supporting variable dt with optimize #152
Conversation
| dxdt_hat[0] = x_hat[1] - x_hat[0] | ||
| dxdt_hat[-1] = x_hat[-1] - x_hat[-2] # use first-order endpoint formulas so as not to amplify noise. See #104 | ||
|
|
||
| x_hat = utility.integrate_dxdt_hat(dxdt_hat, dt=1) # estimate new x_hat by integrating derivative |
There was a problem hiding this comment.
changed the name of this thing, so kwarg breaks.
| t = _t | ||
|
|
||
| # The below does the approximate equivalent of this code, but sparsely in O(N sigma), since the rbf falls off rapidly | ||
| # The below does the approximate equivalent of this code, but sparsely in O(N sigma^2), since the rbf falls off rapidly |
There was a problem hiding this comment.
Solving a sparse linear system with
| :return: **x_hat** (np.array[float]) -- integral of dxdt_hat | ||
| """ | ||
| return np.hstack((0, scipy.integrate.cumulative_trapezoid(dxdt_hat)))*dt | ||
| return cumulative_trapezoid(dxdt_hat, initial=0)*_t if np.isscalar(_t) \ |
There was a problem hiding this comment.
Turns out cumulative_trapezoid can also take an initial value, in which case it returns a result the same length as the thing it's integrating, so there is no need to call hstack.
|
No idea why the optimize test I added is causing the build to take many times longer than usual (I canceled after 15 minutes one time and 6 another, and it usually takes under 2). That test takes about 30 seconds locally and passes fine. Chat thinks it could be a random number generator blocking, waiting for entropy? But the 15 minute run was from before I added seeding in the test. |
Just needed the integration utility to handle variable dt, really, and scipy's function already does, because they're awesome.