Problem with tf.keras and PySyft

[aditya-malte]

Environment: TF 2.0, Python 3, Colab
Hi,
I am facing the following problem while installing pysyft-tensorflow:

When I try to import syft after installation(of pysyft-tensorflow), it returns a "no module found" error.
Please suggest a solution.
Thanks in advance.

[karlhigley]

pysyft-tensorflow doesn't automatically install pysyft, which needs to be installed manually.

[aditya-malte]

Hi,
Thanks for the quick response.
I had tried pip install pysyft too( with and without the udacity flag) but then there were two problems:

1. I lose tf2.0 and am instead am given an older 1.x version of tf.
2. While compiling a model using tf.keras it gives an error that did not occur before:
   “Operation object has no attribute : native_create_with_tf_output”
   Thank you

[jvmncs]

pip install syft[tensorflow] instead of pip install syft[udacity] is intended to fix (1)

it would be helpful if you were to supply steps to reproduce the issue, ideally some code for reference

[Madhura12gj]

pip install syft[tensorflow] instead of pip install syft[udacity] is intended to fix (1)

it would be helpful if you were to supply steps to reproduce the issue, ideally some code for reference

from pandas import read_csv
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder
from sklearn.metrics import confusion_matrix
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Dense
import numpy as np
import pandas as pd
import syft as sy
import tensorflow as tf

hook = sy.TensorFlowHook(tf)
bob = sy.VirtualWorker(hook, id="bob")
alice = sy.VirtualWorker(hook, id="alice")

#downloading the dataset
url='https://raw.githubusercontent.com/Madhura12gj/Hands-on-Pytorch/master/diabetes.csv'
df=pd.read_csv(url)
x = df.iloc[:, :-1].values
y = df.iloc[:, -1].values

#splitting it into testing and training set
x_train, x_test,y_train, y_test= train_test_split(x,y,test_size=0.2,random_state=1)
print(x_train.shape, x_test.shape, y_train.shape, y_test.shape)

x_train, y_train = tf.convert_to_tensor(x_train), tf.convert_to_tensor(y_train)
x_test, y_test = tf.convert_to_tensor(x_test), tf.convert_to_tensor(y_test)

n_features = x_train.shape[1]
print(n_features)

bob_dataset=(x_train[:300].send(bob),y_train[:300].send(bob))
alice_dataset =(x_train[300:].send(alice),y_train[300:].send(alice))

#defining the model
model = tf.keras.models.Sequential([tf.keras.layers.Flatten(input_shape=(n_features,)),
tf.keras.layers.Dense(16, activation='relu'),
tf.keras.layers.Dense(8, activation='relu'),
tf.keras.layers.Dense(1, activation='sigmoid')])

And right after this code we face the error.

[jvmncs]

sorry, I meant the installation code (since this seems to be an installation error)

[aditya-malte]

Hi,
So we had tried the following (and some combinations as well):

1. pip install syft[tensorflow]

2.pip install syft-tensorflow
pip install syft

3.pip install syft-tensorflow
pip install syft[udacity]

And some combinations. None of them seemed to work. and raised an error when the model was declared(something like:
"Operation" has no attribute native_create_tf_output...)
Surprisingly, if we kept the sy.VirtualWorked(id=..,hook) after model declaration the model compiled successfully, so I suppose there is an issue with the TF hooks.
Thanks

[Sharma-Pranav]

Hello,

I am also trying to install this software on a windows computer. I have tried installing in mostly following ways as far as I remember:

1. pip install syft-tensorflow
   pip install syft[udacity]
2. pip install syft-tensorflow
   pip install syft[tensorflow]
3. Installed everything in requirements files individually using Conda, then installed via pip install syft[tensorflow]
4. Installed everything in requirements files individually using Conda, then installed via pip install pysyft
5. pip install syft-tensorflow
   pip install syft[udacity]
6. pip install syft-tensorflow
   pip install pysyft
7. pip install syft[tensorflow]
   pip install syft-tensorflow
8. Also I tried both installation of pytorch 1.3.0 via both anaconda and pip - to check whether errors happen due to this also.

In my case mostly, I got "no module found" error or "ImportError: DLL load failed".

Any suggesions for installation would be a great help.

[nguyenduchuyvn]

Hello,
I got the same issue when installing syft- tensorflow in Colab (TF2.0, Python Version 3.6).
I use :
pip install syft-tensorflow
or
pip install syft[tensorflow]

but in two cases, I always got "no module found" error when import syft.
Any suggestions for this issue? Thanks you in advance

[KCC13]

Hi, I've encountered a similar issue on syft[tensorflow] 0.2.4. The following error message came out when I tried to import syft as sy.

syft.exceptions.UndefinedProtocolTypeError: tensorflow.python.framework.ops.EagerTensor is not defined in the protocol file

My environment setting:

1. mac osx 10.14.6
2. conda 4.8.2
3. python 3.6.10
4. syft[tensorflow] 0.2.4

Currently the latest version that I can successfully import syft is 0.2.0a2.

[Mesilenceki]

I do have the same problem now on google colab. Anyone fix it?

[harshitadd]

Hi, I've encountered a similar issue on syft[tensorflow] 0.2.4. The following error message came out when I tried to import syft as sy.

syft.exceptions.UndefinedProtocolTypeError: tensorflow.python.framework.ops.EagerTensor is not defined in the protocol file

My environment setting:

1. mac osx 10.14.6
2. conda 4.8.2
3. python 3.6.10
4. syft[tensorflow] 0.2.4

Currently the latest version that I can successfully import syft is 0.2.0a2.

Producing the same exception on google colab (with TensorFlow version - 2.2.0):
Tried -

• pip install syft[tensorflow]
  import syft as sy

• pip install syft-tensorflow
  import syft as sy

and the same exception is thrown in both cases. Any suggestions to work around this would be of much help.
Thanks in advance.

[KCC13]

@harshitadd @Mesilenceki
You have to install PySyft manually after pip install syft-tensorflow since PySyft-TensorFlow doesn't include it.

Due to this repository has not been maintained for a long time, it's inconsistent with the latest version of PySyft, so you have to install an older version of PySyft.

In my case, pip install syft==0.2.0a2 solves the import problem, or you can follow the instructions of README.md to build the code from another branch of pysyft (which I haven't tried).

Note that openmined has warned us that this repository will soon be deprecated in favor of PySyft, so it's probably not a good idea to build any decent work on PySyft-TensorFlow.

[harshitadd]

@harshitadd @Mesilenceki
You have to install PySyft manually after pip install syft-tensorflow since PySyft-TensorFlow doesn't include it.

Due to this repository has not been maintained for a long time, it's inconsistent with the latest version of PySyft, so you have to install an older version of PySyft.

In my case, pip install syft==0.2.0a2 solves the import problem, or you can follow the instructions of README.md to build the code from another branch of pysyft (which I haven't tried).

Note that openmined has warned us that this repository will soon be deprecated in favor of PySyft, so it's probably not a good idea to build any decent work on PySyft-TensorFlow.

Thank you for your prompt reply @skywind29. Will bear the suggestions in mind.

[apapadaki]

pip install syft==0.2.9 worked for me for the same issue (python 3.6)