search

tensorflow / privacy

Library for training machine learning models with privacy for training data
Apache License 2.0
1.93k stars 448 forks source link

Minimicrobatch size error, only function when the minibatchsize = = 1 #17

Closed MADONOKOUKI closed 5 years ago

MADONOKOUKI commented 5 years ago

I implemented the other neural network model and take loss function by dp_optimizer.DPGradientDescentGaussianOptimizer.

In that time, I successed when the num_microbatch is 1. But when the num_microbatch is over 1, I got an error.

Traceback (most recent call last):
  File "/home/madono/.pyenv/versions/anaconda3-2018.12/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/common_shapes.py", line 686, in _call_cpp_shape_fn_impl
    input_tensors_as_shapes, status)
  File "/home/madono/.pyenv/versions/anaconda3-2018.12/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/errors_impl.py", line 516, in __exit__
    c_api.TF_GetCode(self.status.status))
tensorflow.python.framework.errors_impl.InvalidArgumentError: Dimension size must be evenly divisible by 2 but is 1 for 'Reshape' (op: 'Reshape') with input shapes: [], [2] and with input tensors computed as partial shapes: input[1] = [2,?].

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "autoencoder_dp.py", line 70, in <module>
    population_size=60000).minimize(cost)
  File "/home/madono/.pyenv/versions/anaconda3-2018.12/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/training/optimizer.py", line 399, in minimize
    grad_loss=grad_loss)
  File "/home/madono/madono/test2/dpgan/privacy/optimizers/dp_optimizer.py", line 68, in compute_gradients
    microbatches_losses = tf.reshape(loss, [self._num_microbatches, -1])
  File "/home/madono/.pyenv/versions/anaconda3-2018.12/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/ops/gen_array_ops.py", line 5782, in reshape
    "Reshape", tensor=tensor, shape=shape, name=name)
  File "/home/madono/.pyenv/versions/anaconda3-2018.12/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
    op_def=op_def)
  File "/home/madono/.pyenv/versions/anaconda3-2018.12/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 3292, in create_op
    compute_device=compute_device)
  File "/home/madono/.pyenv/versions/anaconda3-2018.12/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 3332, in _create_op_helper
    set_shapes_for_outputs(op)
  File "/home/madono/.pyenv/versions/anaconda3-2018.12/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 2496, in set_shapes_for_outputs
    return _set_shapes_for_outputs(op)
  File "/home/madono/.pyenv/versions/anaconda3-2018.12/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 2469, in _set_shapes_for_outputs
    shapes = shape_func(op)
  File "/home/madono/.pyenv/versions/anaconda3-2018.12/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 2399, in call_with_requiring
    return call_cpp_shape_fn(op, require_shape_fn=True)
  File "/home/madono/.pyenv/versions/anaconda3-2018.12/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/common_shapes.py", line 627, in call_cpp_shape_fn
    require_shape_fn)
  File "/home/madono/.pyenv/versions/anaconda3-2018.12/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/framework/common_shapes.py", line 691, in _call_cpp_shape_fn_impl
    raise ValueError(err.message)
ValueError: Dimension size must be evenly divisible by 2 but is 1 for 'Reshape' (op: 'Reshape') with input shapes: [], [2] and with input tensors computed as partial shapes: input[1] = [2,?].

I implement like

optimizer = dp_optimizer.DPGradientDescentGaussianOptimizer(
          l2_norm_clip=1.0,
          noise_multiplier=1.1,
          num_microbatches=2,
          learning_rate=0.0002,
population_size=60000).minimize(cost)
#optimizer = tf.train.RMSPropOptimizer(learning_rate).minimize(cost)

# Initializing the variables
init = tf.initialize_all_variables()

# Explore trainable variables (weight_bias)
var = [v for v in tf.trainable_variables() if 'mimiciii/fc/autoencoder' in v.name] # (784, 128), (128,), (128, 784), (784,)
var_grad = tf.gradients(cost, var) # gradient of cost w.r.t. trainable variables, len(var_grad): 8, type(var_grad): list
norm_gradient_variables = []

# Launch the graph
with tf.Session() as sess:
    writer = tf.summary.FileWriter("./graph/my_graph", sess.graph)
    sess.run(init)
    total_batch = int(mnist.train.num_examples/batch_size)
    # Training cycle
    for epoch in range(training_epochs):
        # Loop over all batches
        for i in range(total_batch):
            batch_xs, batch_ys = mnist.train.next_batch(batch_size)
            # Run optimization op (backprop) and cost op (to get loss value)
            _, c = sess.run([optimizer, cost], feed_dict={X: batch_xs})
            var_grad_val = sess.run(var_grad, feed_dict={X: batch_xs})

            # var_grad_val = [var_grad_val[0], var_grad_val[2]] # no bias, change for different network
            if type(var_grad_val) != type([0]):  # if a is not a list, which indicate it contains only one weight matrix
                var_grad_val = [var_grad_val]
            norm_gradient_variables.append(norm_w(var_grad_val))  # compute the norm of all trainable variables
        # Display logs per epoch step
        if epoch % display_step == 0:
            print("Epoch:", '%04d' % (epoch+1),
                  "cost=", "{:.9f}".format(c))
schien1729 commented 5 years ago

Thanks for bringing this up, MADONOKOUKI. Can I ask what the value of batch_size is?

MADONOKOUKI commented 5 years ago

@schien1729 Thank you for replying. batch size is 256

schien1729 commented 5 years ago

It seems that the code thinks that the loss you're passing (the variable cost) has length 1, and therefore can't split it into two microbatches. cost should be a vector of length 256 (your batch size). Is it possible that you've turned it into a scalar, perhaps by using a reduce function?

MADONOKOUKI commented 5 years ago

I took mean squared error due to minimizing the error between input and output. So I wrote

ae_net = Autoencoder(inputDim, l2scale, compressDims, aeActivation, decompressDims,
                           dataType)  # autoencoder network
clipnorm = 5.0
standard_deviation = 0.0001

# tf Graph input (only pictures)
X = tf.placeholder("float", [None, inputDim])

# Construct model
loss, latent, output  = ae_net(X)
print(output.shape)
# Prediction
y_pred = output
# Targets (Labels) are the input data.
y_true = X

cost = tf.reduce_mean(tf.pow(y_true - y_pred, 2))
print(y_pred)
print(y_true)
cost = tf.losses.mean_squared_error(labels=y_true,
    predictions=y_pred,
)
# Calculate loss as a vector (to support microbatches in DP-SGD).
#vector_loss = tf.nn.sparse_softmax_cross_entropy_with_logits(
#labels=y_true, logits=y_pred)

#cost =  tf.reduce_mean(vector_loss)
# optimizer = tf.train.AdamOptimizer(learning_rate).minimize(cost) # in medgan
      # Use DP version of GradientDescentOptimizer. For illustration purposes,
      # we do that here by calling optimizer_from_args() explicitly, though DP
      # versions of standard optimizers are available in dp_optimizer.
optimizer = dp_optimizer.DPGradientDescentGaussianOptimizer(
          l2_norm_clip=1.0,
          noise_multiplier=1.1,
          num_microbatches=256,
          learning_rate=.15,
          population_size=60000).minimize(cost)

But is this cost function wrong? In this time, which cost function in tensorflow should I use?

schien1729 commented 5 years ago

The issue might be that by default, tf.losses.mean_squared_error aggregates the losses over all the examples it's given. Can you try this instead?

cost = tf.losses.mean_squared_error(labels=y_true,
    predictions=y_pred, reduction=Reduction.NONE
)
MADONOKOUKI commented 5 years ago

Thanks!!!!

cost = tf.losses.mean_squared_error(labels=y_true,
    predictions=y_pred, reduction="none"
)

I can train my code by this function. I can`t import Reduction, So I directlt write "None"

amanoel commented 5 years ago

Hi, I'm having this same error on the MNIST Keras example shipped with the package

ValueError: Dimension size must be evenly divisible by 250 but is 1 for 'training/TFOptimizer/Reshape' (op: 'Reshape') with input shapes: [], [2] and with input tensors computed as partial shapes: input[1] = [250,?].

but only in Tensorflow 1.13, in 1.12 it works fine. Any ideas? Thanks!

npapernot commented 5 years ago

Have you made the one liner modification documented at the top of the MNIST keras example ?

amanoel commented 5 years ago

Sorry, completely missed it :) I just copied the relevant code somewhere else, so I ended up not looking at the docstring.

So I guess this is just #21, never mind my comment!

Viserion-nlper commented 4 years ago

I also met the same problem How to solve this problem, please