Stepping with discrete numerical data

I have this data: nyears_uniqueness.csv.txt

When I use pyGAM to smooth it, I get some strange results:

gam = LinearGAM(n_splines=4)
gam.fit(df[['nyears']], df.uniqueness)
# /home/naught101/miniconda3/envs/science/lib/python3.6/site-packages/pygam/pygam.py:1172: UserWarning: detected catergorical data for feature 0
#  self._validate_data_dep_params(X)

x_pred = np.linspace(min(df.nyears), max(df.nyears), num=100)
y_pred = gam.predict(x_pred)    

df.plot.scatter(x='nyears', y='uniqueness')
ax = gca()
ax.plot(x_pred, y_pred)    

When I try plotting the same data in R, I get a much smoother curve. I’m not sure if that’s just a matter of the defaults being more appropriate for the data, or if this is a bug in pyGAM. If, do you have any suggestions for how to improve the pyGAM plot?

[dswah]

solved! by default we assume all data is numerical. thanks @naught101

[naught101]

I don’t really use R much at the moment, and haven’t used R’s GAM other than as a comparison with pyGAM, to see if I was seeing expected behaviour.

I would think that the best action would be just to assume any numerical data is numerical, unless dtype='categorical' is supplied. Is there any reason to assume categorical for such data, regardless of how discrete the values are?