Abhishek Sinha

The atrous rates are (2,2,2) not (2,4,8) as the stride for the last block is 1 and so the rate gets multiplied by 1 only [here](https://github.com/tensorflow/tensorflow/blob/r1.8/tensorflow/contrib/slim/python/slim/nets/resnet_utils.py#L212). @rishizek what are your...

Hi i think you should increase the number of conv layers and increase the hidden units in fc layers too.

Hi, yeah I had solved the problem. You can check out my pull request - https://github.com/ethereon/caffe-tensorflow/pull/147

I was able to finally do it via the following code - p = featurizer.transform(train_df) data = p.rdd ##converting from dataframe to rdd data type def as_mllib(v): if isinstance(v, ml_linalg.DenseVector):...

@thunterdb So by converting the features to rdd data type and then applying svm over it am I compromising the performance in terms of speed ?

@vipzgy can you please elaborate ? I can't infer what you are saying

+1 to this. By setting average_log_prob to False, we are incentivizing longer answer/response from the model, which seems concerning to me. @eric-mitchell any suggestion on why we want to promote...