zlacker

in the long history of ML, tensorflow is not that significant. realistically, nearly everything it did was available in other systems before. It was really just a grab for Google to try to get everybody to use their framework. But it wasn't a good framework, and it evolved terribly.

MapReduce was not and isn't something that made ML possible. It made data engineering at scale possible.

>>dekhn+(OP)
> in the long history of ML, tensorflow is not that significant.

I don't necessarily disagree, but that's only from a technical pov. However, I think the release of it as a open-source library made it so that it's far easier to learn the tech that already existed. Vast majority of university courses use it to teach lots of different concepts that were far more "mathy" before.

> MapReduce was not and isn't something that made ML possible. It made data engineering at scale possible.

Yeah. Maybe I worded that wrong. I wasn't saying it made ML possible. Just that it's a huge contribution to the Open-Source tech overall.

>>erklik+P3
mapreduce wasn't a contribution to open source tech, except in the sense that it was published with enough detail for Doug Cutting to make an open source version.

Both MR and TF are net-negative for the outside world. I think more unis teach pytorch than TF now.