Linear svr #65
Linear svr #65
Conversation
There was a problem hiding this comment.
Thanks @210057zzh, it's a really good start. There are a few rough edges in the fit() implementation, I'll bring that up in our next meeting.
| end = time.time() | ||
| # print(end - start) | ||
| start = time.time() | ||
| reflsvc = svm.LinearSVC(max_iter=100000) |
There was a problem hiding this comment.
The iterations are not the same for torchml and sklearn?
| lsvc = LinearSVC(max_iter=1000) | ||
| start = time.time() | ||
| lsvc.fit(torch.from_numpy(x), torch.from_numpy(y)) | ||
| end = time.time() |
There was a problem hiding this comment.
Remove timings, or use them in the tests (eg, to make sure our implementation is fast enough).
| end = time.time() | ||
| print(end - start) | ||
| start = time.time() | ||
| reflsvr = svm.LinearSVR(max_iter=100000) |
There was a problem hiding this comment.
- Different max_iter?
- Remove or use timings.
torchml/svm/linear_svc.py
Outdated
| """ | ||
| Predict class labels for samples in X. | ||
|
|
||
| Parameters | ||
| ---------- | ||
| X : {array-like, sparse matrix} of shape (n_samples, n_features) | ||
| The data matrix for which we want to get the predictions. | ||
|
|
||
| Returns | ||
| ------- | ||
| y_pred : ndarray of shape (n_samples,) | ||
| Vector containing the class labels for each sample. | ||
| """ |
| super(LinearSVC, self).__init__() | ||
| self.coef_ = None | ||
| self.intercept_ = None | ||
| self.classes_ = None | ||
| self.dual = dual | ||
| self.tol = tol | ||
| self.C = C | ||
| self.multi_class = multi_class | ||
| self.fit_intercept = fit_intercept | ||
| self.intercept_scaling = intercept_scaling | ||
| self.class_weight = class_weight | ||
| self.verbose = verbose | ||
| self.random_state = random_state | ||
| self.max_iter = max_iter | ||
| self.penalty = penalty | ||
| self.loss = loss |
There was a problem hiding this comment.
Do we support all values for all arguments? If not, can we raise exceptions for the values we don't support?
torchml/svm/linear_svc.py
Outdated
| X_param.value = X.numpy() | ||
| y_param.value = y.numpy() |
There was a problem hiding this comment.
Why do we need .value to be a numpy array?
torchml/svm/linear_svc.py
Outdated
| C_param.value = self.C | ||
| prob.solve(solver="ECOS", abstol=self.tol, max_iters=self.max_iter) | ||
|
|
||
| self.coef_ = torch.cat((self.coef_, torch.t(torch.from_numpy(w.value)))) |
There was a problem hiding this comment.
Here too: can we just use the torch Tensor w instead of numpy? (And below for the bias too)
|
|
||
| Initialize the class with training sets | ||
|
|
||
| ## Arguments |
| import cvxpy as cp | ||
|
|
||
|
|
||
| class LinearSVC(ml.Model): |
torchml/svm/linear_svr.py
Outdated
| constraints = [] | ||
|
|
||
| prob = cp.Problem(cp.Minimize(objective), constraints) | ||
| X_param.value = X.numpy() |
There was a problem hiding this comment.
Also here: do we need the .numpy()?
Description
Fixes #33
Add LinearSVR
Contribution Checklist
If your contribution modifies code in the core library (not docs, tests, or examples), please fill the following checklist.