Complex Synthetic Feature Generation #70
Conversation
Added a suite of functions related to synthetic feature generation.
|
HTML docs are yet to be written. |
Removed .idea, cc_generator and cc_generator_tests are differet files, i've checked. |
pre-commit, code review changes: - added _feature_builder method to avoid duplicate code blocks - added some new parameters to enable random value domains for features
Sorry for not replying earlier, had implemented the necessary changes with the changelog in my commit description, should have replied to the comments as well. Thank you for all the feedback. |
renamed _feature_builder -> _configure_generate_featuer Replace np.ndarray typing with ArrayLike from numpy typing, other typing fixes
| Xn_T = X_noise.T | ||
| n = Xn_T.shape[1] | ||
| n_missing = int(n * p) | ||
| #print("n to delete:", n_missing) |
There was a problem hiding this comment.
Remoevd redundant comments
| #print("n to delete:", n_missing) | ||
|
|
||
| for feature in Xn_T: | ||
| ixs = np.random.choice(n, n_missing, replace=False) |
There was a problem hiding this comment.
Make sure seeds are set for numpy
There was a problem hiding this comment.
Seeds are now set when class is instantiated with cc = CategoricalClassification(), default is 42, pass seed parameter to change.
| ixs = np.random.choice(n, n_missing, replace=False) | ||
|
|
||
| for ix in ixs: | ||
| feature[ix] = missing_val |
There was a problem hiding this comment.
You in general do a lot of for loop-based indexing -- this is really not the optimal way, this can all be vectorized. Not in this PR thought, let's go over separately sometime
| print(f"Number of classes: {self.dataset_info['labels']['n_class']}") | ||
| print(f"Class relation: {self.dataset_info['labels']['class_relation']}") | ||
|
|
||
| print('-------------------------------------') |
There was a problem hiding this comment.
Let's leave this for now, but outrank has proper logging which should be used when time comes, please put as todo
There was a problem hiding this comment.
Removed summarize function, added TODO
| adjustment = samples_per_cluster[i] - cluster_size | ||
| adjustments.append(adjustment) | ||
|
|
||
| if adjustment < 0: # Cluter is too large |
There was a problem hiding this comment.
Comments always above lines, not after pls
There was a problem hiding this comment.
Moved all comments above code block they're referencing.
| :return: array of labels, corresponding to dataset X | ||
| """ | ||
|
|
||
| kmeans = KMeans(n_clusters=n) |
There was a problem hiding this comment.
Added random_state parameter for setting clustering seed.
|
|
||
| for feature_ix in feature_ixs: | ||
| # Filling out the dataset up to feature_ix | ||
| if ix < feature_ix: |
There was a problem hiding this comment.
Also a general comment - it's not math source, you can name variables with fuller/more idiomatic names, not this PR though
| X: ArrayLike, | ||
| feature_indices: list[int] | ArrayLike, | ||
| combination_function: Optional = None, | ||
| combination_type: Literal['linear', 'nonlinear'] = 'linear', |
There was a problem hiding this comment.
Looks like str to me - @miha-jenko why Literal?
Removed unnecessary comments, added global seed, to be set when instantiating the class, added seed for KMeans clustering
Added a suite of functions related to synthetic feature generation and corresponding tests.
Located in outrank/algorithms/synthetic_data_generators
https://jira.outbrain.com/browse/mm_generators