GuyPozner
GuyPozner
Any updates here?
No, since the number of samples in each dataframe is different and I want df1 samples to represent 0.3 of the df distribution. We use this feature to deal with...
No, it is not possible as well, since in some cases we have something like this: ```python batch_size = 2 weights = [0.1, 0.1, 0.8] ```
No, in our current process we do something like this: ```python ds = tf.data.experimental.sample_from_datasets([ds1, ds2, ..., dsn], weights=[w1, w2, ..., wn]) ``` Thus the sampling is a random process and...
Yes to both questions. Regarding the OOM statement(I belive you are refering to #2383 ?), I think you are right :)