Related
I'm trying to have a layer in keras that takes a flat tensor x (doesn't have zero value in it and shape = (batch_size, units)) multiplied by a mask (of the same shape), and it will sort it in the way that masked values will be placed first in the output (the order of the elements value doesn't matter). For clarity here is an example (batch_size = 1, units = 8):
It seems simple but the problem is that I can't find a good solution. Any code or idea is appreciated.
My current code is as below, If you know a more efficient way please let me know.
class Sort(keras.layers.Layer):
def call(self, inputs):
x = inputs.numpy()
nonx, nony = x.nonzero() # idxs of nonzero elements
zero = [np.where(x == 0)[0][0], np.where(x == 0)[1][0]] # idx of first zero
x_shape = tf.shape(inputs)
result = np.zeros((x_shape[0], x_shape[1], 2), dtype = 'int') # mapping matrix
result[:, :, 0] += zero[0]
result[:, :, 1] += zero[1]
p = np.zeros((x_shape[0]), dtype = 'int')
for i, j in zip(nonx, nony):
result[i, p[i]] = [i, j]
p[i] += 1
y = tf.gather_nd(inputs, result)
return y
In a problem I want to solve using Tensorflow, I want to build a n-dimensional rank tensor that is 'diagonal' by blocks. That is, I want to generate a tensor object from a concatenation of low order tensors.
I have tried to define the whole tf.Variable tensor and then to impose the value 0 to some variables but Tensorflow does not allow assignments when working with variable tensors.
Moreover, I would want to create 'diagonal' tensors with the same independent variables, as, for example, using a stacked 2D representation, being A a 2 dimensional tensor:
T = [A, 0;0 , A]
My current source code:
shape1 = [3,3,10,10]
shape2 = [3,3]
i1 = tf.truncated_normal(shape1, stddev=1.0, dtype = tf.float32)
i2 = tf.truncated_normal(shape2, stddev=1.0, dtype = tf.float32)
A = tf.Variable(i1)
V = tf.Variable(i2)
for i in range(10):
for j in range(10):
if i != j:
A[:,:,i,j] = tf.zeros((3,3))
else:
A[:,:,i,j] = V
Of course, this code returns the error Variable object does not support item assignment.
What I want, at the end of the day, is to define a variable tensor such as:
T[:,:,i,j] = tf.zeros([D0,D1]), if i != j
and
T[:,:,i,j] = A, if i = j
with A = tf.variable([D0,D1])
Thank you very much in advance!
One way would be to use tf.stack, which converts a list of tensors of dimension n to a tensor of dimension n+1.
l = []
for i in range(10):
li = [V * 0.0 if i != j else V for j in range(10)]
Ai = tf.stack(li)
l.append(Ai)
A = tf.stack(l)
I found a peculiar property of lstm cell(not limited to lstm but I only examined with this) of tensorflow which has not been reported as far as I know.
I don't know whether it actually has, so I left this post in SO. Below is a toy code for this problem:
import tensorflow as tf
import numpy as np
import time
def network(input_list):
input,init_hidden_c,init_hidden_m = input_list
cell = tf.nn.rnn_cell.BasicLSTMCell(256, state_is_tuple=True)
init_hidden = tf.nn.rnn_cell.LSTMStateTuple(init_hidden_c, init_hidden_m)
states, hidden_cm = tf.nn.dynamic_rnn(cell, input, dtype=tf.float32, initial_state=init_hidden)
net = [v for v in tf.trainable_variables()]
return states, hidden_cm, net
def action(x, h_c, h_m):
t0 = time.time()
outputs, output_h = sess.run([rnn_states[:,-1:,:], rnn_hidden_cm], feed_dict={
rnn_input:x,
rnn_init_hidden_c: h_c,
rnn_init_hidden_m: h_m
})
dt = time.time() - t0
return outputs, output_h, dt
rnn_input = tf.placeholder("float", [None, None, 512])
rnn_init_hidden_c = tf.placeholder("float", [None,256])
rnn_init_hidden_m = tf.placeholder("float", [None,256])
rnn_input_list = [rnn_input, rnn_init_hidden_c, rnn_init_hidden_m]
rnn_states, rnn_hidden_cm, rnn_net = network(rnn_input_list)
feed_input = np.random.uniform(low=-1.,high=1.,size=(1,1,512))
feed_init_hidden_c = np.zeros(shape=(1,256))
feed_init_hidden_m = np.zeros(shape=(1,256))
sess = tf.Session()
sess.run(tf.global_variables_initializer())
for i in range(10000):
_, output_hidden_cm, deltat = action(feed_input, feed_init_hidden_c, feed_init_hidden_m)
if i % 10 == 0:
print 'Running time: ' + str(deltat)
(feed_init_hidden_c, feed_init_hidden_m) = output_hidden_cm
feed_input = np.random.uniform(low=-1.,high=1.,size=(1,1,512))
[Not important]What this code does is to generate an output from 'network()' function containing LSTM where the input's temporal dimension is 1, so output's is also 1, and pull in&out initial state for each step of running.
[Important] Looking the 'sess.run()' part. For some reasons in my real code, I happened to put [:,-1:,:] for 'rnn_states'. What is happening is then the time spent for each 'sess.run()' increases. For some inspection by my own, I found this slow down stems from that [:,-1:,:]. I just wanted to get the output at the last time step. If you do 'outputs, output_h = sess.run([rnn_states, rnn_hidden_cm], feed_dict{~' w/o [:,-1:,:] and take 'last_output = outputs[:,-1:,:]' after the 'sess.run()', then the slow down does not occur.
I do not know why this exponential increment of time happens with that [:,-1:,:] running. Is this the nature of tensorflow hasn't been documented but particularly slows down(may be adding more graph by its own?)?
Thank you, and hope this mistake not happen for other users by this post.
I encountered the same problem, with TensorFlow slowing down for each iteration I ran it, and found this question while trying to debug it. Here's a short description of my situation and how I solved it for future reference. Hopefully it can point someone in the right direction and save them some time.
In my case the problem was mainly that I didn't make use of feed_dict to supply the network state when executing sess.run(). Instead I redeclared outputs, final_state and prediction every iteration. The answer at https://github.com/tensorflow/tensorflow/issues/1439#issuecomment-194405649 made me realize how stupid that was... I was constantly creating new graph nodes in every iteration, making it all slower and slower. The problematic code looked something like this:
# defining the network
lstm_layer = rnn.BasicLSTMCell(num_units, forget_bias=1)
outputs, final_state = rnn.static_rnn(lstm_layer, input, initial_state=rnn_state, dtype='float32')
prediction = tf.nn.softmax(tf.matmul(outputs[-1], out_weights)+out_bias)
for input_data in data_seq:
# redeclaring, stupid stupid...
outputs, final_state = rnn.static_rnn(lstm_layer, input, initial_state=rnn_state, dtype='float32')
prediction = tf.nn.softmax(tf.matmul(outputs[-1], out_weights)+out_bias)
p, rnn_state = sess.run((prediction, final_state), feed_dict={x: input_data})
The solution was of course to only declare the nodes once in the beginning, and supply the new data with feed_dict. The code went from being half slow (> 15 ms in the beginning) and becoming slower for every iteration, to execute every iteration in around 1 ms. My new code looks something like this:
out_weights = tf.Variable(tf.random_normal([num_units, n_classes]), name="out_weights")
out_bias = tf.Variable(tf.random_normal([n_classes]), name="out_bias")
# placeholder for the network state
state_placeholder = tf.placeholder(tf.float32, [2, 1, num_units])
rnn_state = tf.nn.rnn_cell.LSTMStateTuple(state_placeholder[0], state_placeholder[1])
x = tf.placeholder('float', [None, 1, n_input])
input = tf.unstack(x, 1, 1)
# defining the network
lstm_layer = rnn.BasicLSTMCell(num_units, forget_bias=1)
outputs, final_state = rnn.static_rnn(lstm_layer, input, initial_state=rnn_state, dtype='float32')
prediction = tf.nn.softmax(tf.matmul(outputs[-1], out_weights)+out_bias)
# actual network state, which we input with feed_dict
_rnn_state = tf.nn.rnn_cell.LSTMStateTuple(np.zeros((1, num_units), dtype='float32'), np.zeros((1, num_units), dtype='float32'))
it = 0
for input_data in data_seq:
encl_input = [[input_data]]
p, _rnn_state = sess.run((prediction, final_state), feed_dict={x: encl_input, rnn_state: _rnn_state})
print("{} - {}".format(it, p))
it += 1
Moving the declaration out from the for loop also got rid of the problem which the OP sdr2002 had, doing a slice outputs[-1] in sess.run() inside the for loop.
As mentioned above, no sliced output for 'sess.run()' is much appreciated for this case.
def action(x, h_c, h_m):
t0 = time.time()
outputs, output_h = sess.run([rnn_states, rnn_hidden_cm], feed_dict={
rnn_input:x,
rnn_init_hidden_c: h_c,
rnn_init_hidden_m: h_m
})
outputs = outputs[:,-1:,:]
dt = time.time() - t0
return outputs, output_h, dt
I'm trying to train a sequence to sequence model using tensorflow. I see that in the tutorials, buckets help speed up training. So far I'm able to train using just one bucket, and also using just one gpu and multiple buckets using more or less out of the box code, but when I try to use multiple buckets with multiple gpus, I get an error stating
Invalid argument: You must feed a value for placeholder tensor 'gpu_scope_0/encoder50_gpu0' with dtype int32
From the error, I can tell that I'm not declaring the input_feed correctly, so it is expecting the input to be of the size of the largest bucket every time. I'm confused about why this is the case, though, because in the examples that I'm adapting, it does the same thing when initializing the placeholders for the input_feed. As far as I can tell, the tutorials also initialize up to the largest sized bucket, but this error doesn't happen when I use the tutorials' code.
The following is what I think is the relevant initialization code:
self.encoder_inputs = [[] for _ in xrange(self.num_gpus)]
self.decoder_inputs = [[] for _ in xrange(self.num_gpus)]
self.target_weights = [[] for _ in xrange(self.num_gpus)]
self.scope_prefix = "gpu_scope"
for j in xrange(self.num_gpus):
with tf.device("/gpu:%d" % (self.gpu_offset + j)):
with tf.name_scope('%s_%d' % (self.scope_prefix, j)) as scope:
for i in xrange(buckets[-1][0]): # Last bucket is the biggest one.
self.encoder_inputs[j].append(tf.placeholder(tf.int32, shape=[None],
name="encoder{0}_gpu{1}".format(i,j)))
for i in xrange(buckets[-1][1] + 1):
self.decoder_inputs[j].append(tf.placeholder(tf.int32, shape=[None],
name="decoder{0}_gpu{1}".format(i,j)))
self.target_weights[j].append(tf.placeholder(tf.float32, shape=[None],
name="weight{0}_gpu{1}".format(i,j)))
# Our targets are decoder inputs shifted by one.
self.losses = []
self.outputs = []
# The following loss computation creates the neural network. The specified
# device hosts the trainable tf parameters.
bucket = buckets[0]
i = 0
with tf.device(param_device):
output, loss = tf.nn.seq2seq.model_with_buckets(self.encoder_inputs[i], self.decoder_inputs[i],
[self.decoder_inputs[i][k + 1] for k in
xrange(len(self.decoder_inputs[i]) - 1)],
self.target_weights[0], buckets,
lambda x, y: seq2seq_f(x, y, True),
softmax_loss_function=self.softmax_loss_function)
bucket = buckets[0]
self.encoder_states = []
with tf.device('/gpu:%d' % self.gpu_offset):
with variable_scope.variable_scope(variable_scope.get_variable_scope(),
reuse=True):
self.encoder_outputs, self.encoder_states = get_encoder_outputs(self,
self.encoder_inputs[0])
if not forward_only:
self.grads = []
print ("past line 297")
done_once = False
for i in xrange(self.num_gpus):
with tf.device("/gpu:%d" % (self.gpu_offset + i)):
with tf.name_scope("%s_%d" % (self.scope_prefix, i)) as scope:
with variable_scope.variable_scope(variable_scope.get_variable_scope(), reuse=True):
#for j, bucket in enumerate(buckets):
output, loss = tf.nn.seq2seq.model_with_buckets(self.encoder_inputs[i],
self.decoder_inputs[i],
[self.decoder_inputs[i][k + 1] for k in
xrange(len(self.decoder_inputs[i]) - 1)],
self.target_weights[i], buckets,
lambda x, y: seq2seq_f(x, y, True),
softmax_loss_function=self.softmax_loss_function)
self.losses.append(loss)
self.outputs.append(output)
# Training outputs and losses.
if forward_only:
self.outputs, self.losses = tf.nn.seq2seq.model_with_buckets(
self.encoder_inputs, self.decoder_inputs,
[self.decoder_inputs[0][k + 1] for k in xrange(buckets[0][1])],
self.target_weights, buckets, lambda x, y: seq2seq_f(x, y, True),
softmax_loss_function=self.softmax_loss_function)
# If we use output projection, we need to project outputs for decoding.
if self.output_projection is not None:
for b in xrange(len(buckets)):
self.outputs[b] = [
tf.matmul(output, self.output_projection[0]) + self.output_projection[1]
for output in self.outputs[b]
]
else:
self.bucket_grads = []
self.gradient_norms = []
params = tf.trainable_variables()
opt = tf.train.GradientDescentOptimizer(self.learning_rate)
self.updates = []
with tf.device(aggregation_device):
for g in xrange(self.num_gpus):
for b in xrange(len(buckets)):
gradients = tf.gradients(self.losses[g][b], params)
clipped_grads, norm = tf.clip_by_global_norm(gradients, max_gradient_norm)
self.gradient_norms.append(norm)
self.updates.append(
opt.apply_gradients(zip(clipped_grads, params), global_step=self.global_step))
and the following is the relevant code when feeding in data:
input_feed = {}
for i in xrange(self.num_gpus):
for l in xrange(encoder_size):
input_feed[self.encoder_inputs[i][l].name] = encoder_inputs[i][l]
for l in xrange(decoder_size):
input_feed[self.decoder_inputs[i][l].name] = decoder_inputs[i][l]
input_feed[self.target_weights[i][l].name] = target_weights[i][l]
# Since our targets are decoder inputs shifted by one, we need one more.
last_target = self.decoder_inputs[i][decoder_size].name
input_feed[last_target] = np.zeros([self.batch_size], dtype=np.int32)
last_weight = self.target_weights[i][decoder_size].name
input_feed[last_weight] = np.zeros([self.batch_size], dtype=np.float32)
# Output feed: depends on whether we do a backward step or not.
if not forward_only:
output_feed = [self.updates[bucket_id], self.gradient_norms[bucket_id], self.losses[bucket_id]]
else:
output_feed = [self.losses[bucket_id]] # Loss for this batch.
for l in xrange(decoder_size): # Output logits.
output_feed.append(self.outputs[0][l])
Right now I'm considering just padding every input up to the bucket size, but I expect that this would lose some of the advantages of bucketing
Turns out the issue with this was not in the feeding of the placeholders, but was later on in my code where I referred to placeholders that weren't initialized. As far as I can tell when I fixed the later issues this error stopped
I'm trying to visualize the output of a convolutional layer in tensorflow using the function tf.image_summary. I'm already using it successfully in other instances (e. g. visualizing the input image), but have some difficulties reshaping the output here correctly. I have the following conv layer:
img_size = 256
x_image = tf.reshape(x, [-1,img_size, img_size,1], "sketch_image")
W_conv1 = weight_variable([5, 5, 1, 32])
b_conv1 = bias_variable([32])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
So the output of h_conv1 would have the shape [-1, img_size, img_size, 32]. Just using tf.image_summary("first_conv", tf.reshape(h_conv1, [-1, img_size, img_size, 1])) Doesn't account for the 32 different kernels, so I'm basically slicing through different feature maps here.
How can I reshape them correctly? Or is there another helper function I could use for including this output in the summary?
I don't know of a helper function but if you want to see all the filters you can pack them into one image with some fancy uses of tf.transpose.
So if you have a tensor that's images x ix x iy x channels
>>> V = tf.Variable()
>>> print V.get_shape()
TensorShape([Dimension(-1), Dimension(256), Dimension(256), Dimension(32)])
So in this example ix = 256, iy=256, channels=32
first slice off 1 image, and remove the image dimension
V = tf.slice(V,(0,0,0,0),(1,-1,-1,-1)) #V[0,...]
V = tf.reshape(V,(iy,ix,channels))
Next add a couple of pixels of zero padding around the image
ix += 4
iy += 4
V = tf.image.resize_image_with_crop_or_pad(image, iy, ix)
Then reshape so that instead of 32 channels you have 4x8 channels, lets call them cy=4 and cx=8.
V = tf.reshape(V,(iy,ix,cy,cx))
Now the tricky part. tf seems to return results in C-order, numpy's default.
The current order, if flattened, would list all the channels for the first pixel (iterating over cx and cy), before listing the channels of the second pixel (incrementing ix). Going across the rows of pixels (ix) before incrementing to the next row (iy).
We want the order that would lay out the images in a grid.
So you go across a row of an image (ix), before stepping along the row of channels (cx), when you hit the end of the row of channels you step to the next row in the image (iy) and when you run out or rows in the image you increment to the next row of channels (cy). so:
V = tf.transpose(V,(2,0,3,1)) #cy,iy,cx,ix
Personally I prefer np.einsum for fancy transposes, for readability, but it's not in tf yet.
newtensor = np.einsum('yxYX->YyXx',oldtensor)
anyway, now that the pixels are in the right order, we can safely flatten it into a 2d tensor:
# image_summary needs 4d input
V = tf.reshape(V,(1,cy*iy,cx*ix,1))
try tf.image_summary on that, you should get a grid of little images.
Below is an image of what one gets after following all the steps here.
In case someone would like to "jump" to numpy and visualize "there" here is an example how to display both Weights and processing result. All transformations are based on prev answer by mdaoust.
# to visualize 1st conv layer Weights
vv1 = sess.run(W_conv1)
# to visualize 1st conv layer output
vv2 = sess.run(h_conv1,feed_dict = {img_ph:x, keep_prob: 1.0})
vv2 = vv2[0,:,:,:] # in case of bunch out - slice first img
def vis_conv(v,ix,iy,ch,cy,cx, p = 0) :
v = np.reshape(v,(iy,ix,ch))
ix += 2
iy += 2
npad = ((1,1), (1,1), (0,0))
v = np.pad(v, pad_width=npad, mode='constant', constant_values=p)
v = np.reshape(v,(iy,ix,cy,cx))
v = np.transpose(v,(2,0,3,1)) #cy,iy,cx,ix
v = np.reshape(v,(cy*iy,cx*ix))
return v
# W_conv1 - weights
ix = 5 # data size
iy = 5
ch = 32
cy = 4 # grid from channels: 32 = 4x8
cx = 8
v = vis_conv(vv1,ix,iy,ch,cy,cx)
plt.figure(figsize = (8,8))
plt.imshow(v,cmap="Greys_r",interpolation='nearest')
# h_conv1 - processed image
ix = 30 # data size
iy = 30
v = vis_conv(vv2,ix,iy,ch,cy,cx)
plt.figure(figsize = (8,8))
plt.imshow(v,cmap="Greys_r",interpolation='nearest')
you may try to get convolution layer activation image this way:
h_conv1_features = tf.unpack(h_conv1, axis=3)
h_conv1_imgs = tf.expand_dims(tf.concat(1, h_conv1_features_padded), -1)
this gets one vertical stripe with all images concatenated vertically.
if you want them padded (in my case of relu activations to pad with white line):
h_conv1_features = tf.unpack(h_conv1, axis=3)
h_conv1_max = tf.reduce_max(h_conv1)
h_conv1_features_padded = map(lambda t: tf.pad(t-h_conv1_max, [[0,0],[0,1],[0,0]])+h_conv1_max, h_conv1_features)
h_conv1_imgs = tf.expand_dims(tf.concat(1, h_conv1_features_padded), -1)
I personally try to tile every 2d-filter in a single image.
For doing this -if i'm not terribly mistaken since I'm quite new to DL- I found out that it could be helpful to exploit the depth_to_space function, since it takes a 4d tensor
[batch, height, width, depth]
and produces an output of shape
[batch, height*block_size, width*block_size, depth/(block_size*block_size)]
Where block_size is the number of "tiles" in the output image. The only limitation to this is that the depth should be the square of block_size, which is an integer, otherwise it cannot "fill" the resulting image correctly.
A possible solution could be of padding the depth of the input tensor up to a depth that is accepted by the method, but I sill havn't tried this.
Another way, which I think very easy, is using the get_operation_by_name function. I had hard time visualizing the layers with other methods but this helped me.
#first, find out the operations, many of those are micro-operations such as add etc.
graph = tf.get_default_graph()
graph.get_operations()
#choose relevant operations
op_name = '...'
op = graph.get_operation_by_name(op_name)
out = sess.run([op.outputs[0]], feed_dict={x: img_batch, is_training: False})
#img_batch is a single image whose dimensions are (1,n,n,1).
# out is the output of the layer, do whatever you want with the output
#in my case, I wanted to see the output of a convolution layer
out2 = np.array(out)
print(out2.shape)
# determine, row, col, and fig size etc.
for each_depth in range(out2.shape[4]):
fig.add_subplot(rows, cols, each_depth+1)
plt.imshow(out2[0,0,:,:,each_depth], cmap='gray')
For example below is the input(colored cat) and output of the second conv layer in my model.
Note that I am aware this question is old and there are easier methods with Keras but for people who use an old model from other people (such as me), this may be useful.