locuslab / icnn Goto Github PK

Where can I get the CSV files for bibtex data ?

the code define the f function as ：

def f_ficnn(self, x, y, reuse=False):
    fc = tflearn.fully_connected
    xy = tf.concat((x, y), 1)

    prevZ = None
    for i, sz in enumerate([200, 200, 1]):
        z_add = []

        with tf.variable_scope('z_x{}'.format(i)) as s:
            z_x = fc(xy, sz, reuse=reuse, scope=s, bias=True)
            z_add.append(z_x)

        if prevZ is not None:
            with tf.variable_scope('z_z{}_proj'.format(i)) as s:
                z_z = fc(prevZ, sz, reuse=reuse, scope=s, bias=False)
                z_add.append(z_z)

        if sz != 1:
            z = tf.nn.relu(tf.add_n(z_add))
        prevZ = z

    return tf.contrib.layers.flatten(z)

it has two input : x and y. I think there should be only one input and y is output. Do I misunderstand something?

dear teacher, I am a Chinese student. I am surprised and admire your research achievements . After reading your paper, I have benefited a lot. I want to learn more about your technology and apply it to wind turbine control. It's still in the learning phase, but the code can't run because it's true that the files are missing. ''No such file or directory: 'output.random-search/bestParams.json''

Can you update the code and give the complete code? We guarantee that we will not spread it out, that we will quote your achievements in the paper, and that our achievements will be shared with you. Thank you very much.

Thanks to your contributions!

Can you explain the for loop in the code? I could not find any pseudocode in your paper about the momentum.

tf.mul is now deprecated, and needs to be replaced with tf.multiply.
Use tf.initializers.variance_scaling instead with distribution=uniform to get equivalent behavior. Traceback (most recent call last): File "icnn.back.py", line 421, in <module> main() File "icnn.back.py", line 104, in main model = Model(inputSz, outputSz, sess, args.nGdIter) File "icnn.back.py", line 131, in __init__ E0_ = self.f(self.x_, self.y0_) File "icnn.back.py", line 330, in f z_yu = conv(tf.mul(y_red, yu_u), nFilter, kSz, strides=strides, AttributeError: module 'tensorflow' has no attribute 'mul'

Edit: Here is a list of deprecated functions and their updated variants (for tf >= 1.0 users):

• tf.merger_all_summaries => tf.summary.merge_all
• tf.histogram_summary => tf.summary.histogram
• tf.scalar_summary => tf.summary.scalar
• tf.mul => tf.multiply
• tf.train.SummaryWriter => tf.summary.FileWriter

The code in multi-label classification is not aligned with what you mentioned in your paper. The bundle entropy method is buggy, and code is not runnable either despite I revise the deprecated functions. Could you please update it ? Thanks.

I got a lot problems in this code, I think this is my gym and mujoco_py version problem can you guys tell me?

Hi,
I have trouble understanding the proof of Proposition 1 of your paper (https://arxiv.org/pdf/1609.07152.pdf). Can you provide supplementary steps why a fully connected ICNN (defined in equation (2) of your paper) is convex. Especially, why W^(y)_i can have negative values?
For example, when setting all W^(z)_i = 0 I expect the network not to be convex in general.

I would appreciate any help. Thanks.