Model
from tensorflow.keras import backend as K
from tensorflow.keras.losses import MeanSquaredLogarithmicError

def rmsle(y_true, y_pred):
  return K.sqrt(MeanSquaredLogarithmicError(y_true, y_pred))

def build_model():

  i_language = Input(shape=(1,))
  i_year = Input(shape=(1,))
  i_abstract = Input(shape=(100,))

  input = concatenate([i_language, i_year, i_abstract])
  x = Dense(64)(input)
  x = Dense(1, activation='softmax')(x)

  model = Model(inputs=[i_language, i_year, i_abstract], outputs=x)
  model.compile(optimizer = 'adam', loss = rmsle)

  return model

model = build_model()

x1 = np.random.randint(3, size=(100, 1)).astype('float32')
x2 = np.random.randint(59, size=(100, 1)).astype('float32')
x3 = np.random.randn(100, 100)

y = np.random.rand(100,1)

model.fit([x1,x2,x3], y)

where x1,x2,x3 are all sample inputs and y is a sample output. But, he last line model.fit() throws the error:

TypeError                                 Traceback (most recent call last)
<ipython-input-33-66ea59ad4aed> in <module>()
      5 y = np.random.rand(100,1)
      6 
----> 7 model.fit([x1,x2,x3], y)

9 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/func_graph.py in wrapper(*args, **kwargs)
    984           except Exception as e:  # pylint:disable=broad-except
    985             if hasattr(e, "ag_error_metadata"):
--> 986               raise e.ag_error_metadata.to_exception(e)
    987             else:
    988               raise

TypeError: in user code:

    /usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:855 train_function  *
        return step_function(self, iterator)
    <ipython-input-17-6a742f71a83b>:2 rmsle  *
        return K.sqrt(MeanSquaredLogarithmicError(y_true, y_pred))
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/losses.py:506 __init__  **
        mean_squared_logarithmic_error, name=name, reduction=reduction)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/losses.py:241 __init__
        super(LossFunctionWrapper, self).__init__(reduction=reduction, name=name)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/losses.py:102 __init__
        losses_utils.ReductionV2.validate(reduction)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/utils/losses_utils.py:76 validate
        if key not in cls.all():
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/util/dispatch.py:206 wrapper
        return target(*args, **kwargs)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/ops/math_ops.py:1800 tensor_equals
        self, other = maybe_promote_tensors(self, other)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/ops/math_ops.py:1202 maybe_promote_tensors
        ops.convert_to_tensor(tensor, dtype, name="x"))
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/profiler/trace.py:163 wrapped
        return func(*args, **kwargs)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/ops.py:1566 convert_to_tensor
        ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/constant_op.py:339 _constant_tensor_conversion_function
        return constant(v, dtype=dtype, name=name)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/constant_op.py:265 constant
        allow_broadcast=True)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/constant_op.py:283 _constant_impl
        allow_broadcast=allow_broadcast))
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/tensor_util.py:457 make_tensor_proto
        _AssertCompatible(values, dtype)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/tensor_util.py:337 _AssertCompatible
        (dtype.name, repr(mismatch), type(mismatch).__name__))

    TypeError: Expected float32, got 'auto' of type 'str' instead.

I haven't encountered this error before and do not understand what's happening. Could someone please help me get rid of this error?

Answer 1

Quotation from keras manual: "Note that all losses are available both via a class handle and via a function handle. The class handles enable you to pass configuration arguments to the constructor (e.g. loss_fn = CategoricalCrossentropy(from_logits=True)), and they perform reduction by default when used in a standalone way (see details below)."

[I made the same error.]

If one decides to use loss class keras.losses.MeanSquaredLogarithmicError, it needs to be instantiated before use in models. On the other hand if one decides to use function keras.losses.mean_squared_logarithmic_error it would be passed to the model itself.

Answer 2

repalce your custom loss with:

def rmsle(y_true, y_pred):
  msle = MeanSquaredLogarithmicError()
  return K.sqrt(msle(y_true, y_pred))