Unable to fit model: absolute parameter change was below tolerance about prophet

[bletham]

The L-BFGS optimization algorithm uses an approximate Hessian that is calculated from the results of the past few iterations. It is then used to do a line search in the descent direction. Sometimes this approximation can become poor, and the line search fails (fails to find an improvement in the descent direction). When this happens, the approximate Hessian is reset which typically fixes the issue. In situations like the one in the first post here, we can see that after the Hessian reset, the optimization continued to progress and so there is nothing to worry about then. Generally I would not worry about this note, especially if it reports that the optimization terminated normally.

[pldeepesh]

I was wondering what does the LS Failed and Hessian Reset mean. Does that mean the model is working? when I have predicted this data the data seemed more than reasonable, it was pretty much accurate. But I am unable to understand what this means

LS Failed and Hessian Reset

[bletham]

It looks like there wasn't an error, rather this is a normal end of the optimization because it has found the best fit. If you make predictions and plot does it seem reasonable?

[dstaley]

Ah, from the examples I wasn't aware that I needed to output the plot myself. Since there was no output besides the previous, I figured it was an error.

I've updated my script, and this is how I'm now outputting the graphs.

model = Prophet().fit(df)
future = model.make_future_dataframe(periods=30)
forecast = model.predict(future)
figure = model.plot(forecast)
figure.savefig('output')