Derivative query forms cartesian product

[j042]

currently, derivatives queries take 2 forms:

# (1)
ders = spline.derivatives(q, [1, 1])

# (2)
ders = spline.derivatives(q, orders)
assert len(q) == len(orders)

I don't have any use case for (2) yet, but I often encounter the following use case

ders = []
for q_order in q_orders:
    ders.append(spline.derivatives(q, q_order))

A suggestion would be,

ders = spline.derivatives(q, q_orders)
assert len(q_orders) * len(q) == len(ders)
for i in range(len(q_orders):
    assert np.allclose(ders[i::len(q_orders)], spline.derivatives(q, q_orders[i]))

could be nice. what do you think @jzwar

[jzwar]

So a list of derivatives, all computed consecutively?

[jzwar]

For np.eye we have Jacobin already. I honestly do not really see another use-case. I don't like changing behaviour for slightly different inputs... This makes the code to unpredictable. Especially as this can become really expensive fast...

[j042]

It's not about np.eye. This makes a lot more sense than current multi-order query

[jzwar]

But it basically a cartesian product between a list of orders and queries, right? Do you want to drop the old layout?

[j042]

yes (hence, the title) and yes. I guess more direct question would be: do you use "old" layout anywhere?

[jzwar]

no, I don't. Never used it. Go for it

[j042]

@OberGue

[j042]

#337