TypeError: Too many parameter passed to theano function

[nwoeanhinnogaehr]

I'm getting the following error immediately after the first epoch completes. Any idea what the problem might be?

The shape of my inputs is (3, 256, 256).

Epoch 1/5
2976/3000 [============================>.] - ETA: 0s - loss: 0.1243 - acc: 0.9458Traceback (most recent call last):
  File "learn.py", line 36, in <module>
    batch_size=32)
  File "/home/nweninge/.local/lib/python2.7/site-packages/aetros/KerasIntegration.py", line 55, in overwritten_fit
    class_weight, sample_weight, **kwargs)
  File "/home/nweninge/.local/lib/python2.7/site-packages/keras/engine/training.py", line 1124, in fit
    callback_metrics=callback_metrics)
  File "/home/nweninge/.local/lib/python2.7/site-packages/keras/engine/training.py", line 862, in _fit_loop
    callbacks.on_epoch_end(epoch, epoch_logs)
  File "/home/nweninge/.local/lib/python2.7/site-packages/keras/callbacks.py", line 42, in on_epoch_end
    callback.on_epoch_end(epoch, logs)
  File "/home/nweninge/.local/lib/python2.7/site-packages/aetros/KerasLogger.py", line 218, in on_epoch_end
    images = self.build_insight_images()
  File "/home/nweninge/.local/lib/python2.7/site-packages/aetros/KerasLogger.py", line 300, in build_insight_images
    Y = fn(input_data_x_sample)[0]
  File "/home/nweninge/.local/lib/python2.7/site-packages/keras/backend/theano_backend.py", line 792, in __call__
    return self.function(*inputs)
  File "/usr/local/lib/python2.7/dist-packages/theano/compile/function_module.py", line 770, in __call__
    raise TypeError("Too many parameter passed to theano function")
TypeError: Too many parameter passed to theano function```

[gabrieldemarmiesse]

I had this bug too.
The strange thing is that, I tried to run my script without "KerasIntegration('gabrieldemarmiesse/test', model, insights = True)" and it worked fine so I guess the bug comes from either aetros, keras or theano. For more information, here is my script:

from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation
from keras.optimizers import SGD
import numpy as np
from aetros.KerasIntegration import KerasIntegration
import os

os.environ['API_KEY'] = "my api key"

data_dim = 20
nb_classes = 4

model = Sequential()

# Dense(64) is a fully-connected layer with 64 hidden units.
# in the first layer, you must specify the expected input data shape:
# here, 20-dimensional vectors.
model.add(Dense(64, input_dim=data_dim, init='uniform'))
model.add(Activation('tanh'))
model.add(Dropout(0.5))
model.add(Dense(64, init='uniform'))
model.add(Activation('tanh'))
model.add(Dropout(0.5))
model.add(Dense(nb_classes, init='uniform'))
model.add(Activation('softmax'))

model.compile(loss='categorical_crossentropy',
              optimizer='sgd',  
              metrics=["accuracy"])

# generate dummy training data
x_train = np.random.random((1000, data_dim))
y_train = np.random.random((1000, nb_classes))

# generate dummy test data
x_test = np.random.random((100, data_dim))
y_test = np.random.random((100, nb_classes))

KerasIntegration('gabrieldemarmiesse/test', model, insights = True )

model.fit(x_train, y_train,
          nb_epoch=200,
          batch_size=5)

score = model.evaluate(x_test, y_test, batch_size=16)

and here is the stacktrace:

C:\Users\smith>python y.py
Using Theano backend.
Training status changed to TRAINING
Training status changed to TRAINING
Epoch 1/200
 990/1000 [============================>.] - ETA: 0s - loss: 2.7636 - acc: 0.2444Traceback (most recent call last):
  File "y.py", line 43, in <module>
    batch_size=5)
  File "C:\Program Files\Anaconda3\lib\site-packages\aetros\KerasIntegration.py", line 55, in overwritten_fit
    class_weight, sample_weight, **kwargs)
  File "C:\Program Files\Anaconda3\lib\site-packages\keras\keras\models.py", line 652, in fit
    sample_weight=sample_weight)
  File "C:\Program Files\Anaconda3\lib\site-packages\keras\keras\engine\training.py", line 1111, in fit
    initial_epoch=initial_epoch)
  File "C:\Program Files\Anaconda3\lib\site-packages\keras\keras\engine\training.py", line 846, in _fit_loop
    callbacks.on_epoch_end(epoch, epoch_logs)
  File "C:\Program Files\Anaconda3\lib\site-packages\keras\keras\callbacks.py", line 43, in on_epoch_end
    callback.on_epoch_end(epoch, logs)
  File "C:\Program Files\Anaconda3\lib\site-packages\aetros\KerasLogger.py", line 218, in on_epoch_end
    images = self.build_insight_images()
  File "C:\Program Files\Anaconda3\lib\site-packages\aetros\KerasLogger.py", line 326, in build_insight_images
    Y = fn(input_data_x_sample)[0]
  File "C:\Program Files\Anaconda3\lib\site-packages\keras\keras\backend\theano_backend.py", line 811, in __call__
    return self.function(*inputs)
  File "C:\Program Files\Anaconda3\lib\site-packages\theano\theano\compile\function_module.py", line 772, in __call__
    raise TypeError("Too many parameter passed to theano function")
TypeError: Too many parameter passed to theano function

My system:
windows 10
Anaconda with python 3 (latest version right now)
keras (latest version right now)
Theano (latest version right now)
I'm working on a CPU. Cuda is not installed.

Here is the content of the .theanorc file:

[global]
floatX = float32
device = cpu

[nvcc]
compiler_bindir=C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\bin

and here is the content of the .keras/keras.json file:

{
    "floatx": "float32",
    "epsilon": 1e-07,
    "image_dim_ordering": "th",
    "backend": "theano"
}

Thank you.

[nwoeanhinnogaehr]

I think the issue might be that no validation data was passed to model.fit().

Try setting either validation_data or validation_split.

[gabrieldemarmiesse]

I changed my code and added the validation_split parameter. Here is the result:

Training status changed to TRAINING
0.00GB GPU memory used of 0.00GB, GeForce GTX 970, device id 0
Train on 900 samples, validate on 100 samples
Training status changed to TRAINING
Epoch 1/200
870/900 [============================>.] - ETA: 0s - loss: 2.7474 - acc: 0.2609Traceback (most recent call last):
  File "C:\Program Files\Anaconda3\lib\site-packages\PIL\ImageFile.py", line 466, in _save
    fh = fp.fileno()
io.UnsupportedOperation: fileno

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<stdin>", line 3, in <module>
  File "C:\Program Files\Anaconda3\lib\site-packages\aetros-0.3.5-py3.5.egg\aetros\KerasIntegration.py", line 55, in overwritten_fit
    class_weight, sample_weight, **kwargs)
  File "C:\Program Files\Anaconda3\lib\site-packages\keras\keras\models.py", line 652, in fit
    sample_weight=sample_weight)
  File "C:\Program Files\Anaconda3\lib\site-packages\keras\keras\engine\training.py", line 1113, in fit
    initial_epoch=initial_epoch)
  File "C:\Program Files\Anaconda3\lib\site-packages\keras\keras\engine\training.py", line 848, in _fit_loop
    callbacks.on_epoch_end(epoch, epoch_logs)
  File "C:\Program Files\Anaconda3\lib\site-packages\keras\keras\callbacks.py", line 45, in on_epoch_end
    callback.on_epoch_end(epoch, logs)
  File "C:\Program Files\Anaconda3\lib\site-packages\aetros-0.3.5-py3.5.egg\aetros\KerasLogger.py", line 218, in on_epoch_end
    images = self.build_insight_images()
  File "C:\Program Files\Anaconda3\lib\site-packages\aetros-0.3.5-py3.5.egg\aetros\KerasLogger.py", line 341, in build_insight_images
    'image': self.to_base64(image)
  File "C:\Program Files\Anaconda3\lib\site-packages\aetros-0.3.5-py3.5.egg\aetros\KerasLogger.py", line 469, in to_base64
    image.save(buffer, format="JPEG", optimize=True, quality=80)
  File "C:\Program Files\Anaconda3\lib\site-packages\PIL\Image.py", line 1698, in save
    save_handler(self, fp, filename)
  File "C:\Program Files\Anaconda3\lib\site-packages\PIL\JpegImagePlugin.py", line 725, in _save
    ImageFile._save(im, fp, [("jpeg", (0, 0)+im.size, 0, rawmode)], bufsize)
  File "C:\Program Files\Anaconda3\lib\site-packages\PIL\ImageFile.py", line 481, in _save
    fp.write(d)
TypeError: string argument expected, got 'bytes'

Maybe it's because insights=True only work with images and that this script is not using images?

[marcj]

Insights works currently only with images as input. This is a rather soft limitation that will be removed once we touch this functionality again.