How to implement the time sequence shown as following - tensorflow

The ST(spatial transform) has two input. the first one is Fi, which is fixed. The another is M, which is varied according to the output of last LSTM. The input of LTSM depends on the output of ST and the state of last LSTM.

The easiest way to do this is probably to write your own RNN cell. Another way is to use tf.raw_rnn. Check out this post or this excellent article.

Actually,I implement the network as following:
def build_model(self):
lstm_cell = tf.contrib.rnn.BasicLSTMCell(
num_units=self.config.num_lstm_units, state_is_tuple=True, reuse=True)
if self.mode == "train":
lstm_cell = tf.contrib.rnn.DropoutWrapper(
lstm_cell,
input_keep_prob=self.config.lstm_dropout_keep_prob,
output_keep_prob=self.config.lstm_dropout_keep_prob)
with tf.variable_scope("lstm", initializer=self.initializer) as lstm_scope:
zero_state = lstm_cell.zero_state(
batch_size=self.image_embeddings.get_shape()[0], dtype=tf.float32)
K = 5
C = 80
scores = tf.Variable(tf.random_normal(shape=[K, self.config.batch_size, C]), name="scores")
M = tf.Variable(tf.random_normal(shape=[K+1, self.config.batch_size, 2, 3]), name="M")
tf.assign(M[0], tf.convert_to_tensor([[1., 0., 0.], [0., 1., 0.]]))
lstm_input_size = 14
zk_size = 4096
hidden = zero_state
for k in range(0, K+1):
# Allow the LSTM variables to be reused.
if k > 0:
lstm_scope.reuse_variables()
f_k = spatial_transformer_network.spatial_transformer_network(self.image_embeddings, M[k])
f_k = tf.nn.max_pool(f_k, [1,2,2,1], [1,1,1,1], padding='VALID')
f_k = tf.layers.dense(tf.reshape(f_k, [self.config.batch_size, int(lstm_input_size * lstm_input_size / 4 * 512)]), 4096)
lstm_outputs, hidden = lstm_cell(f_k, hidden)
z_k = tf.layers.dense(hidden[0], zk_size, activation=tf.nn.relu)
if k != 0:
tf.assign(scores[k - 1], (tf.layers.dense(z_k, C)))
if k != K:
tf.assign(M[k + 1], (tf.reshape(tf.layers.dense(z_k, 6), [self.config.batch_size, 2, 3])))
tf.assign(M[k + 1, :, 0, 1], (tf.convert_to_tensor(0.)))
tf.assign(M[k + 1, :, 1, 0], (tf.convert_to_tensor(0.)))
But it throws error when it runs to
lstm_outputs, hidden = lstm_cell(f_k, hidden).
The error info is:
ValueError: Variable lstm/basic_lstm_cell/kernel does not exist, or was not created with tf.get_variable(). Did you mean to set reuse=tf.AUTO_REUSE in VarScope?
So what's the problem?

Related

tf.math.bincount - use min/max weight instead of weight sum

I would like to get a max/min value in tf.math.bincount instead of the weight sum. Basically currently it works as:
values = tf.constant([1,1,2,3,2,4,4,5])
weights = tf.constant([1,5,0,1,0,5,4,5])
tf.math.bincount(values, weights=weights) #[0 6 0 1 9 5]
However, I would like to get max/min for the conflicting weights instead, e.g. for max it should return:
[0 5 0 1 5 5]
It requires some finessing, but you can accomplish this as follows:
def bincount_with_max_weight(values: tf.Tensor, weights: tf.Tensor) -> tf.Tensor:
_range = tf.range(tf.reduce_max(values) + 1)
return tf.map_fn(lambda x: tf.maximum(
tf.reduce_max(tf.gather(weights, tf.where(tf.equal(values, x)))), 0), _range)
The output for the example case is:
[0 5 0 1 5 5]
Breaking it down, the first line computes the range of values in values:
_range = tf.range(tf.reduce_max(values) + 1)
and in the second line, the maximum of weight is computed per element in _range using tf.map_fn with tf.where, which retrieves indices where the clause is true, and tf.gather, which retrieves the values corresponding to supplied indices.
The tf.maximum wraps the output to handle the case where the element does not exist in values i.e; in the example case, 0 does not exist in values so the output without tf.maximum would be INT_MIN for 0:
[-2147483648 5 0 1 5 5]
This could also be applied on the final result tensor instead of per element:
def bincount_with_max_weight(values: tf.Tensor, weights: tf.Tensor) -> tf.Tensor:
_range = tf.range(tf.reduce_max(values) + 1)
result = tf.map_fn(lambda x:
tf.reduce_max(tf.gather(weights, tf.where(tf.equal(values, x)))), _range)
return tf.maximum(result, 0)
Note that this would not work if negative weights are utilized - in that case, tf.where can be used for comparing against the minimum integer value (tf.int32.min in the example, although this can be applied for any numeric dtype) instead of applying tf.maximum:
def bincount_with_max_weight(values: tf.Tensor, weights: tf.Tensor) -> tf.Tensor:
_range = tf.range(tf.reduce_max(values) + 1)
result = tf.map_fn(lambda x:
tf.reduce_max(tf.gather(weights, tf.where(tf.equal(values, x)))), _range)
return tf.where(tf.equal(result, tf.int32.min), 0, result)
Update
For handling the 2D Tensor case, we can use tf.map_fn to apply the maximum weight function to each pair of values and weights in the batch:
def bincount_with_max_weight(values: tf.Tensor, weights: tf.Tensor, axis: Optional[int] = None) -> tf.Tensor:
_range = tf.range(tf.reduce_max(values) + 1)
def mapping_function(x: int, _values: tf.Tensor, _weights: tf.Tensor) -> tf.Tensor:
return tf.reduce_max(tf.gather(_weights, tf.where(tf.equal(_values, x))))
if axis == -1:
result = tf.map_fn(lambda pair: tf.map_fn(lambda x: mapping_function(x, *pair), _range), (values, weights),
dtype=tf.int32)
else:
result = tf.map_fn(lambda x: mapping_function(x, values, weights), _range)
return tf.where(tf.equal(result, tf.int32.min), 0, result)
For the 2D example provided:
values = tf.constant([[1, 1, 2, 3], [2, 1, 4, 5]])
weights = tf.constant([[1, 5, 0, 1], [0, 5, 4, 5]])
print(bincount_with_max_weight(values, weights, axis=-1))
The output is:
tf.Tensor(
[[0 5 0 1 0 0]
[0 5 0 0 4 5]], shape=(2, 6), dtype=int32)
This implementation is a generalization of the approach originally described - if axis is omitted, it will compute results for the 1D case.
For Faster Execution try this,
values = tf.constant([[1,1,2,3], [2,1,4,5]])
weights = tf.constant([[1,5,0,1], [0,5,4,5]])
def find_max_bins(output , values , weights):
np.maximum.at(output , values , weights)
return output
#tf.function(input_signature=[tf.TensorSpec(shape=[None], dtype = tf.float32),
tf.TensorSpec(shape=[None], dtype = tf.int32),
tf.TensorSpec(shape=[None], dtype = tf.int32)
])
def tf_function(output , values , weights):
print(values)
y = tf.numpy_function(find_max_bins, [output , values , weights], tf.float32)
return y
length = np.max(values)+1
initial_value = [0 for x in range(length)]
variable = tf.Variable(initial_value = initial_value, shape=(length) , dtype=tf.float32)
for i , (value , weight) in enumerate(zip(values , weights)):
if(i > 0):
output = tf.stack([output , tf_function(variable , value , weight)] , 0)
else:
output = tf_function(variable , value , weight)
variable.assign_sub(initial_value)
Output:
<tf.Tensor: shape=(2, 6), dtype=float32, numpy=
array([[0., 5., 0., 1., 0., 0.],
[0., 5., 0., 0., 4., 5.]], dtype=float32)>

How to implement tf.gather_nd in Pytorch with the argument batch_dims?

I have been doing a project on image matching, so I need to find correspondences between 2 images. To get descriptors, I will need a interpolate function. However, when I read about a equivalent function which is done in Tensorflow, I still don’t get how to implement tf.gather_nd(parmas, indices, barch_dims) in Pytorch. Especially when there is a argument: batch_dims. I have gone through stackoverflow and there is no perfect equivalence yet.
The referred interpolate function in Tensorflow is below and I have been trying to implement this in Pytorch Arguments' information is below:
inputs is a dense feature map[i] from a for loop of batch size, which means it is 3D[H, W, C](in pytorch is [C, H, W])
pos is a set of random point coordinate shapes like [[i, j], [i, j],...,[i, j]], so it is 2D when it goes in interpolate function(in pytorch is [[i,i,...,i], [j,j,...,j]])
and it then expands both of their dimensions when they get into this function
I just want a perfect implement of tf.gather_nd with argument batch_dims. Thank you!
And here's a simple example of using it:
pos = tf.ones((12, 2)) ## stands for a set of coordinates [[i, i,…, i], [j, j,…, j]]
inputs = tf.ones((4, 4, 128)) ## stands for [H, W, C] of dense feature map
outputs = interpolate(pos, inputs, batched=False)
print(outputs.get_shape()) # We get (12, 128) here
interpolate function (tf version):
def interpolate(pos, inputs, nd=True):
pos = tf.expand_dims(pos, 0)
inputs = tf.expand_dims(inputs, 0)
h = tf.shape(inputs)[1]
w = tf.shape(inputs)[2]
i = pos[:, :, 0]
j = pos[:, :, 1]
i_top_left = tf.clip_by_value(tf.cast(tf.math.floor(i), tf.int32), 0, h - 1)
j_top_left = tf.clip_by_value(tf.cast(tf.math.floor(j), tf.int32), 0, w - 1)
i_top_right = tf.clip_by_value(tf.cast(tf.math.floor(i), tf.int32), 0, h - 1)
j_top_right = tf.clip_by_value(tf.cast(tf.math.ceil(j), tf.int32), 0, w - 1)
i_bottom_left = tf.clip_by_value(tf.cast(tf.math.ceil(i), tf.int32), 0, h - 1)
j_bottom_left = tf.clip_by_value(tf.cast(tf.math.floor(j), tf.int32), 0, w - 1)
i_bottom_right = tf.clip_by_value(tf.cast(tf.math.ceil(i), tf.int32), 0, h - 1)
j_bottom_right = tf.clip_by_value(tf.cast(tf.math.ceil(j), tf.int32), 0, w - 1)
dist_i_top_left = i - tf.cast(i_top_left, tf.float32)
dist_j_top_left = j - tf.cast(j_top_left, tf.float32)
w_top_left = (1 - dist_i_top_left) * (1 - dist_j_top_left)
w_top_right = (1 - dist_i_top_left) * dist_j_top_left
w_bottom_left = dist_i_top_left * (1 - dist_j_top_left)
w_bottom_right = dist_i_top_left * dist_j_top_left
if nd:
w_top_left = w_top_left[..., None]
w_top_right = w_top_right[..., None]
w_bottom_left = w_bottom_left[..., None]
w_bottom_right = w_bottom_right[..., None]
interpolated_val = (
w_top_left * tf.gather_nd(inputs, tf.stack([i_top_left, j_top_left], axis=-1), batch_dims=1) +
w_top_right * tf.gather_nd(inputs, tf.stack([i_top_right, j_top_right], axis=-1), batch_dims=1) +
w_bottom_left * tf.gather_nd(inputs, tf.stack([i_bottom_left, j_bottom_left], axis=-1), batch_dims=1) +
w_bottom_right * tf.gather_nd(inputs, tf.stack([i_bottom_right, j_bottom_right], axis=-1), batch_dims=1)
)
interpolated_val = tf.squeeze(interpolated_val, axis=0)
return interpolated_val
As far as I'm aware there is no directly equivalent of tf.gather_nd in PyTorch and implementing a generic version with batch_dims is not that simple. However, you likely don't need a generic version, and given the context of your interpolate function, a version for [C, H, W] would suffice.
At the beginning of interpolate you add a singular dimension to the front, which is the batch dimension. Setting batch_dims=1 in tf.gather_nd means there is one batch dimension at the beginning, therefore it applies it per batch, i.e. it indexes inputs[0] with pos[0] etc. There is no benefit of adding a singular batch dimension, because you could have just used the direct computation.
# Adding singular batch dimension
# Shape: [1, num_pos, 2]
pos = tf.expand_dims(pos, 0)
# Shape: [1, H, W, C]
inputs = tf.expand_dims(inputs, 0)
batched_result = tf.gather_nd(inputs, pos, batch_dims=1)
single_result = tf.gater_nd(inputs[0], pos[0])
# The first element in the batched result is the same as the single result
# Hence there is no benefit to adding a singular batch dimension.
tf.reduce_all(batched_result[0] == single_result) # => True
Single version
In PyTorch the implementation for [H, W, C] can be done with Python's indexing. While PyTorch usually uses [C, H, W] for images, it's only a matter of what dimension to index, but let's keep them the same as in TensorFlow for the sake of comparison. If you were to index them manually, you would do it as such: inputs[pos_h[0], pos_w[0]], inputs[pos_h[1], pos_w[1]] and so on. PyTorch allows you to do that automatically by providing the indices as lists: inputs[pos_h, pos_w], where pos_h and pos_w have the same length. All you need to do is split your pos into two separate tensors, one for the indices along the height dimension and the other along the width dimension, which you also did in the TensorFlow version.
inputs = torch.randn(4, 4, 128)
# Random positions 0-3, shape: [12, 2]
pos = torch.randint(4, (12, 2))
# Positions split by dimension
pos_h = pos[:, 0]
pos_w = pos[:, 1]
# Index the inputs with the indices per dimension
gathered = inputs[pos_h, pos_w]
# Verify that it's identical to TensorFlow's output
inputs_tf = tf.convert_to_tensor(inputs.numpy())
pos_tf = tf.convert_to_tensor(pos.numpy())
gathered_tf = tf.gather_nd(inputs_tf, pos_tf)
gathered_tf = torch.from_numpy(gathered_tf.numpy())
torch.equal(gathered_tf, gathered) # => True
If you want to apply it to a tensor of size [C, H, W] instead, you only need to change the dimensions you want to index:
# For [H, W, C]
gathered = inputs[pos_h, pos_w]
# For [C, H, W]
gathered = inputs[:, pos_h, pos_w]
Batched version
Making it a batched batched version (for [N, H, W, C] or [N, C, H, W]) is not that difficult, and using that is more appropriate, since you're dealing with batches anyway. The only tricky part is that each element in the batch should only be applied to the corresponding batch. For this the batch dimensions needs to be enumerated, which can be done with torch.arange. The batch enumeration is just the list with the batch indices, which will be combined with the pos_h and pos_w indices, resulting in inputs[0, pos_h[0, 0], pos_h[0, 0]], inputs[0, pos_h[0, 1], pos_h[0, 1]] ... inputs[1, pos_h[1, 0], pos_h[1, 0]] etc.
batch_size = 3
inputs = torch.randn(batch_size, 4, 4, 128)
# Random positions 0-3, different for each batch, shape: [3, 12, 2]
pos = torch.randint(4, (batch_size, 12, 2))
# Positions split by dimension
pos_h = pos[:, :, 0]
pos_w = pos[:, :, 1]
batch_enumeration = torch.arange(batch_size) # => [0, 1, 2]
# pos_h and pos_w have shape [3, 12], so the batch enumeration needs to be
# repeated 12 times per batch.
# Unsqueeze to get shape [3, 1], now the 1 could be repeated to 12, but
# broadcasting will do that automatically.
batch_enumeration = batch_enumeration.unsqueeze(1)
# Index the inputs with the indices per dimension
gathered = inputs[batch_enumeration, pos_h, pos_w]
# Again, verify that it's identical to TensorFlow's output
inputs_tf = tf.convert_to_tensor(inputs.numpy())
pos_tf = tf.convert_to_tensor(pos.numpy())
# This time with batch_dims=1
gathered_tf = tf.gather_nd(inputs_tf, pos_tf, batch_dims=1)
gathered_tf = torch.from_numpy(gathered_tf.numpy())
torch.equal(gathered_tf, gathered) # => True
Again, for [N, C, H, W], only the dimensions that are indexed need to be changed:
# For [N, H, W, C]
gathered = inputs[batch_enumeration, pos_h, pos_w]
# For [N, C, H, W]
gathered = inputs[batch_enumeration, :, pos_h, pos_w]
Just a little side note on the interpolate implementation, rounding the positions (floor and ceil respectively) doesn't make sense, because indices must be integers, so it has no effect, as long as your positions are actual indices. That also results in i_top_left and i_bottom_left being the same value, but even if they are to be rounded differently, they are always 1 position apart. Furthermore, i_top_left and i_top_right are literally the same. I don't think that this function produces a meaningful output. I don't know what you're trying to achieve, but if you're looking for image interpolation you could have a look at torch.nn.functional.interpolate.
This is just an extension of Michael Jungo's batched version answer when pos is 2D array instead of 1D array (excluding batch dimension).
bs = 2
H = 4
W = 6
C = 3
inputs = torch.randn(bs, H, W, C)
pos_h = torch.randint(H, (bs, H, W))
pos_w = torch.randint(W, (bs, H, W))
batch_enumeration = torch.arange(bs)
batch_enumeration = batch_enumeration.unsqueeze(1).unsqueeze(2)
inputs.shape
Out[34]: torch.Size([2, 4, 6, 3])
pos_h.shape
Out[35]: torch.Size([2, 4, 6])
pos_w.shape
Out[36]: torch.Size([2, 4, 6])
batch_enumeration.shape
Out[37]: torch.Size([2, 1, 1])
gathered = inputs[batch_enumeration, pos_h, pos_w]
For channel first, we also need to enumerate channels
inputs = torch.randn(bs, C, H, W)
pos_h = torch.randint(H, (bs, 1, H, W))
pos_w = torch.randint(W, (bs, 1, H, W))
batch_enumeration = torch.arange(bs)
batch_enumeration = batch_enumeration.unsqueeze(1).unsqueeze(2).unsqueeze(3)
channel_enumeration = torch.arange(C)
channel_enumeration = channel_enumeration.unsqueeze(0).unsqueeze(2).unsqueeze(3)
inputs.shape
Out[49]: torch.Size([2, 3, 4, 6])
pos_h.shape
Out[50]: torch.Size([2, 1, 4, 6])
pos_w.shape
Out[51]: torch.Size([2, 1, 4, 6])
batch_enumeration.shape
Out[52]: torch.Size([2, 1, 1, 1])
channel_enumeration.shape
Out[57]: torch.Size([1, 3, 1, 1])
gathered = inputs[batch_enumeration, channel_enumeration, pos_h, pos_w]
gathered.shape
Out[59]: torch.Size([2, 3, 4, 6])
Let's verify
inputs_np = inputs.numpy()
pos_h_np = pos_h.numpy()
pos_w_np = pos_w.numpy()
gathered_np = gathered.numpy()
pos_h_np[0,0,0,0]
Out[68]: 0
pos_w_np[0,0,0,0]
Out[69]: 3
inputs_np[0,:,0,3]
Out[71]: array([ 0.79122806, -2.190181 , -0.16741803], dtype=float32)
gathered_np[0,:,0,0]
Out[72]: array([ 0.79122806, -2.190181 , -0.16741803], dtype=float32)
pos_h_np[1,0,3,4]
Out[73]: 1
pos_w_np[1,0,3,4]
Out[74]: 2
inputs_np[1,:,1,2]
Out[75]: array([ 0.9282498 , -0.34945545, 0.9136222 ], dtype=float32)
gathered_np[1,:,3,4]
Out[77]: array([ 0.9282498 , -0.34945545, 0.9136222 ], dtype=float32)
I improved the answer from Michael Jungo's implementation. Now it supports arbitrary leading batch dimensions.
def gather_nd_torch(params, indices, batch_dim=1):
""" A PyTorch porting of tensorflow.gather_nd
This implementation can handle leading batch dimensions in params, see below for detailed explanation.
The majority of this implementation is from Michael Jungo # https://stackoverflow.com/a/61810047/6670143
I just ported it compatible to leading batch dimension.
Args:
params: a tensor of dimension [b1, ..., bn, g1, ..., gm, c].
indices: a tensor of dimension [b1, ..., bn, x, m]
batch_dim: indicate how many batch dimension you have, in the above example, batch_dim = n.
Returns:
gathered: a tensor of dimension [b1, ..., bn, x, c].
Example:
>>> batch_size = 5
>>> inputs = torch.randn(batch_size, batch_size, batch_size, 4, 4, 4, 32)
>>> pos = torch.randint(4, (batch_size, batch_size, batch_size, 12, 3))
>>> gathered = gather_nd_torch(inputs, pos, batch_dim=3)
>>> gathered.shape
torch.Size([5, 5, 5, 12, 32])
>>> inputs_tf = tf.convert_to_tensor(inputs.numpy())
>>> pos_tf = tf.convert_to_tensor(pos.numpy())
>>> gathered_tf = tf.gather_nd(inputs_tf, pos_tf, batch_dims=3)
>>> gathered_tf.shape
TensorShape([5, 5, 5, 12, 32])
>>> gathered_tf = torch.from_numpy(gathered_tf.numpy())
>>> torch.equal(gathered_tf, gathered)
True
"""
batch_dims = params.size()[:batch_dim] # [b1, ..., bn]
batch_size = np.cumprod(list(batch_dims))[-1] # b1 * ... * bn
c_dim = params.size()[-1] # c
grid_dims = params.size()[batch_dim:-1] # [g1, ..., gm]
n_indices = indices.size(-2) # x
n_pos = indices.size(-1) # m
# reshape leadning batch dims to a single batch dim
params = params.reshape(batch_size, *grid_dims, c_dim)
indices = indices.reshape(batch_size, n_indices, n_pos)
# build gather indices
# gather for each of the data point in this "batch"
batch_enumeration = torch.arange(batch_size).unsqueeze(1)
gather_dims = [indices[:, :, i] for i in range(len(grid_dims))]
gather_dims.insert(0, batch_enumeration)
gathered = params[gather_dims]
# reshape back to the shape with leading batch dims
gathered = gathered.reshape(*batch_dims, n_indices, c_dim)
return gathered
I have also made a demo Colab notebook, you can check it here. This implementation is way faster than TF's original implementation according to my poor speed test on Colab server with a GPU instance.

Discrepancy between tensorflow's conv1d and pytorch's conv1d

I am trying to import some pytorch code to tensorflow, I came to know that torch.nn.functional.conv1d() is tf.nn.conv1d() but I am afraid there are still some discrepancies in tf's versions. Specifically, I cannot find the group parameter in tf.conv1d. For example: the following codes output two different results:
Pytorch:
inputs = torch.Tensor([[[1, 1, 1, 1],[2, 2, 2, 2],[3, 3, 3, 3]]]) #batch_sizex seq_length x embed_dim,
inputs = inputs.transpose(2,1) #batch_size x embed_dim x seq_length
batch_size, embed_dim, seq_length = inputs.size()
kernel_size = 3
in_channels = 2
out_channels = in_channels
weight = torch.ones(out_channels, 1, kernel_size)
inputs = inputs.contiguous().view(-1, in_channels, seq_length) #batch_size*embed_dim/in_channels x in_channels x seq_length
inputs = F.pad(inputs, (kernel_size-1,0), 'constant', 0)
output = F.conv1d(inputs, weight, padding=0, groups=in_channels)
output = output.contiguous().view(batch_size, embed_dim, seq_length).transpose(2,1)
Output:
tensor([[[1., 1., 1., 1.],
[3., 3., 3., 3.],
[6., 6., 6., 6.]]])
Tensorflow:
inputs = tf.constant([[[1, 1, 1, 1],[2, 2, 2, 2],[3, 3, 3, 3]]], dtype=tf.float32) #batch_sizex seq_length x embed_dim
inputs = tf.transpose(inputs, perm=[0,2,1])
batch_size, embed_dim, seq_length = inputs.get_shape()
print(batch_size, seq_length, embed_dim)
kernel_size = 3
in_channels = 2
out_channels = in_channels
weight = tf.ones([kernel_size, in_channels, out_channels])
inputs = tf.reshape(inputs, [(batch_size*embed_dim)//in_channels, in_channels, seq_length], name='inputs')
inputs = tf.transpose(inputs, perm=[0, 2, 1])
padding = [[0, 0], [(kernel_size - 1), 0], [0, 0]]
padded = tf.pad(inputs, padding)
res = tf.nn.conv1d(padded, weight, 1, 'VALID')
res = tf.transpose(res, perm=[0, 2, 1])
res = tf.reshape(res, [batch_size, embed_dim, seq_length])
res = tf.transpose(res, perm=[0, 2, 1])
print(res)
Output:
[[[ 2. 2. 2. 2.]
[ 6. 6. 6. 6.]
[12. 12. 12. 12.]]], shape=(1, 3, 4), dtype=float32)
Different results
There is no discrepancy between those versions, you are just setting up different things. To get exactly same results as in Tensorflow change the lines specifying weights to:
weight = torch.ones(out_channels, 2, kernel_size)
, because your input has two input channels, as you have correctly declared in TF:
weight = tf.ones([kernel_size, in_channels, out_channels])
Groups parameter
You have misunderstood what is groups parameter responsible for in pytorch. It restricts the number of channels each filter uses (in this case only one as 2 input_channels divided by 2 give us one).
See here for more intuitive explanation for 2D convolution.

Missing placeholder value crash occurs during global variable initialization?

On tensorflow 1.4. I'm getting the you need a value for placeholder tensor... error. Thing is I am feeding this tensor like
feats = np.reshape(feats, (-1, var1, feat_dim, 1))
_, outlogits = sess.run([train_step, logits], feed_dict={inp_layer: feats,
targs: targets,
eta: 1e-4})
(normally I'd want to reshape inside of the graph but for debugging purposes I've taken that out)
placeholder:
inp_layer = tf.placeholder(tf.float32, shape=[None, var1, feat_dim, 1])
The error says: You must feed a value for placeholder tensor 'Placeholder' with dtype float and shape [?,66,200,1]
And this error occurs when I run sess.run(tf.global_variables_initializer()), so it hasn't even gotten to the part where it should start thinking about placeholders and yet it's complaining about them?!
I think it might have to do with the fact that one of my layer sizes is dependent on a placeholder.. (I have validate_shape=False for the weights though). Will add more code.
edit: example code that fails, point out where I think the problem is coming from (remember code fails on global variable init):
!edit2: YUP the problem is that line. The question then becomes how can I have a graph where the dimensions of one the weights (and therefore the outputs) is dynamic.
train_feats = '..'
train_json = '..'
feat_dim = 200
var1 = 20
batch_size = 64
inp_layer = tf.placeholder(tf.float32, shape=[None, var1, feat_dim, 1])
targs = tf.placeholder(tf.int64, shape=[None])
eta = tf.placeholder(tf.float32)
chunk_size = 3
w1 = init_weight([chunk_size, feat_dim, 1, 32])
b1 = tf.zeros([32])
a1 = conv_layer(inp_layer, w1, b1, stride=3, padding='VALID')
chunk_size = tf.shape(a1)[1] <==== # ! IS THE PROBLEM !
w5 = init_weight([chunk_size, 1, 32, 12])
b5 = tf.zeros([12])
a5 = conv_layer(a1, w5, b5, stride=1, padding='VALID', act=False)
logits_ = tf.reshape(a5, [-1, 12])
softmax = tf.nn.softmax(logits_)
cross_ent = tf.reduce_sum(tf.nn.sparse_softmax_cross_entropy_with_logits(labels=targs,
logits=logits_))
train_step = tf.train.AdamOptimizer(eta).minimize(cross_ent)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for feats, targets in batch_gen(train_feats, train_json, var1, feat_dim):
feats = np.reshape(feats, (var1, var1, feat_dim, 1))
sess.run(train_step, feed_dict={inp_layer: bla,
targs: targets,
eta: 1e-4})
def init_weight(shape):
return tf.Variable(tf.truncated_normal(shape, stddev=0.01), validate_shape=False)
def conv_layer(x, w, b, stride, padding, act=True):
# striding over the features
if act:
return tf.nn.elu(tf.nn.conv2d(x, w, [1, stride, 1, 1], padding) + b)
else:
return tf.nn.conv2d(x, w, [1, stride, 1, 1], padding) + b
The line
chunk_size = tf.shape(a1)[1]
tf.shape extracts the runtime shape of a1, not the static shape known at graph definition time. Since a1 is the result of the convolution between inp_layer and w1, when you reference a1 you need also to resolve inp_layer. Because inp_layer is a placeholder, your error follows.
Since you're interested in the second dimension of a1 that is know at graph definition time, you can just use:
chunk_size = a1.shape[1].value
to extract the correct dimension value.

TensorFlow: Plotting a graph that describes data points and decision boundary

I would like to achieve something similar:
https://rootpy.github.io/root_numpy/_images/plot_multiclass_1.png
What would be the most elegant solution? Get the weights, bias, function and data and plot it with some other tool or does TensorFlow have support for that?
As far as I know, Tensorflow does not directly support plotting decision boundaries.
It is certainly not the most elegant solution, but you can create a grid. Classify each point of the grid and then plot it. For example:
#!/usr/bin/env python
"""
Solve the XOR problem with Tensorflow.
The XOR problem is a two-class classification problem. You only have four
datapoints, all of which are given during training time. Each datapoint has
two features:
x o
o x
As you can see, the classifier has to learn a non-linear transformation of
the features to find a propper decision boundary.
"""
__author__ = "Martin Thoma"
import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np
# The training data
XOR_X = [[0, 0], [0, 1], [1, 0], [1, 1]] # Features
XOR_Y = [[0], [1], [1], [0]] # Class labels
XOR_Y = [[1, 0], [0, 1], [0, 1], [1, 0]] # Target values
assert len(XOR_X) == len(XOR_Y) # sanity check
# The network
nb_classes = 2
input_ = tf.placeholder(tf.float32,
shape=[None, len(XOR_X[0])],
name="input")
target = tf.placeholder(tf.float32,
shape=[None, nb_classes],
name="output")
nb_hidden_nodes = 2
# enc = tf.one_hot([0, 1], 2)
w1 = tf.Variable(tf.random_uniform([2, nb_hidden_nodes], -1, 1),
name="Weights1")
w2 = tf.Variable(tf.random_uniform([nb_hidden_nodes, nb_classes], -1, 1),
name="Weights2")
b1 = tf.Variable(tf.zeros([nb_hidden_nodes]), name="Biases1")
b2 = tf.Variable(tf.zeros([nb_classes]), name="Biases2")
activation2 = tf.sigmoid(tf.matmul(input_, w1) + b1)
hypothesis = tf.nn.softmax(tf.matmul(activation2, w2) + b2)
cross_entropy = -tf.reduce_sum(target * tf.log(hypothesis))
train_step = tf.train.GradientDescentOptimizer(0.1).minimize(cross_entropy)
# Start training
init = tf.initialize_all_variables()
with tf.Session() as sess:
sess.run(init)
for i in range(100000):
sess.run(train_step, feed_dict={input_: XOR_X, target: XOR_Y})
if i % 10000 == 0:
print('Epoch ', i)
print('Hypothesis ', sess.run(hypothesis,
feed_dict={input_: XOR_X,
target: XOR_Y}))
print('w1 ', sess.run(w1))
print('b1 ', sess.run(b1))
print('w2 ', sess.run(w2))
print('b2 ', sess.run(b2))
print('cost (ce)', sess.run(cross_entropy,
feed_dict={input_: XOR_X,
target: XOR_Y}))
# Visualize classification boundary
xs = np.linspace(-5, 5)
ys = np.linspace(-5, 5)
pred_classes = []
for x in xs:
for y in ys:
pred_class = sess.run(hypothesis,
feed_dict={input_: [[x, y]]})
pred_classes.append((x, y, pred_class.argmax()))
xs_p, ys_p = [], []
xs_n, ys_n = [], []
for x, y, c in pred_classes:
if c == 0:
xs_n.append(x)
ys_n.append(y)
else:
xs_p.append(x)
ys_p.append(y)
plt.plot(xs_p, ys_p, 'ro', xs_n, ys_n, 'bo')
plt.show()
which gives