Keras: Loading model with custom loss function: ValueError: 'Unknown loss function'
I trained and saved a model that uses a custom loss function (Keras version: 2.0.2):
model.compile(optimizer=adam, loss=SSD_Loss(neg_pos_ratio=neg_pos_ratio, alpha=alpha).compute_loss)
When I try to load the model, I get this error:
ValueError: ('Unknown loss function', ':compute_loss')
This is the stack trace:
ValueError Traceback (most recent call last)
<ipython-input-76-52ca495a8e09> in <module>()
----> 1 model, layer_dict, classifier_sizes = load_model('./model_0.h5')
/Users/pierluigiferrari/anaconda/envs/carnd-term1/lib/python3.5/site-packages/keras/models.py in load_model(filepath, custom_objects)
258 metrics=metrics,
259 loss_weights=loss_weights,
--> 260 sample_weight_mode=sample_weight_mode)
261
262 # Set optimizer weights.
/Users/pierluigiferrari/anaconda/envs/carnd-term1/lib/python3.5/site-packages/keras/engine/training.py in compile(self, optimizer, loss, metrics, loss_weights, sample_weight_mode, **kwargs)
738 loss_functions = [losses.get(l) for l in loss]
739 else:
--> 740 loss_function = losses.get(loss)
741 loss_functions = [loss_function for _ in range(len(self.outputs))]
742 self.loss_functions = loss_functions
/Users/pierluigiferrari/anaconda/envs/carnd-term1/lib/python3.5/site-packages/keras/losses.py in get(identifier)
88 if isinstance(identifier, six.string_types):
89 identifier = str(identifier)
---> 90 return deserialize(identifier)
91 elif callable(identifier):
92 return identifier
/Users/pierluigiferrari/anaconda/envs/carnd-term1/lib/python3.5/site-packages/keras/losses.py in deserialize(name, custom_objects)
80 module_objects=globals(),
81 custom_objects=custom_objects,
---> 82 printable_module_name='loss function')
83
84
/Users/pierluigiferrari/anaconda/envs/carnd-term1/lib/python3.5/site-packages/keras/utils/generic_utils.py in deserialize_keras_object(identifier, module_objects, custom_objects, printable_module_name)
155 if fn is None:
156 raise ValueError('Unknown ' + printable_module_name,
--> 157 ':' + function_name)
158 return fn
159 else:
ValueError: ('Unknown loss function', ':compute_loss')
[x] Check that you are up-to-date with the master branch of Keras. You can update with:
pip install git+git://github.com/fchollet/keras.git --upgrade --no-deps[x] If running on TensorFlow, check that you are up-to-date with the latest version. The installation instructions can be found here.
pierluigiferrari
All 56 comments
I have this exact same error and just noticed this morning.
My metrics are very simple:
#-----------------------------------------------------------------------------------------------------------------------------------------------------
# PFA, prob false alert for binary classifier
def binary_PFA(y_true, y_pred):
# N = total number of negative labels
N = K.sum(1 - K.round(y_true))
# FP = total number of false alerts, alerts from the negative class labels
FP = K.sum(K.round(y_pred) - K.round(y_pred) * K.round(y_true))
return FP/N
#-----------------------------------------------------------------------------------------------------------------------------------------------------
# P_TA prob true alerts for binary classifier
def binary_PTA(y_true, y_pred):
# P = total number of positive labels
P = K.sum(K.round(y_true))
# TP = total number of correct alerts, alerts from the positive class labels
TP = K.sum(K.round(y_pred) * K.round(y_true))
return TP/P
#-----------------------------------------------------------------------------------------------------------------------------------------------------
myOptimizer = keras.optimizers.adadelta(lr=1.0)
model.compile(loss='binary_crossentropy', optimizer = myOptimizer, metrics=[keras.metrics.binary_accuracy, metrics.binary_PTA, metrics.binary_PFA])
isaacgerg
on 22 Mar 2017
It looks like its trying to find the fn in generic_utils.py for a function_name which is binary_PTA which isnt found in custom_objects. @fchollet How do we add our metric to custom_objects?
isaacgerg
on 22 Mar 2017
I did this as a work around:
model = model_from_json(open(modelFile).read())
model.load_weights(os.path.join(os.path.dirname(modelFile), 'model_weights.h5'))
@isaacgerg what Keras version are you running? I got the error described in the old title in Keras 2.0.0, now after updating to 2.0.2 I'm getting a new error (as described in the new title).
But yeah, for the moment, saving and loading the weights separately is the way to go as a workaround.
pierluigiferrari
on 22 Mar 2017
I use 2.0.1
isaacgerg
on 22 Mar 2017
Hi, the same problem here.
One ugly solution that worked for me is to include the custom objective into keras:
import keras.losses
keras.losses.custom_loss = custom_loss
This is a know issue on keras 1 #3977.
On keras 2.0 you have to replace keras.objectives to keras.losses.
dluvizon
on 30 Mar 2017
My work around is to load the json first and then load the weights. I keep my load function in a local lib with the rest of my keras workarounds ;)
isaacgerg
on 30 Mar 2017
This PR should have fixed this issue.
Can use:
from keras.utils.generic_utils import get_custom_objects
import SSD_Loss
loss = SSD_Loss(neg_pos_ratio=neg_pos_ratio, alpha=alpha)
get_custom_objects().update({"SSD_Loss": loss.computeloss})
joeyearsley
on 16 Apr 2017
Same here.
I tried @joeyearsley's workaround and seems loading correctly.
inspiralpatterns
on 28 Apr 2017
I solved this problem by adding 'custom_bojects'
model = load_model('model/multi_task/try.h5', custom_objects={'loss_max': loss_max})
my loss function:
def loss_max(y_true, y_pred):
from keras import backend as K
return K.max(K.abs(y_pred - y_true), axis=-1)
Bisgates
on 9 May 2017
My workaround is like @dluvizon except I assigned keras.losses.loss. That was the missing function name according to the error message.
/Users/apiccolboni/anaconda/lib/python2.7/site-packages/keras/utils/generic_utils.pyc in deserialize_keras_object(identifier, module_objects, custom_objects, printable_module_name)
155 if fn is None:
156 raise ValueError('Unknown ' + printable_module_name,
--> 157 ':' + function_name)
158 return fn
159 else:
ValueError: ('Unknown loss function', ':loss')
keras 2.0.4
piccolbo
on 23 May 2017
Hi, I have a similar problem with:
model.compile( loss=lambda x,y: custom_loss_function(x,y,third_argument), optimizer=optimizer)
Keras 2.0.3 (Python 2.7.6) gives me the error:
('Unknown loss function', ':<lambda>')
Does anyone can help me ? Many thanks!
pigna90
on 30 May 2017
Hi @pigna90 , the easiest way is to define a python function in the form:
def custom_loss_function(y_true, y_pred):
# Compute loss
return loss
Then you pass it as your loss in model.compile(loss=custom_loss_function, [...]).
dluvizon
on 30 May 2017
Thanks @dluvizon, but I need three parameters as I wrote in my first post. How can I handle it?
The error is raised only when I try to reload the model that has been saved.
pigna90
on 30 May 2017
Checkout functools.partial This is really basic Python, so I think this is
not the appropriate forum to discuss it.
On Tue, May 30, 2017, 8:55 AM Alessandro Romano notifications@github.com
wrote:
Thanks @dluvizon https://github.com/dluvizon, but I need three
parameters as I wrote in my first post. How can I handle it?
The error is raised only when I try to reload the model that has been
saved.—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
https://github.com/fchollet/keras/issues/5916#issuecomment-304924075,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AA3RW8F8kgL35_f7D3iXjj9kuVW6h0Opks5r_DwCgaJpZM4Mki1T
.
piccolbo
on 30 May 2017
@piccolbo actually I'm using also functools.partial, but I need to execute the following code in save/serialization time and in loading time:
custom_loss_partial = functools.partial(custom_loss_function, third_argument=third_argument)
custom_loss_partial.__name__ = "custom_loss_function"
I'm searching for a way that allows me to load the model without declare custom_loss_partial two times.
I hope that my issue it's relevant and clear.
pigna90
on 30 May 2017
@Bisgates
Hi, I tried your way but sadly it doesn't work. It just gives me the same error message:
File "/home/gnahzuy/.conda/envs/gpu/lib/python3.5/site-packages/keras/losses.py", line 94, in deserialize
printable_module_name='loss function')
File "/home/gnahzuy/.conda/envs/gpu/lib/python3.5/site-packages/keras/utils/generic_utils.py", line 159, in deserialize_keras_object
':' + function_name)
ValueError: Unknown loss function:sorenson_dice
I compile my model like:
model.compile(optimizer='adam', loss=sorenson_dice)
and I load my model like:
model = keras.models.load_model('/home/gnahzuy/U-net/scripts/val_loss=.-0.88.hdf5',custom_objecta={'val_loss': sorenson_dice})
I define the loss function in the same file.
I think that might because I give the worry name 'val_loss'? I tried 'loss' but it doesn't work too.
Do you have any idea about that?
Thanks
ZY0422
on 23 Jul 2017
@ZY0422 You want custom_objects={'sorenson_dice': sorenson_dice}
wkm
on 3 Aug 2017
This issue has been automatically marked as stale because it has not had recent activity. It will be closed after 30 days if no further activity occurs, but feel free to re-open a closed issue if needed.
![stale[bot] picture](https://avatars3.githubusercontent.com/in/1724?v=4&s=40)
stale[bot]
on 1 Nov 2017
I have a similar problem
I compile my model like this:
model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer=sgd, metrics=['accuracy'])
when load_model('model.hdf5', custom_objects={'ctc': lambda y_true, y_pred: y_pred})
it raise "Unknown loss function: '
Can anyone help me ,thank U very much!!!
caocao1989
on 5 Jan 2018
@caocao1989
I'd guess you're using the CTC ocr example here, too.
I was able to use custom_objects = {'<lambda>': lambda y_true, y_pred: y_pred} as a work-around.
I hope this helps!
SimulatedANeal
on 10 Jan 2018
@SimulatedANeal
Yes,Thanks for your reply,I have solved the problem like your answer
caocao1989
on 10 Jan 2018
@SimulatedANeal
Wow, great! But i wanna to know why this can work? I never see such kind of custom_objects structure before. Where did you see it?
LCorleone
on 10 Feb 2018
@SimulatedANeal I am using ssd_keras loss from keras_loss_function.keras_ssd_loss import SSDLoss and again while loading the model I'm getting this error ValueError: Unknown loss function:compute_loss . I tried above methods to resolve but none of these worked. I am using Keras 2.1.3.
pymit
on 21 Jul 2018
@piccolbo, As @pigna90 referred earlier, I am as well using a custom partial function which requires additional arguments. The third argument is actually an input node in the model. Attaching a snippet from the model corresponding to it:
def sparse_weighted_loss(target, output, weights):
return tf.multiply(tf.keras.backend.sparse_categorical_crossentropy(target, output), weights)
weights_tensor = Input(shape=(None,), dtype='float32', name='weights_input')
lossFct = partial(sparse_weighted_loss, weights=weights_tensor)
update_wrapper(lossFct, sparse_weighted_loss)
I use lossFct as my custom loss function (which is basically a example-wise weighted cross-entropy loss). Now I redefine sparse_weighted_loss in the custom_objects as follows:
def sparse_weighted_loss(target, output, weights):
return tf.multiply(tf.keras.backend.sparse_categorical_crossentropy(target, output), weights)
custom_obj = {}
custom_obj['sparse_weighted_loss'] = sparse_weighted_loss
model = keras.models.load_model(modelPath, custom_objects=custom_obj)
While loading the model, it still throws this error:
Traceback (most recent call last):
File "Train_Product_NER_weighted_softmax.py", line 112, in <module>
model = BiLSTM.loadModel(sys.argv[2])
File "/BiLSTM_weightedloss.py", line 653, in loadModel
model = keras.models.load_model(modelPath, custom_objects=custom_obj)
File "/usr/local/lib/python3.4/site-packages/Keras-2.1.6-py3.4.egg/keras/models.py", line 388, in load_model
File "/usr/local/lib/python3.4/site-packages/Keras-2.1.6-py3.4.egg/keras/engine/training.py", line 837, in compile
File "/usr/local/lib/python3.4/site-packages/Keras-2.1.6-py3.4.egg/keras/engine/training.py", line 429, in weighted
TypeError: sparse_weighted_loss() missing 1 required positional argument: 'weights'
Any insights anyone here?
TanmayParekh
on 11 Sep 2018
@TanmayParekh It's an unrelated issue to the one discussed here. I don't think you are going to find what you need if you use the issue tracking system this way. If it were my project, you'd have to open a separate issue. Sometimes user groups are a first line of support: https://groups.google.com/forum/#!forum/keras-users Good luck.
piccolbo
on 11 Sep 2018
@piccolbo Thanks a lot. New to the community. This helps! :)
TanmayParekh
on 11 Sep 2018
@TanmayParekh Have you solved your problem? I have the same error. Thanks.
shuaiw24
on 18 Sep 2018
I am getting same issue.
*** Model is compiled and saved: ******
`def triplet_loss(y_true, y_pred, alpha = 0.3):
anchor, positive, negative = y_pred[0], y_pred[1], y_pred[2]
pos_dist = tf.reduce_sum(tf.square(tf.subtract(anchor, positive)), axis=-1)
neg_dist = tf.reduce_sum(tf.square(tf.subtract(anchor, negative)), axis=-1)
basic_loss = tf.add(tf.subtract(pos_dist, neg_dist), alpha)
loss = tf.reduce_sum(tf.maximum(basic_loss, 0.0))
return loss
FRmodel.compile(optimizer = 'adam', loss = triplet_loss, metrics = ['accuracy'])
FRmodel.save('model.h5')`
**** loading model using: ****
FRmodel = load_model('model.h5')
****Error getting: *******
ValueError: Unknown loss function:triplet_loss
Can someone suggest solution for this please?
Thanks
Nish2288
on 25 Sep 2018
@shuaiw24 I used a different hack here. Couldn't solve the original problem though.
I am only saving the weights of the model right now. During loading, I re-create the model and load the weights into the model. That works well for my problem.
TanmayParekh
on 25 Sep 2018
I solved this problem by adding 'custom_bojects'
model = load_model('model/multi_task/try.h5', custom_objects={'loss_max': loss_max})my loss function:
def loss_max(y_true, y_pred): from keras import backend as K return K.max(K.abs(y_pred - y_true), axis=-1)
This is the most elegant solution I've seen, thanks a lot!
soulmachine
on 30 Nov 2018
https://github.com/keras-team/keras/issues/11821
keras 2.0.8, tf1.4
my loss is written something like this:
def segnet_loss_graph(y_true, y_pred):
return K.categorical_crossentropy(y_true, y_pred)
and the code is:
segnet_loss = KL.Lambda(
lambda x: segnet_loss_graph(*x), name="segnet_loss")(
[input_gt_segmap,seg_score_map])
outputs = [segnet_loss]
model = KM.Model(inputs, outputs, name='multi_task_segment')
and loss is adding like this
keras_model.add_loss(loss)
=======
cuz i save the weights and structure, i load model directly
keras.callbacks.ModelCheckpoint(checkpoint_path, verbose=0, save_weights_only=False)
...
...
here is the inference code
from keras.models import load_model
import tensorflow as tf
import keras
import keras.losses
keras.losses.custom_loss ="segnet_loss"
custom_objects= {
"backend":keras.backend,
"tf": tf,
"segnet_loss_graph": segnet_loss_graph
}
model = load_model(model_path,custom_objects=custom_objects)
it's no use if i change "segnet_loss_graph": segnet_loss_graph to any of the following stuff
"segnet_loss": segnet_loss_graph
"segnet_loss": lambda x: segnet_loss_graph(x)
"segnet_loss_graph": lambda x: segnet_loss_graph(x)
the error will be following appears with different parameter i pass
1.ValueError: The model cannot be compiled because it has no loss to optimize.
2.NameError: name 'segnet_loss_graph' is not defined
3.TypeError: 'NoneType' object is not callable
is there any hint to solve this problem????
world4jason
on 7 Dec 2018
@SimulatedANeal @dluvizon
I compile my model like this:
model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer=adam)
After that I load model:
load_model('model.hdf5', custom_objects={'\
Before save:
Epoch 00001: val_loss improved from inf to 228.31433, saving model to ../model.hdf5
Epoch 00002: val_loss improved from 228.31433 to 176.43184
After save:
Epoch 00001: val_loss improved from inf to 117.94428, saving model to ../model.hdf5
==> why model is improved from 'inf' ?. Is it a problem ? how to fix ?
ducvm
on 29 Dec 2018
If you load model to continue training, you need to define your custom loss function:
def custom_loss(y_true, y_pred):
# code
return loss
model = load_model('model.hdf5', custom_objects={'custom_loss': custom_loss})
If you load model only for prediction (without training), you need to set compile flag to False:
model = load_model('model.hdf5', compile=False)
and you don`t need to define your custom_loss, because the loss function is not necessary for prediction
only.
Pepslee
on 25 Jan 2019
I solved this problem by adding 'custom_bojects'
model = load_model('model/multi_task/try.h5', custom_objects={'loss_max': loss_max})my loss function:
def loss_max(y_true, y_pred): from keras import backend as K return K.max(K.abs(y_pred - y_true), axis=-1)My loss function is like this.
def custom_loss_function(inputs):
def custom_loss(y_true, y_pred):
print (inputs.shape, y_true.shape, y_pred.shape)
x = k.exp(inputs[:,12,:])
y_t = k.log(k.exp(y_true)*x)
y_p = k.log(k.exp(y_pred)*x)
# y_p[y_p==0] = 1e-6
l = y_t - y_p
return k.square(l)
return custom_loss
I am getting the following error
model=load_model('lstm_fftm_custom_loss.h5', custom_objects={'custom_loss':custom_loss})
NameError: name 'custom_loss' is not defined
Can you please help?
sayZeeel
on 4 Mar 2019
I solved this problem by adding 'custom_bojects'
model = load_model('model/multi_task/try.h5', custom_objects={'loss_max': loss_max})my loss function:
def loss_max(y_true, y_pred): from keras import backend as K return K.max(K.abs(y_pred - y_true), axis=-1)My loss function is like this.
def custom_loss_function(inputs):
def custom_loss(y_true, y_pred): print (inputs.shape, y_true.shape, y_pred.shape) x = k.exp(inputs[:,12,:]) y_t = k.log(k.exp(y_true)*x) y_p = k.log(k.exp(y_pred)*x) # y_p[y_p==0] = 1e-6 l = y_t - y_p return k.square(l) return custom_lossI am getting the following error
model=load_model('lstm_fftm_custom_loss.h5', custom_objects={'custom_loss':custom_loss})
NameError: name 'custom_loss' is not definedCan you please help?
@sayZeeel Try in your loss function to give it a name:
def custom_loss_function(inputs):
def custom_loss(y_true, y_pred):
print (inputs.shape, y_true.shape, y_pred.shape)
x = k.exp(inputs[:,12,:])
y_t = k.log(k.exp(y_true)*x)
y_p = k.log(k.exp(y_pred)*x)
# y_p[y_p==0] = 1e-6
l = y_t - y_p
return k.square(l)
custom_loss.__name__ = "Custom Loss"
return custom_loss
and then:
model = load_model('model/multi_task/try.h5', custom_objects={'Custom Loss': custom_loss_function})
sofiaherrero
on 7 Mar 2019
I solved this problem by adding 'custom_bojects'
model = load_model('model/multi_task/try.h5', custom_objects={'loss_max': loss_max})my loss function:
def loss_max(y_true, y_pred): from keras import backend as K return K.max(K.abs(y_pred - y_true), axis=-1)
what if I have many losses!!! I have 7 clasifier with seven loss functions:
losses = {
"digit1_output": custom_loss.weighted_categorical_crossentropy(np.float32(weights[0])),
"digit2_output":custom_loss.weighted_categorical_crossentropy(np.float32(weights[1])),
"alphaNum_output": custom_loss.weighted_categorical_crossentropy(np.float32(weights_c)),
"digit3_output": custom_loss.weighted_categorical_crossentropy(np.float32(weights[2])),
"digit4_output": custom_loss.weighted_categorical_crossentropy(np.float32(weights[3])),
"digit5_output": custom_loss.weighted_categorical_crossentropy(np.float32(weights[4])),
"digit6_output": custom_loss.weighted_categorical_crossentropy(np.float32(weights[5])),
"digit7_output": custom_loss.weighted_categorical_crossentropy(np.float32(weights[6]))
}
marz869
on 21 Apr 2019
Thanks @isaacgerg, your solution worked well. However, I had to re-train the model to save the json file since I had only saved my model as a h5 file.
I also tried @Pepslee's direct solution with compile=False option. It works perfectly, thanks a lot! This was the solution I actually needed, because I load the model only for prediction.
In short, both solutions work perfectly for different cases. Use @isaacgerg's solution if you're loading the model for training again (i.e. for transfer learning). Don't forget to store your model as json after training using:
model_json = model.to_json()
with open('model.json', "w") as json_file:
json_file.write(model_json)
If your only aim is to predict, then use @Pepslee's solution. It's more straightforward.
cememreakbas
on 31 May 2019
I solved this problem by adding 'custom_bojects'
model = load_model('model/multi_task/try.h5', custom_objects={'loss_max': loss_max})my loss function:
def loss_max(y_true, y_pred): from keras import backend as K return K.max(K.abs(y_pred - y_true), axis=-1)
@Bisgates I'm loading my model using the weights as such:
model = yolo_body(Input(shape=(None, None, 3)), 3, num_classes)
model.load_weights('logs/000/trained_weights_stage_1.h5')
model.compile(optimizer=Adam(lr=1e-4), loss={'yolo_loss': lambda y_true, y_pred: y_pred})
I can't add custom objects, is there a way to add yolo_loss, when loading the model using load_weights.
I am using https://github.com/qqwweee/keras-yolo3 github repository, with yolo_body and yolo_loss defined in yolov3.models.py file.
sirius0503
on 10 Jun 2019
Hi, the same problem here.
One ugly solution that worked for me is to include the custom objective into keras:
import keras.losses
keras.losses.custom_loss = custom_lossThis is a know issue on keras 1 #3977.
On keras 2.0 you have to replacekeras.objectivestokeras.losses.
@dluvizon: Can you elaborate on how to add the custom_loss in keras2.0?
sirius0503
on 10 Jun 2019
@TanmayParekh @shuaiw24, this is how you do it.
def sparse_weighted_loss_func(weights):
def sparse_weighted_loss(target, output):
return tf.multiply(tf.keras.backend.sparse_categorical_crossentropy(target, output), weights)
return sparse_weighted_loss
model.compile(loss=sparse_weighted_loss_func(weights), ...)
# during loading time, it will expect sparse_weighted_loss not sparse_weighted_loss_func
model = load_model('pathtomodel',
custom_objects: {'sparse_weighted_loss': sparse_weighted_loss_func(weights)})
lminer
on 16 Jun 2019
I trained and saved a model that uses a custom loss function (Keras version: 2.0.2):
model.compile(optimizer=adam, loss=SSD_Loss(neg_pos_ratio=neg_pos_ratio, alpha=alpha).compute_loss)When I try to load the model, I get this error:
ValueError: ('Unknown loss function', ':compute_loss')This is the stack trace:
ValueError Traceback (most recent call last) <ipython-input-76-52ca495a8e09> in <module>() ----> 1 model, layer_dict, classifier_sizes = load_model('./model_0.h5') /Users/pierluigiferrari/anaconda/envs/carnd-term1/lib/python3.5/site-packages/keras/models.py in load_model(filepath, custom_objects) 258 metrics=metrics, 259 loss_weights=loss_weights, --> 260 sample_weight_mode=sample_weight_mode) 261 262 # Set optimizer weights. /Users/pierluigiferrari/anaconda/envs/carnd-term1/lib/python3.5/site-packages/keras/engine/training.py in compile(self, optimizer, loss, metrics, loss_weights, sample_weight_mode, **kwargs) 738 loss_functions = [losses.get(l) for l in loss] 739 else: --> 740 loss_function = losses.get(loss) 741 loss_functions = [loss_function for _ in range(len(self.outputs))] 742 self.loss_functions = loss_functions /Users/pierluigiferrari/anaconda/envs/carnd-term1/lib/python3.5/site-packages/keras/losses.py in get(identifier) 88 if isinstance(identifier, six.string_types): 89 identifier = str(identifier) ---> 90 return deserialize(identifier) 91 elif callable(identifier): 92 return identifier /Users/pierluigiferrari/anaconda/envs/carnd-term1/lib/python3.5/site-packages/keras/losses.py in deserialize(name, custom_objects) 80 module_objects=globals(), 81 custom_objects=custom_objects, ---> 82 printable_module_name='loss function') 83 84 /Users/pierluigiferrari/anaconda/envs/carnd-term1/lib/python3.5/site-packages/keras/utils/generic_utils.py in deserialize_keras_object(identifier, module_objects, custom_objects, printable_module_name) 155 if fn is None: 156 raise ValueError('Unknown ' + printable_module_name, --> 157 ':' + function_name) 158 return fn 159 else: ValueError: ('Unknown loss function', ':compute_loss')
- [x] Check that you are up-to-date with the master branch of Keras. You can update with:
pip install git+git://github.com/fchollet/keras.git --upgrade --no-deps- [x] If running on TensorFlow, check that you are up-to-date with the latest version. The installation instructions can be found here.
The following code worked for me --
generator.compile(loss='mse', optimizer=opt, metrics=[perceptual_distance])
model = load_model("../input/srresnet-epoch-120/pretrain/model00000120.h5", custom_objects={'perceptual_distance': perceptual_distance})
weights = model.get_weights()
generator.set_weights(weights)
khushal1996
on 10 Jul 2019
CA4GitHub
on 5 Dec 2019
I solved this problem by adding 'custom_bojects'
model = load_model('model/multi_task/try.h5', custom_objects={'loss_max': loss_max})my loss function:
def loss_max(y_true, y_pred): from keras import backend as K return K.max(K.abs(y_pred - y_true), axis=-1)My loss function is like this.
def custom_loss_function(inputs):def custom_loss(y_true, y_pred): print (inputs.shape, y_true.shape, y_pred.shape) x = k.exp(inputs[:,12,:]) y_t = k.log(k.exp(y_true)*x) y_p = k.log(k.exp(y_pred)*x) # y_p[y_p==0] = 1e-6 l = y_t - y_p return k.square(l) return custom_lossI am getting the following error
model=load_model('lstm_fftm_custom_loss.h5', custom_objects={'custom_loss':custom_loss})
NameError: name 'custom_loss' is not defined
Can you please help?@sayZeeel Try in your loss function to give it a name:
def custom_loss_function(inputs): def custom_loss(y_true, y_pred): print (inputs.shape, y_true.shape, y_pred.shape) x = k.exp(inputs[:,12,:]) y_t = k.log(k.exp(y_true)*x) y_p = k.log(k.exp(y_pred)*x) # y_p[y_p==0] = 1e-6 l = y_t - y_p return k.square(l) custom_loss.__name__ = "Custom Loss" return custom_lossand then:
model = load_model('model/multi_task/try.h5', custom_objects={'Custom Loss': custom_loss_function})
Giving the custom loss a name was the only thing that worked in tf 2.1
ajaceves
on 17 Jan 2020
None of the above worked for me, but this did:
Load the model with compile=False and then compile it manually.
import tensorflow as tf
import tensorflow_hub as hub
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
from tensorflow.keras.optimizers import Adam
IMG_WIDTH = 224
IMG_HEIGHT = IMG_WIDTH
CHANNELS = 3
LEARNING_RATE = 1e-5
NUM_LABELS = 128
def create_model():
feature_extractor_url = 'https://tfhub.dev/google/imagenet/mobilenet_v2_100_224/feature_vector/4'
feature_extractor_layer = hub.KerasLayer(feature_extractor_url,
input_shape=(IMG_HEIGHT, IMG_WIDTH, CHANNELS))
feature_extractor_layer.trainable = False
return Sequential([
feature_extractor_layer,
Dense(1024, activation='relu', name='hidden_layer'),
Dense(NUM_LABELS, activation='sigmoid', name='output')
])
def load_model(model_dir):
# got the compile=False idea from @Pepslee's comment:
# https://github.com/keras-team/keras/issues/5916#issuecomment-457624404
return tf.keras.models.load_model(model_dir,
compile=False,
custom_objects={'KerasLayer': hub.KerasLayer})
# this didn't work.
# neither did the custom_loss_function with __name__ thing.
# custom_objects={'KerasLayer': hub.KerasLayer,
# 'custom_loss': custom_loss,
# 'custom_metric': custom_metric})
def train(prev_model):
# the trick is to load the model with compile=False
if prev_model:
model = load_model(prev_model)
else:
model = create_model()
# and then compile manually,
# the same way it does with a new model.
model.compile(
optimizer=Adam(learning_rate=LEARNING_RATE),
loss=custom_loss,
metrics=[custom_metric]
)
# model.fit ...
# def custom_loss(y_true, y_pred):
# ...
# def custom_metric(y_true, y_pred, threshold=0.5):
# ...
blvz
on 28 Feb 2020
I have a problem when I load this model: ValueError: Unknown loss function:
import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)
Input data files are available in the "../input/" directory.
For example, running this (by clicking run or pressing Shift+Enter) will list the files in the input directory
import matplotlib.pyplot as plt
import os
Imports
import os
import fnmatch
import cv2
import numpy as np
import string
import time
from keras.preprocessing.sequence import pad_sequences
from keras.layers import Dense, LSTM, Reshape, BatchNormalization, Input, Conv2D
from keras.layers import MaxPool2D, Lambda, Bidirectional
from keras.models import Model
from keras.activations import relu, sigmoid, softmax
import keras.backend as K
from keras.utils import to_categorical
from keras.callbacks import ModelCheckpoint
import tensorflow as tf
from tensorflow.python.client import device_lib
import xml.etree.ElementTree as ET
read xml file
doc = ET.parse("/home/yosra/Downloads/IAMDataset/xml/a01-000u.xml")
root = doc.getroot()
dic = []
label = []
for i in root.iter('word'):
dic.append(i.get('id'))
label.append(i.get('text'))
print(dic, ' ' ,label)
Global Variables
char_list = string.ascii_letters + string.digits
print("Character List: ", char_list)
function to decode the text into indice of char list
def encode_to_labels(text):
# We encode each output word into digits
digit_list = []
for index, character in enumerate(text):
try:
digit_list.append(char_list.index(character))
except:
print("Error in finding index for character ", character)
#End For
return digit_list
preprocess the data
read the image from IAM Dataset
n_samples = len(os.listdir('/home/yosra/Desktop/imagetest'))
#Number of samples in xml file
xml_samples = len(dic)
list of trining_set
training_img = []
training_txt=[]
train_input_length = []
train_label_length = []
orig_txt = []
lists for validation dataset
valid_img = []
valid_txt = []
valid_input_length = []
valid_label_length = []
valid_orig_txt = []
max_label_len = 0
Training Variables
batch_size = 256
epochs = 10
k=1
for i, pic in enumerate(os.listdir('/home/yosra/Desktop/imagetest')):
# Read image as grayscale
img = cv2.imread(os.path.join('/home/yosra/Desktop/imagetest', pic), cv2.IMREAD_GRAYSCALE)
pic_target = pic[:-4]
# convert each image of shape (32, 128, 1)
w, h = img.shape
if h > 128 or w > 32:
continue
# endif
# Process the images to bring them to scale
if w < 32:
add_zeros = np.ones((32-w, h))*255
img = np.concatenate((img, add_zeros))
# endif
if h < 128:
add_zeros = np.ones((32, 128-h))*255
img = np.concatenate((img, add_zeros), axis=1)
# endif
img = np.expand_dims(img , axis = 2)
# Normalise the image
img = img/255.
# Get the text for the image
txt = pic_target.split('_')[1]
# compute maximum length of the text
if len(txt) > max_label_len:
max_label_len = len(txt)
if k%10 == 0:
valid_orig_txt.append(txt)
valid_label_length.append(len(txt))
valid_input_length.append(31)
valid_img.append(img)
valid_txt.append(encode_to_labels(txt))
else:
orig_txt.append(txt)
train_label_length.append(len(txt))
train_input_length.append(31)
training_img.append(img)
training_txt.append(encode_to_labels(txt))
k+=1
print('kamlna')
pad each output label to maximum text length
train_padded_txt = pad_sequences(training_txt, maxlen=max_label_len, padding='post', value = len(char_list))
valid_padded_txt = pad_sequences(valid_txt, maxlen=max_label_len, padding='post', value = len(char_list))
input with shape of height=32 and width=128
input with shape of height=32 and width=128
inputs = Input(shape=(32,128,1))
convolution layer with kernel size (3,3)
conv_1 = Conv2D(64, (3,3), activation = 'relu', padding='same')(inputs)
poolig layer with kernel size (2,2)
pool_1 = MaxPool2D(pool_size=(2, 2), strides=2)(conv_1)
conv_2 = Conv2D(128, (3,3), activation = 'relu', padding='same')(pool_1)
pool_2 = MaxPool2D(pool_size=(2, 2), strides=2)(conv_2)
conv_3 = Conv2D(256, (3,3), activation = 'relu', padding='same')(pool_2)
conv_4 = Conv2D(256, (3,3), activation = 'relu', padding='same')(conv_3)
poolig layer with kernel size (2,1)
pool_4 = MaxPool2D(pool_size=(2, 1))(conv_4)
conv_5 = Conv2D(512, (3,3), activation = 'relu', padding='same')(pool_4)
Batch normalization layer
batch_norm_5 = BatchNormalization()(conv_5)
conv_6 = Conv2D(512, (3,3), activation = 'relu', padding='same')(batch_norm_5)
batch_norm_6 = BatchNormalization()(conv_6)
pool_6 = MaxPool2D(pool_size=(2, 1))(batch_norm_6)
conv_7 = Conv2D(512, (2,2), activation = 'relu')(pool_6)
squeezed = Lambda(lambda x: K.squeeze(x, 1))(conv_7)
bidirectional LSTM layers with units=128
blstm_1 = Bidirectional(LSTM(128, return_sequences=True, dropout = 0.2))(squeezed)
blstm_2 = Bidirectional(LSTM(128, return_sequences=True, dropout = 0.2))(blstm_1)
outputs = Dense(len(char_list)+1, activation = 'softmax')(blstm_2)
model to be used at test time
act_model = Model(inputs, outputs)
act_model.summary()
the CTC loss fnction is to predict the output text, it is very helpfull for the
text recognition topic.
labels = Input(name='the_labels', shape=[max_label_len], dtype='float32')
input_length = Input(name='input_length', shape=[1], dtype='int64')
label_length = Input(name='label_length', shape=[1], dtype='int64')
def ctc_lambda_func(args):
y_pred, labels, input_length, label_length = args
return K.ctc_batch_cost(labels, y_pred, input_length, label_length)
loss_out = Lambda(ctc_lambda_func, output_shape=(1,), name='ctc')([outputs, labels, input_length, label_length])
model to be used at training time
model = Model(inputs=[inputs, labels, input_length, label_length], outputs=loss_out)
train the model
model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer = 'adam')
filepath= "/home/yosra/Downloads/best_model.hdf5"
checkpoint = ModelCheckpoint(filepath=filepath, monitor='val_loss', verbose=1, save_best_only=True, mode='auto')
callbacks_list = [checkpoint]
training_img = np.array(training_img)
train_input_length = np.array(train_input_length)
train_label_length = np.array(train_label_length)
valid_img = np.array(valid_img)
valid_input_length = np.array(valid_input_length)
valid_label_length = np.array(valid_label_length)
model.fit(x=[training_img, train_padded_txt, train_input_length, train_label_length],
y=np.zeros(len(training_img)), batch_size=batch_size,
epochs = epochs,
validation_data = ([valid_img, valid_padded_txt, valid_input_length, valid_label_length],[np.zeros(len(valid_img))]),
verbose = 1, callbacks = callbacks_list)
model.save(filepath)
test the model
from keras.models import load_model
load the saved best model weights
new_model = load_model(filepath)
yosraammar
on 4 May 2020
When I load my model, I have this error: ValueError: Unknown loss function:
pad each output label to maximum text length
train_padded_txt = pad_sequences(training_txt, maxlen=max_label_len, padding='post', value = len(char_list))
valid_padded_txt = pad_sequences(valid_txt, maxlen=max_label_len, padding='post', value = len(char_list))
input with shape of height=32 and width=128
input with shape of height=32 and width=128
inputs = Input(shape=(32,128,1))
convolution layer with kernel size (3,3)
conv_1 = Conv2D(64, (3,3), activation = 'relu', padding='same')(inputs)
poolig layer with kernel size (2,2)
pool_1 = MaxPool2D(pool_size=(2, 2), strides=2)(conv_1)
conv_2 = Conv2D(128, (3,3), activation = 'relu', padding='same')(pool_1)
pool_2 = MaxPool2D(pool_size=(2, 2), strides=2)(conv_2)
conv_3 = Conv2D(256, (3,3), activation = 'relu', padding='same')(pool_2)
conv_4 = Conv2D(256, (3,3), activation = 'relu', padding='same')(conv_3)
poolig layer with kernel size (2,1)
pool_4 = MaxPool2D(pool_size=(2, 1))(conv_4)
conv_5 = Conv2D(512, (3,3), activation = 'relu', padding='same')(pool_4)
Batch normalization layer
batch_norm_5 = BatchNormalization()(conv_5)
conv_6 = Conv2D(512, (3,3), activation = 'relu', padding='same')(batch_norm_5)
batch_norm_6 = BatchNormalization()(conv_6)
pool_6 = MaxPool2D(pool_size=(2, 1))(batch_norm_6)
conv_7 = Conv2D(512, (2,2), activation = 'relu')(pool_6)
squeezed = Lambda(lambda x: K.squeeze(x, 1))(conv_7)
bidirectional LSTM layers with units=128
blstm_1 = Bidirectional(LSTM(128, return_sequences=True, dropout = 0.2))(squeezed)
blstm_2 = Bidirectional(LSTM(128, return_sequences=True, dropout = 0.2))(blstm_1)
outputs = Dense(len(char_list)+1, activation = 'softmax')(blstm_2)
model to be used at test time
act_model = Model(inputs, outputs)
act_model.summary()
the CTC loss fnction is to predict the output text, it is very helpfull for the
text recognition topic.
labels = Input(name='the_labels', shape=[max_label_len], dtype='float32')
input_length = Input(name='input_length', shape=[1], dtype='int64')
label_length = Input(name='label_length', shape=[1], dtype='int64')
def ctc_lambda_func(args):
y_pred, labels, input_length, label_length = args
return K.ctc_batch_cost(labels, y_pred, input_length, label_length)
loss_out = Lambda(ctc_lambda_func, output_shape=(1,), name='ctc')([outputs, labels, input_length, label_length])
model to be used at training time
model = Model(inputs=[inputs, labels, input_length, label_length], outputs=loss_out)
train the model
model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer = 'adam')
filepath= "/home/yosra/Downloads/best_model.hdf5"
checkpoint = ModelCheckpoint(filepath=filepath, monitor='val_loss', verbose=1, save_best_only=True, mode='auto')
callbacks_list = [checkpoint]
training_img = np.array(training_img)
train_input_length = np.array(train_input_length)
train_label_length = np.array(train_label_length)
valid_img = np.array(valid_img)
valid_input_length = np.array(valid_input_length)
valid_label_length = np.array(valid_label_length)
model.fit(x=[training_img, train_padded_txt, train_input_length, train_label_length],
y=np.zeros(len(training_img)), batch_size=batch_size,
epochs = epochs,
validation_data = ([valid_img, valid_padded_txt, valid_input_length, valid_label_length],[np.zeros(len(valid_img))]),
verbose = 1, callbacks = callbacks_list)
model.save(filepath)
test the model
from keras.models import load_model
load the saved best model weights
new_model = load_model(filepath)
yosraammar
on 4 May 2020
Stop posting into this thread!!! There's an answer above, and you're making it hard to find!
WORKAROUND
model = tf.keras.models.load_model(path_here, compile=False)
Demetrio92
on 25 Jun 2020
could anyone solve this problem please?
This is the definition of my loss function:
def my_loss(y_true,y_pred,lambda_const,i,T,task_size=2):
y_trueSoft=y_true[:,:(i)task_size]
y_predSoft=y_pred[:,:(i)task_size]/T
y_trueHard=y_true[:,(i)task_size:]
y_predHard=y_pred[:,(i)task_size:]
return lambda_constcategorical_crossentropy(y_trueSoft,y_predSoft) + (1-lambda_const)categorical_crossentropy(y_trueHard,y_predHard)
model_final.compile(loss=lambda y_true, y_pred: my_loss(y_true, y_pred, lambda_const,i,T), optimizer="adam", metrics=["acc"])
And when I want to load this model , I got this error,ValueError: Unknown loss function:
I have changed the way I load this model, it writes below, but I still got the same error, I am so confused
### model_tmp=load_model(model_path_old,custom_objects={'angle bracket lambda angle bracket': lambda y_true, y_pred : my_loss(y_true, y_pred, lambda_const,i,T)})
Can anyone help me ? Thanks a lot
weilinapple
on 22 Jul 2020
could anyone solve this problem please?
This is the definition of my loss function:def my_loss(y_true,y_pred,lambda_const,i,T,task_size=2):
y_trueSoft=y_true[:,:(i)task_size]
y_predSoft=y_pred[:,:(i)task_size]/T
y_trueHard=y_true[:,(i)task_size:]
y_predHard=y_pred[:,(i)task_size:]return lambda_constcategorical_crossentropy(y_trueSoft,y_predSoft) + (1-lambda_const)categorical_crossentropy(y_trueHard,y_predHard)
model_final.compile(loss=lambda y_true, y_pred: my_loss(y_true, y_pred, lambda_const,i,T), optimizer="adam", metrics=["acc"])
And when I want to load this model , I got this error,ValueError: Unknown loss function:.
I have changed the way I load this model, it writes below, but I still got the same error, I am so confusedmodel_tmp=load_model(model_path_old,custom_objects={'angle bracket lambda angle bracket': lambda y_true, y_pred : my_loss(y_true, y_pred, lambda_const,i,T)})
Can anyone help me ? Thanks a lot
You first need to load your model with compile=False, then compile it. Like this:
model=load_model(model_path_old,custom_objects={'angle bracket lambda angle bracket': lambda y_true, y_pred : my_loss(y_true, y_pred, lambda_const,i,T)}, compile=False) \\
model.compile(loss=lambda y_true, y_pred: my_loss(y_true, y_pred, lambda_const,i,T), optimizer="adam", metrics=["acc"])
cememreakbas
on 22 Jul 2020
could anyone solve this problem please?
This is the definition of my loss function:
def my_loss(y_true,y_pred,lambda_const,i,T,task_size=2):
y_trueSoft=y_true[:,:(i)task_size]
y_predSoft=y_pred[:,:(i)task_size]/T
y_trueHard=y_true[:,(i)task_size:]
y_predHard=y_pred[:,(i)task_size:]
return lambda_constcategorical_crossentropy(y_trueSoft,y_predSoft) + (1-lambda_const)categorical_crossentropy(y_trueHard,y_predHard)
model_final.compile(loss=lambda y_true, y_pred: my_loss(y_true, y_pred, lambda_const,i,T), optimizer="adam", metrics=["acc"])
And when I want to load this model , I got this error,ValueError: Unknown loss function:.
I have changed the way I load this model, it writes below, but I still got the same error, I am so confusedmodel_tmp=load_model(model_path_old,custom_objects={'angle bracket lambda angle bracket': lambda y_true, y_pred : my_loss(y_true, y_pred, lambda_const,i,T)})
Can anyone help me ? Thanks a lot
You first need to load your model with compile=False, then compile it. Like this:
model=load_model(model_path_old,custom_objects={'angle bracket lambda angle bracket': lambda y_true, y_pred : my_loss(y_true, y_pred, lambda_const,i,T)}, compile=False) \\ model.compile(loss=lambda y_true, y_pred: my_loss(y_true, y_pred, lambda_const,i,T), optimizer="adam", metrics=["acc"])
Thank you so much! But it didn't work. I still got this Unknown loss function:
weilinapple
on 22 Jul 2020
could anyone solve this problem please?
This is the definition of my loss function:
def my_loss(y_true,y_pred,lambda_const,i,T,task_size=2):
y_trueSoft=y_true[:,:(i)task_size]
y_predSoft=y_pred[:,:(i)task_size]/T
y_trueHard=y_true[:,(i)task_size:]
y_predHard=y_pred[:,(i)_task_size:]
return lambda_const_categorical_crossentropy(y_trueSoft,y_predSoft) + (1-lambda_const)categorical_crossentropy(y_trueHard,y_predHard)
model_final.compile(loss=lambda y_true, y_pred: my_loss(y_true, y_pred, lambda_const,i,T), optimizer="adam", metrics=["acc"])
And when I want to load this model , I got this error,ValueError: Unknown loss function:.
I have changed the way I load this model, it writes below, but I still got the same error, I am so confusedmodel_tmp=load_model(model_path_old,custom_objects={'angle bracket lambda angle bracket': lambda y_true, y_pred : my_loss(y_true, y_pred, lambda_const,i,T)})
Can anyone help me ? Thanks a lot
You first need to load your model with compile=False, then compile it. Like this:
model=load_model(model_path_old,custom_objects={'angle bracket lambda angle bracket': lambda y_true, y_pred : my_loss(y_true, y_pred, lambda_const,i,T)}, compile=False) \\ model.compile(loss=lambda y_true, y_pred: my_loss(y_true, y_pred, lambda_const,i,T), optimizer="adam", metrics=["acc"])Thank you so much! But it didn't work. I still got this Unknown loss function:
Custom objects should be called while compiling, so they should not be called in load_model. The above code piece is not the right answer, sorry. It should be:
model=load_model(model_path, compile=False) \\ model.compile(loss=lambda y_true, y_pred: my_loss(y_true, y_pred, lambda_const,i,T), optimizer="adam", metrics=["acc"])
cememreakbas
on 22 Jul 2020
could anyone solve this problem please?
This is the definition of my loss function:
def my_loss(y_true,y_pred,lambda_const,i,T,task_size=2):
y_trueSoft=y_true[:,:(i)task_size]
y_predSoft=y_pred[:,:(i)task_size]/T
y_trueHard=y_true[:,(i)task_size:]
y_predHard=y_pred[:,(i)_task_size:]
return lambda_const_categorical_crossentropy(y_trueSoft,y_predSoft) + (1-lambda_const)categorical_crossentropy(y_trueHard,y_predHard)
model_final.compile(loss=lambda y_true, y_pred: my_loss(y_true, y_pred, lambda_const,i,T), optimizer="adam", metrics=["acc"])
And when I want to load this model , I got this error,ValueError: Unknown loss function:.
I have changed the way I load this model, it writes below, but I still got the same error, I am so confusedmodel_tmp=load_model(model_path_old,custom_objects={'angle bracket lambda angle bracket': lambda y_true, y_pred : my_loss(y_true, y_pred, lambda_const,i,T)})
Can anyone help me ? Thanks a lot
You first need to load your model with compile=False, then compile it. Like this:
model=load_model(model_path_old,custom_objects={'angle bracket lambda angle bracket': lambda y_true, y_pred : my_loss(y_true, y_pred, lambda_const,i,T)}, compile=False) \\ model.compile(loss=lambda y_true, y_pred: my_loss(y_true, y_pred, lambda_const,i,T), optimizer="adam", metrics=["acc"])Thank you so much! But it didn't work. I still got this Unknown loss function:
Custom objects should be called while compiling, so they should not be called in load_model. The above code piece is not the right answer, sorry. It should be:
model=load_model(model_path, compile=False) \\ model.compile(loss=lambda y_true, y_pred: my_loss(y_true, y_pred, lambda_const,i,T), optimizer="adam", metrics=["acc"])
Thank you so much. It works! Hope you have a nice day！
weilinapple
on 22 Jul 2020
@weilinapple you reading comprehension is way below required minimum for an engineer, consider a different career path.
Demetrio92
on 22 Jul 2020
@Demetrio92 your communication and social skills are below required minimum for an engineer, consider a different career path.
r0l1
on 11 Oct 2020
model.compile(optimizer='adam', metrics=['accuracy'],
loss=tf.keras.losses.SparseCategoricalCrossentropy())
this code works
SirErnestt
on 27 Nov 2020
Related issues
NancyZxll
·
3Comments
kylemcdonald
·
3Comments
farizrahman4u
·
3Comments
rantsandruse
·
3Comments
Imorton-zd
·
3Comments
Most helpful comment
I solved this problem by adding 'custom_bojects'
my loss function: