just encountered this.
collating what worked for me.
update fastai/metrics.py to this version
def accuracy_np(preds, targs):
preds = np.argmax(preds, 1)
#then in the workbook, after
log_preds, y = learn.TTA()
probs = np.mean(np.exp(log_preds),0)
print (“accuracy_np(probs, y):”, accuracy_np(probs, y))
print (“metrics.log_loss(y, probs):”, metrics.log_loss(y, probs))
many thanks to pandeyaah for putting us on track.
note to self: keep up to date with Jeremies’ git updates