mortendahl / privateml Goto Github PK

I am running the Feature Extraction.ipynb code. I am getting this error in extractor defining code.

Code:
 extractor = keras.models.Model(
    inputs=model.input, 
    outputs=flatten_layer.output
)

AssertionError Traceback (most recent call last)
in ()
1 flatten_layer = model.get_layer(index=7)
----> 2 assert flatten_layer.name.startswith('flatten_')
3
4 extractor = keras.models.Model(
5 inputs=model.input,

AssertionError:

There is no clear indication of what error i am facing. Can anybody help.

Hi,

I tried to run the image analysis on 64-bit Windows. When the process goes to decode(), the overflow error arises. I changed
map_negative_range = np.vectorize(lambda element: float(element) if element <= Q/2 else element - Q)
to
map_negative_range = np.vectorize(lambda element: float(element) if element <= Q/2 else element - Q, otypes=[np.float64])
The error goes away, but two warnings arise:
tensor.py:102: RuntimeWarning: overflow encountered in exp
tensor.py:81: RuntimeWarning: invalid value encountered in true_divide
I tried to ignore the warnings, but the prediction results are not good (50% - 60% I guess).
Any idea how to deal with?

Thanks!

Hello. I am running the spdz/Basic SPDZ.ipynb code. When I modified the test data, I was getting this error.

The following code is the test case that you gave. I changed the w to 6497992661811505122. In fact, this will trigger mod operation and causing the result of PublicElement is not equal to v+w.

Correspondingly, some operate has similar problems with triggering mod Q operation.

#  test case
v = 5
w =6497992661811505122

for x_type in [PublicElement, PrivateElement]:
    for y_type in [PublicElement, PrivateElement]:
        
        x = x_type(v)
        y = y_type(w)
       print('unwrap：%d---------(v + w)：%d' % (z.unwrap(), v + w))
        z = x + y; assert z.unwrap() == v + w
        z = x - y; assert z.unwrap() == v - w
        z = x * y; assert z.unwrap() == v * w
        z = x.square(); assert z.unwrap() == v * v

# log
unwrap：4---------(v + w)：6497992661811505127
  File "privateml/spdz/aa.py", line 196, in <module>
    assert z.unwrap() == v + w
AssertionError