Open tactycHQ opened 4 years ago
Thats funny... I get a different answer like africa, and south_america ... etc every time I train. The only fact in the file on canada is that it is is a neighbor of united_states ... and united_states has neighbors canada and mexico but no info on where the US is located. .... some countries are neighbors of other countries that end up being located in south_america so I could see the system saying there is a chance Candada is in south_america ... but I never get to that even after retraining.
Still this is all very interesting and look forward to learning more and learning how to analyze/diagnose the predictions.
Really cool project and loved your podcast!
I was tinkering around with the get_most_likely function as follows:
And this results in:
[{'triple': ('canada', 'locatedin', 'central_asia'), 'prob': 0.8538}]
Unless I have my geography wrong :), do you think this is a result of the data being faulty? Or could I have done something wrong?