cgastaud

Results 5 comments of cgastaud

ValueError: Dimension size must be evenly divisible by 10 but is 1 for '{{node training/Reshape}} = Reshape[T=DT_FLOAT, Tshape=DT_INT32](loss/mul, training/Reshape/shape)' with input shapes: [], [2] and with input tensors computed as partial shapes: input[1] = [10,?]

Loss must have `reduction = 'none'`. See [this tutorial](https://github.com/tensorflow/privacy/blob/master/tutorials/mnist_dpsgd_tutorial_keras.py#L122) for instance

Paper or formula used in rdp_accountant.py

I think this is what you are looking for: https://arxiv.org/abs/1908.10530

IndexError in `_ecg_findpeaks_manikandan()`

we can, but after tinkering a bit with it seems like the whole for loop is not used ? we build the `id_maxes` array but then return `idx` which was...

IndexError in `_ecg_findpeaks_manikandan()`

Ok I circled back to the paper and the goal of this loop seems to be finding the "True R-peak" (see section 2.4), aka the max amplitude in the vicinity...

IndexError in `_ecg_findpeaks_manikandan()`

I think you're right about the window size, and I'm ok with leaving it as is. Maybe we could add it as a parameter ? Finding the local maximum in...