tre
tre copied to clipboard
Measuring compositionality in representation learning
eg, lets say I have `[red, green, pink]`. I came up with two ideas, not sure which of these are best, or perhaps they are both best not the best...
for some versions of scipy, need a guard around `sig = scipy.stats.pearsonr([d[2] for d in datapoints], [d[3] for d in datapoints])` ... otherwise scipy throws an exception about need at...
I believe there is a bug at "eval2.py" file. RMSprop run (lines 50-60) is done updating the embedding weights after one iteration, and the converged results is just +- the...