Datasets:

EdwardoSunny
/

mle-rubrics-clean

Modalities:

Text

Formats:

Size:

Libraries:

Dataset card Data Studio Files Files and versions

xet

Community

Dataset Viewer

Auto-converted to Parquet Duplicate

Split (1)

train · 4.29k rows

rubric_id stringlengths 12 12	text stringlengths 745 1.73k	retrieval_text stringlengths 91 325	input_condition stringclasses 2 values	family stringclasses 1 value	source_instance_id stringlengths 28 68	source_repo stringclasses 3 values	source_dataset stringclasses 3 values	library stringclasses 1 value	manifest_rank int64 0 4.29k	title stringlengths 27 126	prompt_block stringlengths 745 1.73k
6ef73b6f89d9	### Persistence (last-observed-value) used as the forecast for a multi-week extrapolation horizon - Kind: methodology - Applies when: `code` -- the task asks for future cumulative/monotone quantities over a horizon that extends beyond the last training date. - Pattern: For every prediction row the script lo...	Persistence (last-observed-value) used as the forecast for a multi-week extrapolation horizon -- `code` -- the task asks for future cumulative/monotone quantities over a horizon that extends beyond the last training date.	code	mle_clean	da-pool-clean/dsbench-covid19-global-forecasting-week-3#0	dsbench	clean:methodology	mle_clean	0	Persistence (last-observed-value) used as the forecast for a multi-week extrapolation horizon	### Persistence (last-observed-value) used as the forecast for a multi-week extrapolation horizon - Kind: methodology - Applies when: `code` -- the task asks for future cumulative/monotone quantities over a horizon that extends beyond the last training date. - Pattern: For every prediction row the script lo...
eec0bc932202	### Fitted-model helpers defined but never invoked - Kind: methodology - Applies when: `code` -- the script contains fitting/prediction functions alongside a simpler fallback path. - Pattern: Functions implementing the intended model (e.g. log-linear or curve fitting) are defined at the top, but the predict...	Fitted-model helpers defined but never invoked -- `code` -- the script contains fitting/prediction functions alongside a simpler fallback path.	code	mle_clean	da-pool-clean/dsbench-covid19-global-forecasting-week-3#1	dsbench	clean:methodology	mle_clean	1	Fitted-model helpers defined but never invoked	### Fitted-model helpers defined but never invoked - Kind: methodology - Applies when: `code` -- the script contains fitting/prediction functions alongside a simpler fallback path. - Pattern: Functions implementing the intended model (e.g. log-linear or curve fitting) are defined at the top, but the predict...
a45d74e6ab6e	### No holdout evaluation against the task's own metric or a trivial baseline - Kind: methodology - Applies when: `task` -- an explicit scoring metric is given and a submission is produced. - Pattern: The script builds predictions and immediately writes the submission; nowhere does it hold back the last k d...	No holdout evaluation against the task's own metric or a trivial baseline -- `task` -- an explicit scoring metric is given and a submission is produced.	task	mle_clean	da-pool-clean/dsbench-covid19-global-forecasting-week-3#2	dsbench	clean:methodology	mle_clean	2	No holdout evaluation against the task's own metric or a trivial baseline	### No holdout evaluation against the task's own metric or a trivial baseline - Kind: methodology - Applies when: `task` -- an explicit scoring metric is given and a submission is produced. - Pattern: The script builds predictions and immediately writes the submission; nowhere does it hold back the last k d...
30cb2709443d	### Silent zero/default fill when the entity lookup finds no history - Kind: methodology - Applies when: `code` -- predictions are produced by matching a constructed entity key from the prediction set against the training set. - Pattern: Keys are built by string concatenation of possibly-missing fields (`fi...	Silent zero/default fill when the entity lookup finds no history -- `code` -- predictions are produced by matching a constructed entity key from the prediction set against the training set.	code	mle_clean	da-pool-clean/dsbench-covid19-global-forecasting-week-3#3	dsbench	clean:methodology	mle_clean	3	Silent zero/default fill when the entity lookup finds no history	### Silent zero/default fill when the entity lookup finds no history - Kind: methodology - Applies when: `code` -- predictions are produced by matching a constructed entity key from the prediction set against the training set. - Pattern: Keys are built by string concatenation of possibly-missing fields (`fi...
af6e1ab35d31	### Several scripts overwriting one submission path, with the last one left unfinished - Kind: causal - Applies when: `code` -- multiple iterations of a model live as separate files that all write to the same output filename. - Pattern: `explore/…_model/advanced_…/improved_…` scripts each end with `to_csv(<...	Several scripts overwriting one submission path, with the last one left unfinished -- `code` -- multiple iterations of a model live as separate files that all write to the same output filename.	code	mle_clean	da-pool-clean/dsbench-covid19-global-forecasting-week-3#4	dsbench	clean:causal	mle_clean	4	Several scripts overwriting one submission path, with the last one left unfinished	### Several scripts overwriting one submission path, with the last one left unfinished - Kind: causal - Applies when: `code` -- multiple iterations of a model live as separate files that all write to the same output filename. - Pattern: `explore/…_model/advanced_…/improved_…` scripts each end with `to_csv(<...
6eceb812ff33	### Per-row `DataFrame` filtering inside a Python loop over the prediction set - Kind: operational - Applies when: `code` -- predictions are generated with `for idx, row in test.iterrows()` and each iteration filters the full training table. - Pattern: Inside the loop, `train[train['key'] == row['key']].sor...	Per-row `DataFrame` filtering inside a Python loop over the prediction set -- `code` -- predictions are generated with `for idx, row in test.iterrows()` and each iteration filters the full training table.	code	mle_clean	da-pool-clean/dsbench-covid19-global-forecasting-week-3#5	dsbench	clean:operational	mle_clean	5	Per-row `DataFrame` filtering inside a Python loop over the prediction set	### Per-row `DataFrame` filtering inside a Python loop over the prediction set - Kind: operational - Applies when: `code` -- predictions are generated with `for idx, row in test.iterrows()` and each iteration filters the full training table. - Pattern: Inside the loop, `train[train['key'] == row['key']].sor...
61a3b916ca7b	### Feature selection fitted outside the cross-validation loop - Kind: methodology - Applies when: `code` -- a supervised feature selector / scaler / target-dependent transform is fit on the full training set and the reduced matrix is then passed to `cross_val_score` / `KFold` - Pattern: `X_sel = SelectKBes...	Feature selection fitted outside the cross-validation loop -- `code` -- a supervised feature selector / scaler / target-dependent transform is fit on the full training set and the reduced matrix is then passed to `cross_val_score` / `KFold`	code	mle_clean	da-pool-clean/dsbench-dont-overfit-ii#0	dsbench	clean:methodology	mle_clean	6	Feature selection fitted outside the cross-validation loop	### Feature selection fitted outside the cross-validation loop - Kind: methodology - Applies when: `code` -- a supervised feature selector / scaler / target-dependent transform is fit on the full training set and the reduced matrix is then passed to `cross_val_score` / `KFold` - Pattern: `X_sel = SelectKBes...
8d1e4fb5115c	### Submission rows not reconciled with the provided sample submission - Kind: methodology - Applies when: `code` -- the answer file is constructed from the ids of the feature/test file while a sample submission file is provided - Pattern: `pd.DataFrame({'id': test['id'], 'target': pred}).to_csv(...)` with ...	Submission rows not reconciled with the provided sample submission -- `code` -- the answer file is constructed from the ids of the feature/test file while a sample submission file is provided	code	mle_clean	da-pool-clean/dsbench-dont-overfit-ii#1	dsbench	clean:methodology	mle_clean	7	Submission rows not reconciled with the provided sample submission	### Submission rows not reconciled with the provided sample submission - Kind: methodology - Applies when: `code` -- the answer file is constructed from the ids of the feature/test file while a sample submission file is provided - Pattern: `pd.DataFrame({'id': test['id'], 'target': pred}).to_csv(...)` with ...
0a7d8dd5e248	### Hyperparameters picked by argmax of a noisy single-split CV - Kind: methodology - Applies when: `task` -- training rows are few (tens to a few hundred) relative to feature count, and the task explicitly warns against overfitting - Pattern: one `StratifiedKFold(n_splits=5, random_state=fixed)` sweep over...	Hyperparameters picked by argmax of a noisy single-split CV -- `task` -- training rows are few (tens to a few hundred) relative to feature count, and the task explicitly warns against overfitting	task	mle_clean	da-pool-clean/dsbench-dont-overfit-ii#2	dsbench	clean:methodology	mle_clean	8	Hyperparameters picked by argmax of a noisy single-split CV	### Hyperparameters picked by argmax of a noisy single-split CV - Kind: methodology - Applies when: `task` -- training rows are few (tens to a few hundred) relative to feature count, and the task explicitly warns against overfitting - Pattern: one `StratifiedKFold(n_splits=5, random_state=fixed)` sweep over...
acb66f3abf05	### Final script hard-codes a setting the exploration script was supposed to choose - Kind: methodology - Applies when: `code` -- exploration and final-fit live in separate scripts and only the final one writes the artifact - Pattern: the exploration file sweeps several axes (number of features, penalty, mo...	Final script hard-codes a setting the exploration script was supposed to choose -- `code` -- exploration and final-fit live in separate scripts and only the final one writes the artifact	code	mle_clean	da-pool-clean/dsbench-dont-overfit-ii#3	dsbench	clean:methodology	mle_clean	9	Final script hard-codes a setting the exploration script was supposed to choose	### Final script hard-codes a setting the exploration script was supposed to choose - Kind: methodology - Applies when: `code` -- exploration and final-fit live in separate scripts and only the final one writes the artifact - Pattern: the exploration file sweeps several axes (number of features, penalty, mo...
8951ad19231b	### Blanket warning suppression around iterative solvers on wide data - Kind: operational - Applies when: `code` -- `warnings.filterwarnings('ignore')` at module top and linear models fit with iterative solvers (`saga`, `liblinear`, `lbfgs`) on standardised data with features ≥ samples - Pattern: `LogisticR...	Blanket warning suppression around iterative solvers on wide data -- `code` -- `warnings.filterwarnings('ignore')` at module top and linear models fit with iterative solvers (`saga`, `liblinear`, `lbfgs`) on standardised data with features ≥ samples	code	mle_clean	da-pool-clean/dsbench-dont-overfit-ii#4	dsbench	clean:operational	mle_clean	10	Blanket warning suppression around iterative solvers on wide data	### Blanket warning suppression around iterative solvers on wide data - Kind: operational - Applies when: `code` -- `warnings.filterwarnings('ignore')` at module top and linear models fit with iterative solvers (`saga`, `liblinear`, `lbfgs`) on standardised data with features ≥ samples - Pattern: `LogisticR...
8414430af3c2	### Inline heredoc code execution when the brief mandates `.py` files - Kind: operational - Applies when: `code` -- the environment instructions state that all code must live in `.py` files invoked as `python file.py`, with no inline heredocs - Pattern: an ad-hoc verification/EDA step is run as `python3 << ...	Inline heredoc code execution when the brief mandates `.py` files -- `code` -- the environment instructions state that all code must live in `.py` files invoked as `python file.py`, with no inline heredocs	code	mle_clean	da-pool-clean/dsbench-nlp-getting-started@s2#0	dsbench	clean:operational	mle_clean	11	Inline heredoc code execution when the brief mandates `.py` files	### Inline heredoc code execution when the brief mandates `.py` files - Kind: operational - Applies when: `code` -- the environment instructions state that all code must live in `.py` files invoked as `python file.py`, with no inline heredocs - Pattern: an ad-hoc verification/EDA step is run as `python3 << ...
880438f90e25	### Row-count check against a hardcoded literal instead of the reference submission - Kind: methodology - Applies when: `code` -- a sample/reference submission file is provided and the final artifact must match it - Pattern: the validation code prints/asserts `len(submission) == <magic integer>` (or compare...	Row-count check against a hardcoded literal instead of the reference submission -- `code` -- a sample/reference submission file is provided and the final artifact must match it	code	mle_clean	da-pool-clean/dsbench-nlp-getting-started@s2#1	dsbench	clean:methodology	mle_clean	12	Row-count check against a hardcoded literal instead of the reference submission	### Row-count check against a hardcoded literal instead of the reference submission - Kind: methodology - Applies when: `code` -- a sample/reference submission file is provided and the final artifact must match it - Pattern: the validation code prints/asserts `len(submission) == <magic integer>` (or compare...
26520378b9f0	### Single random holdout used as the only validation, then discarded - Kind: methodology - Applies when: `code` -- a small tabular/text dataset (a few thousand rows) is modeled and only one configuration is finally fit - Pattern: one `train_test_split(test_size=0.2)` produces a single F1 estimate, the mode...	Single random holdout used as the only validation, then discarded -- `code` -- a small tabular/text dataset (a few thousand rows) is modeled and only one configuration is finally fit	code	mle_clean	da-pool-clean/dsbench-nlp-getting-started@s2#2	dsbench	clean:methodology	mle_clean	13	Single random holdout used as the only validation, then discarded	### Single random holdout used as the only validation, then discarded - Kind: methodology - Applies when: `code` -- a small tabular/text dataset (a few thousand rows) is modeled and only one configuration is finally fit - Pattern: one `train_test_split(test_size=0.2)` produces a single F1 estimate, the mode...
9ba979f46ae3	### No decision-threshold tuning when the metric is F1 - Kind: methodology - Applies when: `task` -- the evaluation metric is F1 (or another asymmetric metric) on a binary target - Pattern: probabilities are computed but predictions come from the default `.predict()` 0.5 cutoff, with class imbalance handled...	No decision-threshold tuning when the metric is F1 -- `task` -- the evaluation metric is F1 (or another asymmetric metric) on a binary target	task	mle_clean	da-pool-clean/dsbench-nlp-getting-started@s2#3	dsbench	clean:methodology	mle_clean	14	No decision-threshold tuning when the metric is F1	### No decision-threshold tuning when the metric is F1 - Kind: methodology - Applies when: `task` -- the evaluation metric is F1 (or another asymmetric metric) on a binary target - Pattern: probabilities are computed but predictions come from the default `.predict()` 0.5 cutoff, with class imbalance handled...
2932e0663bcd	### Over-aggressive text normalization that deletes informative tokens - Kind: methodology - Applies when: `code` -- short user-generated text (social posts, messages) is vectorized for classification - Pattern: a regex chain strips everything except letters and a couple of punctuation marks (`[^a-zA-Z\s!?]...	Over-aggressive text normalization that deletes informative tokens -- `code` -- short user-generated text (social posts, messages) is vectorized for classification	code	mle_clean	da-pool-clean/dsbench-nlp-getting-started@s2#4	dsbench	clean:methodology	mle_clean	15	Over-aggressive text normalization that deletes informative tokens	### Over-aggressive text normalization that deletes informative tokens - Kind: methodology - Applies when: `code` -- short user-generated text (social posts, messages) is vectorized for classification - Pattern: a regex chain strips everything except letters and a couple of punctuation marks (`[^a-zA-Z\s!?]...
97b963731ea5	### Target not transformed for a log-scale error metric - Kind: methodology - Applies when: `task` -- the evaluation metric is RMSLE (or any log-scale / relative error) on a skewed non-negative count target - Pattern: Regressors are fit directly on the raw target with default squared-error loss, and predict...	Target not transformed for a log-scale error metric -- `task` -- the evaluation metric is RMSLE (or any log-scale / relative error) on a skewed non-negative count target	task	mle_clean	da-pool-clean/dsbench-bike-sharing-demand@s2#0	dsbench	clean:methodology	mle_clean	16	Target not transformed for a log-scale error metric	### Target not transformed for a log-scale error metric - Kind: methodology - Applies when: `task` -- the evaluation metric is RMSLE (or any log-scale / relative error) on a skewed non-negative count target - Pattern: Regressors are fit directly on the raw target with default squared-error loss, and predict...
ac669ff73c56	### No held-out validation before submitting - Kind: methodology - Applies when: `code` -- a model is trained and its predictions written straight to the submission file - Pattern: The script fits on 100% of the labelled data and never computes any error estimate; validation utilities may even be imported (...	No held-out validation before submitting -- `code` -- a model is trained and its predictions written straight to the submission file	code	mle_clean	da-pool-clean/dsbench-bike-sharing-demand@s2#1	dsbench	clean:methodology	mle_clean	17	No held-out validation before submitting	### No held-out validation before submitting - Kind: methodology - Applies when: `code` -- a model is trained and its predictions written straight to the submission file - Pattern: The script fits on 100% of the labelled data and never computes any error estimate; validation utilities may even be imported (...
ad5f0bc5ac2c	### Features whose range in training never overlaps the test range - Kind: methodology - Applies when: `task` -- the train/test split is defined by a deterministic rule on a calendar or index field (e.g. first N days of each period train, remainder test) - Pattern: Feature engineering derives a column that ...	Features whose range in training never overlaps the test range -- `task` -- the train/test split is defined by a deterministic rule on a calendar or index field (e.g. first N days of each period train, remainder test)	task	mle_clean	da-pool-clean/dsbench-bike-sharing-demand@s2#2	dsbench	clean:methodology	mle_clean	18	Features whose range in training never overlaps the test range	### Features whose range in training never overlaps the test range - Kind: methodology - Applies when: `task` -- the train/test split is defined by a deterministic rule on a calendar or index field (e.g. first N days of each period train, remainder test) - Pattern: Feature engineering derives a column that ...
209558205275	### Truncating continuous predictions to integers - Kind: methodology - Applies when: `code` -- predictions of a continuous target are cast with `astype(int)` before writing the submission - Pattern: `astype(int)` truncates toward zero rather than rounding, biasing every prediction downward by up to 1; unde...	Truncating continuous predictions to integers -- `code` -- predictions of a continuous target are cast with `astype(int)` before writing the submission	code	mle_clean	da-pool-clean/dsbench-bike-sharing-demand@s2#3	dsbench	clean:methodology	mle_clean	19	Truncating continuous predictions to integers	### Truncating continuous predictions to integers - Kind: methodology - Applies when: `code` -- predictions of a continuous target are cast with `astype(int)` before writing the submission - Pattern: `astype(int)` truncates toward zero rather than rounding, biasing every prediction downward by up to 1; unde...
beafab6d75c3	### Ignoring the stated execution protocol (inline heredoc instead of a .py file) - Kind: methodology - Applies when: `task` -- the instructions explicitly require code to live in `.py` files run with `python file.py` and forbid inline heredocs - Pattern: Verification or exploration steps are executed as `p...	Ignoring the stated execution protocol (inline heredoc instead of a .py file) -- `task` -- the instructions explicitly require code to live in `.py` files run with `python file.py` and forbid inline heredocs	task	mle_clean	da-pool-clean/dsbench-bike-sharing-demand@s2#4	dsbench	clean:methodology	mle_clean	20	Ignoring the stated execution protocol (inline heredoc instead of a .py file)	### Ignoring the stated execution protocol (inline heredoc instead of a .py file) - Kind: methodology - Applies when: `task` -- the instructions explicitly require code to live in `.py` files run with `python file.py` and forbid inline heredocs - Pattern: Verification or exploration steps are executed as `p...
10780d3dc99d	### F1 metric optimized with class re-weighting instead of a tuned decision threshold - Kind: methodology - Applies when: `code` -- the metric is F1 on the positive class and the model is a probabilistic classifier fit with `class_weight='balanced'` (or similar resampling) and hard `.predict()`. - Pattern: ...	F1 metric optimized with class re-weighting instead of a tuned decision threshold -- `code` -- the metric is F1 on the positive class and the model is a probabilistic classifier fit with `class_weight='balanced'` (or similar resampling) and hard `.predict()`.	code	mle_clean	da-pool-clean/dsbench-nlp-getting-started#0	dsbench	clean:methodology	mle_clean	21	F1 metric optimized with class re-weighting instead of a tuned decision threshold	### F1 metric optimized with class re-weighting instead of a tuned decision threshold - Kind: methodology - Applies when: `code` -- the metric is F1 on the positive class and the model is a probabilistic classifier fit with `class_weight='balanced'` (or similar resampling) and hard `.predict()`. - Pattern: ...
76419a376590	### Single arbitrary holdout accepted as proof, no model or hyperparameter comparison - Kind: methodology - Applies when: `code` -- validation is one `train_test_split(..., random_state=k)` evaluation of exactly one pipeline configuration. - Pattern: The script builds one feature extractor + one estimator, ...	Single arbitrary holdout accepted as proof, no model or hyperparameter comparison -- `code` -- validation is one `train_test_split(..., random_state=k)` evaluation of exactly one pipeline configuration.	code	mle_clean	da-pool-clean/dsbench-nlp-getting-started#1	dsbench	clean:methodology	mle_clean	22	Single arbitrary holdout accepted as proof, no model or hyperparameter comparison	### Single arbitrary holdout accepted as proof, no model or hyperparameter comparison - Kind: methodology - Applies when: `code` -- validation is one `train_test_split(..., random_state=k)` evaluation of exactly one pipeline configuration. - Pattern: The script builds one feature extractor + one estimator, ...
882ffde4b9d4	### Aggressive text pruning on short-document classification - Kind: methodology - Applies when: `code` -- TF-IDF/count vectorization of very short documents with `stop_words='english'`, a small `max_features` cap, and `min_df`/`max_df` cutoffs all set at once. - Pattern: Several vocabulary-shrinking knobs ...	Aggressive text pruning on short-document classification -- `code` -- TF-IDF/count vectorization of very short documents with `stop_words='english'`, a small `max_features` cap, and `min_df`/`max_df` cutoffs all set at once.	code	mle_clean	da-pool-clean/dsbench-nlp-getting-started#2	dsbench	clean:methodology	mle_clean	23	Aggressive text pruning on short-document classification	### Aggressive text pruning on short-document classification - Kind: methodology - Applies when: `code` -- TF-IDF/count vectorization of very short documents with `stop_words='english'`, a small `max_features` cap, and `min_df`/`max_df` cutoffs all set at once. - Pattern: Several vocabulary-shrinking knobs ...
e8c075411c88	### Scripts delivered as inline heredocs instead of runnable files - Kind: operational - Applies when: `code` -- the environment instructs that all logic be written as `.py` files in the work directory and executed with `python file.py` - Pattern: The whole pipeline (EDA, feature engineering, training, subm...	Scripts delivered as inline heredocs instead of runnable files -- `code` -- the environment instructs that all logic be written as `.py` files in the work directory and executed with `python file.py`	code	mle_clean	da-pool-clean/dsbench-tmdb-box-office-prediction#0	dsbench	clean:operational	mle_clean	24	Scripts delivered as inline heredocs instead of runnable files	### Scripts delivered as inline heredocs instead of runnable files - Kind: operational - Applies when: `code` -- the environment instructs that all logic be written as `.py` files in the work directory and executed with `python file.py` - Pattern: The whole pipeline (EDA, feature engineering, training, subm...
3e7d6cf21ec9	### Submission row/ID set never reconciled with the provided sample or test file - Kind: methodology - Applies when: `code` -- the task supplies a `sample_submission` file and a separate test file and requires exact format matching - Pattern: Validation of the output consists only of checking column names, ...	Submission row/ID set never reconciled with the provided sample or test file -- `code` -- the task supplies a `sample_submission` file and a separate test file and requires exact format matching	code	mle_clean	da-pool-clean/dsbench-tmdb-box-office-prediction#1	dsbench	clean:methodology	mle_clean	25	Submission row/ID set never reconciled with the provided sample or test file	### Submission row/ID set never reconciled with the provided sample or test file - Kind: methodology - Applies when: `code` -- the task supplies a `sample_submission` file and a separate test file and requires exact format matching - Pattern: Validation of the output consists only of checking column names, ...
4d7c9928b22c	### Reading the labeled file and treating an internal split as the competition test set - Kind: methodology - Applies when: `code` -- the task provides distinct labeled-train and unlabeled-test files - Pattern: The script builds "train" and "test" frames whose combined size equals the labeled file's size (e...	Reading the labeled file and treating an internal split as the competition test set -- `code` -- the task provides distinct labeled-train and unlabeled-test files	code	mle_clean	da-pool-clean/dsbench-tmdb-box-office-prediction#2	dsbench	clean:methodology	mle_clean	26	Reading the labeled file and treating an internal split as the competition test set	### Reading the labeled file and treating an internal split as the competition test set - Kind: methodology - Applies when: `code` -- the task provides distinct labeled-train and unlabeled-test files - Pattern: The script builds "train" and "test" frames whose combined size equals the labeled file's size (e...
e7ac2345f6fb	### Train and test concatenated before fitting feature transforms - Kind: methodology - Applies when: `code` -- feature engineering uses aggregates, frequency counts, encodings or scalers - Pattern: `pd.concat([train.drop(target), test])` is the input to the feature builder, so any group statistic, category...	Train and test concatenated before fitting feature transforms -- `code` -- feature engineering uses aggregates, frequency counts, encodings or scalers	code	mle_clean	da-pool-clean/dsbench-tmdb-box-office-prediction#3	dsbench	clean:methodology	mle_clean	27	Train and test concatenated before fitting feature transforms	### Train and test concatenated before fitting feature transforms - Kind: methodology - Applies when: `code` -- feature engineering uses aggregates, frequency counts, encodings or scalers - Pattern: `pd.concat([train.drop(target), test])` is the input to the feature builder, so any group statistic, category...
dd41140d11b3	### No holdout score reported in the competition's own metric - Kind: methodology - Applies when: `task` -- the metric is a log-space error (e.g. RMSLE) on a heavily skewed positive target - Pattern: Models are fit and predictions written without printing a held-out score computed exactly as the metric defi...	No holdout score reported in the competition's own metric -- `task` -- the metric is a log-space error (e.g. RMSLE) on a heavily skewed positive target	task	mle_clean	da-pool-clean/dsbench-tmdb-box-office-prediction#4	dsbench	clean:methodology	mle_clean	28	No holdout score reported in the competition's own metric	### No holdout score reported in the competition's own metric - Kind: methodology - Applies when: `task` -- the metric is a log-space error (e.g. RMSLE) on a heavily skewed positive target - Pattern: Models are fit and predictions written without printing a held-out score computed exactly as the metric defi...
3aa73fdef5ad	### Final model never refit on all labelled data - Kind: methodology - Applies when: `code` -- a holdout split is carved out for validation and the chosen estimator is used directly for test predictions - Pattern: `train_test_split` produces a training subset, the pipeline is `fit` only on that subset, and ...	Final model never refit on all labelled data -- `code` -- a holdout split is carved out for validation and the chosen estimator is used directly for test predictions	code	mle_clean	da-pool-clean/dsbench-nlp-getting-started@s3#0	dsbench	clean:methodology	mle_clean	29	Final model never refit on all labelled data	### Final model never refit on all labelled data - Kind: methodology - Applies when: `code` -- a holdout split is carved out for validation and the chosen estimator is used directly for test predictions - Pattern: `train_test_split` produces a training subset, the pipeline is `fit` only on that subset, and ...
018b8c054c0b	### Model choice and reported score from one small holdout - Kind: methodology - Applies when: `code` -- two or more candidate models are compared and the better one is picked automatically - Pattern: a single `train_test_split` (one random_state) is used both to score candidates and to select `best_pipelin...	Model choice and reported score from one small holdout -- `code` -- two or more candidate models are compared and the better one is picked automatically	code	mle_clean	da-pool-clean/dsbench-nlp-getting-started@s3#1	dsbench	clean:methodology	mle_clean	30	Model choice and reported score from one small holdout	### Model choice and reported score from one small holdout - Kind: methodology - Applies when: `code` -- two or more candidate models are compared and the better one is picked automatically - Pattern: a single `train_test_split` (one random_state) is used both to score candidates and to select `best_pipelin...
d0579e9a27e6	### Tree boosting fitted on high-dimensional sparse text features inside one timed command - Kind: operational - Applies when: `code` -- a gradient-boosting / random-forest estimator is placed directly after a TF-IDF vectorizer with thousands of features - Pattern: `Pipeline([('tfidf', TfidfVectorizer(max_f...	Tree boosting fitted on high-dimensional sparse text features inside one timed command -- `code` -- a gradient-boosting / random-forest estimator is placed directly after a TF-IDF vectorizer with thousands of features	code	mle_clean	da-pool-clean/dsbench-nlp-getting-started@s3#3	dsbench	clean:operational	mle_clean	31	Tree boosting fitted on high-dimensional sparse text features inside one timed command	### Tree boosting fitted on high-dimensional sparse text features inside one timed command - Kind: operational - Applies when: `code` -- a gradient-boosting / random-forest estimator is placed directly after a TF-IDF vectorizer with thousands of features - Pattern: `Pipeline([('tfidf', TfidfVectorizer(max_f...
21bbc219f0df	### Irreversible text normalisation applied without an ablation - Kind: methodology - Applies when: `code` -- raw short-text input is heavily pre-processed before vectorisation - Pattern: a `clean_text` function strips URLs, mentions, hashes and case, combined with `stop_words='english'`, `min_df`, `max_df`...	Irreversible text normalisation applied without an ablation -- `code` -- raw short-text input is heavily pre-processed before vectorisation	code	mle_clean	da-pool-clean/dsbench-nlp-getting-started@s3#4	dsbench	clean:methodology	mle_clean	32	Irreversible text normalisation applied without an ablation	### Irreversible text normalisation applied without an ablation - Kind: methodology - Applies when: `code` -- raw short-text input is heavily pre-processed before vectorisation - Pattern: a `clean_text` function strips URLs, mentions, hashes and case, combined with `stop_words='english'`, `min_df`, `max_df`...
ec6f524dc252	### Optimizing squared error when the metric is log-scaled - Kind: methodology - Applies when: `task` states a root-mean-squared-logarithmic error (or any log/ratio metric) and `code` fits a regressor on the raw target. - Pattern: `model.fit(X, y)` with default squared-error loss on an unbounded count t...	Optimizing squared error when the metric is log-scaled -- `task` states a root-mean-squared-logarithmic error (or any log/ratio metric) and `code` fits a regressor on the raw target.	task	mle_clean	da-pool-clean/dsbench-bike-sharing-demand#1	dsbench	clean:methodology	mle_clean	33	Optimizing squared error when the metric is log-scaled	### Optimizing squared error when the metric is log-scaled - Kind: methodology - Applies when: `task` states a root-mean-squared-logarithmic error (or any log/ratio metric) and `code` fits a regressor on the raw target. - Pattern: `model.fit(X, y)` with default squared-error loss on an unbounded count t...
bc74b534ef68	### Feature whose value range is disjoint between train and test - Kind: methodology - Applies when: `task` describes a split defined by a calendar/index rule (e.g. first part of each period is train, remainder is test) and `code` derives that very quantity as a feature. - Pattern: time/index decomposition ...	Feature whose value range is disjoint between train and test -- `task` describes a split defined by a calendar/index rule (e.g. first part of each period is train, remainder is test) and `code` derives that very quantity as a feature.	task	mle_clean	da-pool-clean/dsbench-bike-sharing-demand#2	dsbench	clean:methodology	mle_clean	34	Feature whose value range is disjoint between train and test	### Feature whose value range is disjoint between train and test - Kind: methodology - Applies when: `task` describes a split defined by a calendar/index rule (e.g. first part of each period is train, remainder is test) and `code` derives that very quantity as a feature. - Pattern: time/index decomposition ...
9f6134d94759	### Ensemble weights asserted in a comment rather than measured - Kind: methodology - Applies when: `code` blends two or more model outputs with fixed numeric weights. - Pattern: `pred = am1 + bm2` with `a`, `b` justified by a comment claiming one model "typically performs better", and no validation metri...	Ensemble weights asserted in a comment rather than measured -- `code` blends two or more model outputs with fixed numeric weights.	code	mle_clean	da-pool-clean/dsbench-bike-sharing-demand#3	dsbench	clean:methodology	mle_clean	35	Ensemble weights asserted in a comment rather than measured	### Ensemble weights asserted in a comment rather than measured - Kind: methodology - Applies when: `code` blends two or more model outputs with fixed numeric weights. - Pattern: `pred = am1 + bm2` with `a`, `b` justified by a comment claiming one model "typically performs better", and no validation metri...
71245b6f492c	### Quantizing predictions for a continuous metric - Kind: methodology - Applies when: `task` scores continuous predictions and `code` casts them to integers or rounds before writing the submission. - Pattern: `predictions.astype(int)` (truncation) applied to a non-negative regression output whose true valu...	Quantizing predictions for a continuous metric -- `task` scores continuous predictions and `code` casts them to integers or rounds before writing the submission.	task	mle_clean	da-pool-clean/dsbench-bike-sharing-demand#4	dsbench	clean:methodology	mle_clean	36	Quantizing predictions for a continuous metric	### Quantizing predictions for a continuous metric - Kind: methodology - Applies when: `task` scores continuous predictions and `code` casts them to integers or rounds before writing the submission. - Pattern: `predictions.astype(int)` (truncation) applied to a non-negative regression output whose true valu...
da40c0c6d24d	### Verification run through an inline heredoc instead of a script file - Kind: operational - Applies when: `task`/instructions require code to live in `.py` files executed as `python file.py`, and the agent pipes code into the interpreter via `python << 'EOF'`. - Pattern: analysis or sanity-check logic exi...	Verification run through an inline heredoc instead of a script file -- `task`/instructions require code to live in `.py` files executed as `python file.py`, and the agent pipes code into the interpreter via `python << 'EOF'`.	task	mle_clean	da-pool-clean/dsbench-bike-sharing-demand#5	dsbench	clean:operational	mle_clean	37	Verification run through an inline heredoc instead of a script file	### Verification run through an inline heredoc instead of a script file - Kind: operational - Applies when: `task`/instructions require code to live in `.py` files executed as `python file.py`, and the agent pipes code into the interpreter via `python << 'EOF'`. - Pattern: analysis or sanity-check logic exi...
a0635a60bc02	### Metric computed on already-transformed targets - Kind: methodology - Applies when: `code` -- the target is log-transformed before fitting and validation uses a log-based error function - Pattern: `y_log = np.log1p(y)` is used for training/validation, and the held-out score is then obtained by feeding th...	Metric computed on already-transformed targets -- `code` -- the target is log-transformed before fitting and validation uses a log-based error function	code	mle_clean	da-pool-clean/dsbench-santander-value-prediction-challenge@s2#0	dsbench	clean:methodology	mle_clean	38	Metric computed on already-transformed targets	### Metric computed on already-transformed targets - Kind: methodology - Applies when: `code` -- the target is log-transformed before fitting and validation uses a log-based error function - Pattern: `y_log = np.log1p(y)` is used for training/validation, and the held-out score is then obtained by feeding th...
2716dd41f605	### No baseline comparison before shipping predictions - Kind: methodology - Applies when: `code` -- a single model's predictions are written to the submission without any reference point - Pattern: The script trains one model, prints a validation number, and writes predictions; it never evaluates the trivi...	No baseline comparison before shipping predictions -- `code` -- a single model's predictions are written to the submission without any reference point	code	mle_clean	da-pool-clean/dsbench-santander-value-prediction-challenge@s2#1	dsbench	clean:methodology	mle_clean	39	No baseline comparison before shipping predictions	### No baseline comparison before shipping predictions - Kind: methodology - Applies when: `code` -- a single model's predictions are written to the submission without any reference point - Pattern: The script trains one model, prints a validation number, and writes predictions; it never evaluates the trivi...
ff64c900d177	### High-dimensional, small-sample fit with no regularization or feature reduction - Kind: methodology - Applies when: `code` -- the training table has far more feature columns than rows (wide, mostly-sparse anonymized features) - Pattern: All columns are passed as-is to a default-parameter tree ensemble (e...	High-dimensional, small-sample fit with no regularization or feature reduction -- `code` -- the training table has far more feature columns than rows (wide, mostly-sparse anonymized features)	code	mle_clean	da-pool-clean/dsbench-santander-value-prediction-challenge@s2#2	dsbench	clean:methodology	mle_clean	40	High-dimensional, small-sample fit with no regularization or feature reduction	### High-dimensional, small-sample fit with no regularization or feature reduction - Kind: methodology - Applies when: `code` -- the training table has far more feature columns than rows (wide, mostly-sparse anonymized features) - Pattern: All columns are passed as-is to a default-parameter tree ensemble (e...
0d25f93cdc7d	### Final model refit on all data with no retained holdout check - Kind: methodology - Applies when: `code` -- a separate script/section retrains on the full training set for the final predictions - Pattern: Validation happens in one run; the final run refits on 100% of rows with the same (untuned) configur...	Final model refit on all data with no retained holdout check -- `code` -- a separate script/section retrains on the full training set for the final predictions	code	mle_clean	da-pool-clean/dsbench-santander-value-prediction-challenge@s2#3	dsbench	clean:methodology	mle_clean	41	Final model refit on all data with no retained holdout check	### Final model refit on all data with no retained holdout check - Kind: methodology - Applies when: `code` -- a separate script/section retrains on the full training set for the final predictions - Pattern: Validation happens in one run; the final run refits on 100% of rows with the same (untuned) configur...
a3cef8beff38	### Analysis run through inline heredocs instead of the required .py files - Kind: operational - Applies when: `task` -- the instructions state that code must be written as `.py` files in the work directory and executed with `python file.py` - Pattern: Every modeling step is executed as `python << 'EOF' ......	Analysis run through inline heredocs instead of the required .py files -- `task` -- the instructions state that code must be written as `.py` files in the work directory and executed with `python file.py`	task	mle_clean	da-pool-clean/dsbench-santander-value-prediction-challenge@s2#4	dsbench	clean:operational	mle_clean	42	Analysis run through inline heredocs instead of the required .py files	### Analysis run through inline heredocs instead of the required .py files - Kind: operational - Applies when: `task` -- the instructions state that code must be written as `.py` files in the work directory and executed with `python file.py` - Pattern: Every modeling step is executed as `python << 'EOF' ......
23732f0c234a	### Final artifact chosen by execution order, not by validation score - Kind: methodology - Applies when: `code` -- two or more scripts (v1, v2, …) each write the same required output path. - Pattern: Successive iterations of the solution each end with `to_csv('.../submission.csv')`, and no step records/com...	Final artifact chosen by execution order, not by validation score -- `code` -- two or more scripts (v1, v2, …) each write the same required output path.	code	mle_clean	da-pool-clean/dsbench-dont-overfit-ii@s2#0	dsbench	clean:methodology	mle_clean	43	Final artifact chosen by execution order, not by validation score	### Final artifact chosen by execution order, not by validation score - Kind: methodology - Applies when: `code` -- two or more scripts (v1, v2, …) each write the same required output path. - Pattern: Successive iterations of the solution each end with `to_csv('.../submission.csv')`, and no step records/com...
b5348f7db0e4	### Model selection on the same folds used to score them - Kind: methodology - Applies when: `code` -- many candidate models/hyperparameters are ranked by one `cross_val_score` pass on a small training set and the top-k are then blended. - Pattern: `for name, model in models: scores = cross_val_score(model,...	Model selection on the same folds used to score them -- `code` -- many candidate models/hyperparameters are ranked by one `cross_val_score` pass on a small training set and the top-k are then blended.	code	mle_clean	da-pool-clean/dsbench-dont-overfit-ii@s2#1	dsbench	clean:methodology	mle_clean	44	Model selection on the same folds used to score them	### Model selection on the same folds used to score them - Kind: methodology - Applies when: `code` -- many candidate models/hyperparameters are ranked by one `cross_val_score` pass on a small training set and the top-k are then blended. - Pattern: `for name, model in models: scores = cross_val_score(model,...
f5d9332cdae1	### Averaging raw probabilities from heterogeneous, uncalibrated models for a rank metric - Kind: methodology - Applies when: `task` scores with a rank-based metric (AUC) and `code` blends models of different families (linear, tree, kernel). - Pattern: `np.mean(predictions, axis=0)` over `predict_proba` out...	Averaging raw probabilities from heterogeneous, uncalibrated models for a rank metric -- `task` scores with a rank-based metric (AUC) and `code` blends models of different families (linear, tree, kernel).	task	mle_clean	da-pool-clean/dsbench-dont-overfit-ii@s2#2	dsbench	clean:methodology	mle_clean	45	Averaging raw probabilities from heterogeneous, uncalibrated models for a rank metric	### Averaging raw probabilities from heterogeneous, uncalibrated models for a rank metric - Kind: methodology - Applies when: `task` scores with a rank-based metric (AUC) and `code` blends models of different families (linear, tree, kernel). - Pattern: `np.mean(predictions, axis=0)` over `predict_proba` out...
bc02b22fb9de	### Submission rows taken from an assumed id range instead of the provided test file - Kind: methodology - Applies when: `task` states a test-set size or supplies a sample submission whose row count disagrees with the actual test file shipped in the data directory. - Pattern: Building the output frame from ...	Submission rows taken from an assumed id range instead of the provided test file -- `task` states a test-set size or supplies a sample submission whose row count disagrees with the actual test file shipped in the data directory.	task	mle_clean	da-pool-clean/dsbench-dont-overfit-ii@s2#3	dsbench	clean:methodology	mle_clean	46	Submission rows taken from an assumed id range instead of the provided test file	### Submission rows taken from an assumed id range instead of the provided test file - Kind: methodology - Applies when: `task` states a test-set size or supplies a sample submission whose row count disagrees with the actual test file shipped in the data directory. - Pattern: Building the output frame from ...
2589f2881fb8	### Inline heredoc Python used for the verification steps - Kind: operational - Applies when: `task` instructions require all code to live in `.py` files executed with `python file.py`. - Pattern: Validation/inspection logic run as `python3 << 'EOF' ... EOF` in the shell, so the checks are not part of any s...	Inline heredoc Python used for the verification steps -- `task` instructions require all code to live in `.py` files executed with `python file.py`.	task	mle_clean	da-pool-clean/dsbench-dont-overfit-ii@s2#4	dsbench	clean:operational	mle_clean	47	Inline heredoc Python used for the verification steps	### Inline heredoc Python used for the verification steps - Kind: operational - Applies when: `task` instructions require all code to live in `.py` files executed with `python file.py`. - Pattern: Validation/inspection logic run as `python3 << 'EOF' ... EOF` in the shell, so the checks are not part of any s...
4f161eedcbd4	### Single-threaded sklearn GradientBoosting on a wide, tens-of-thousands-row table inside one time-capped command - Kind: operational - Applies when: `code` -- a tabular model is trained with `sklearn.ensemble.GradientBoostingClassifier`/`Regressor` on a dataset with tens of thousands of rows and hundreds of c...	Single-threaded sklearn GradientBoosting on a wide, tens-of-thousands-row table inside one time-capped command -- `code` -- a tabular model is trained with `sklearn.ensemble.GradientBoostingClassifier`/`Regressor` on a dataset with tens of thousands of rows and hundreds of columns under a hard per-command timeout.	code	mle_clean	da-pool-clean/dsbench-santander-customer-satisfaction#0	dsbench	clean:operational	mle_clean	48	Single-threaded sklearn GradientBoosting on a wide, tens-of-thousands-row table inside one time-capped command	### Single-threaded sklearn GradientBoosting on a wide, tens-of-thousands-row table inside one time-capped command - Kind: operational - Applies when: `code` -- a tabular model is trained with `sklearn.ensemble.GradientBoostingClassifier`/`Regressor` on a dataset with tens of thousands of rows and hundreds of c...
ad6cdbf77d28	### Imputation statistics recomputed from an already-imputed frame - Kind: methodology - Applies when: `code` -- missing values are filled with column statistics and the same statistic expression is reused for the test frame. - Pattern: `X = X.fillna(X.median())` followed by `X_test = X_test.fillna(X.median...	Imputation statistics recomputed from an already-imputed frame -- `code` -- missing values are filled with column statistics and the same statistic expression is reused for the test frame.	code	mle_clean	da-pool-clean/dsbench-santander-customer-satisfaction#1	dsbench	clean:methodology	mle_clean	49	Imputation statistics recomputed from an already-imputed frame	### Imputation statistics recomputed from an already-imputed frame - Kind: methodology - Applies when: `code` -- missing values are filled with column statistics and the same statistic expression is reused for the test frame. - Pattern: `X = X.fillna(X.median())` followed by `X_test = X_test.fillna(X.median...
eda646db38a7	### Preprocessing statistics fitted on the whole training set before the validation split - Kind: methodology - Applies when: `code` -- a held-out split is used to estimate the metric and any imputation/scaling/target-derived statistic is computed earlier in the script. - Pattern: Column medians/means/scale...	Preprocessing statistics fitted on the whole training set before the validation split -- `code` -- a held-out split is used to estimate the metric and any imputation/scaling/target-derived statistic is computed earlier in the script.	code	mle_clean	da-pool-clean/dsbench-santander-customer-satisfaction#2	dsbench	clean:methodology	mle_clean	50	Preprocessing statistics fitted on the whole training set before the validation split	### Preprocessing statistics fitted on the whole training set before the validation split - Kind: methodology - Applies when: `code` -- a held-out split is used to estimate the metric and any imputation/scaling/target-derived statistic is computed earlier in the script. - Pattern: Column medians/means/scale...
4123073b16ad	### Submission built from test-file row order without aligning to the sample submission - Kind: methodology - Applies when: `task` -- a sample submission file is provided and the instruction is to match it exactly. - Pattern: The output frame is assembled as `DataFrame({'ID': test_ids, 'TARGET': preds})` st...	Submission built from test-file row order without aligning to the sample submission -- `task` -- a sample submission file is provided and the instruction is to match it exactly.	task	mle_clean	da-pool-clean/dsbench-santander-customer-satisfaction#3	dsbench	clean:methodology	mle_clean	51	Submission built from test-file row order without aligning to the sample submission	### Submission built from test-file row order without aligning to the sample submission - Kind: methodology - Applies when: `task` -- a sample submission file is provided and the instruction is to match it exactly. - Pattern: The output frame is assembled as `DataFrame({'ID': test_ids, 'TARGET': preds})` st...
d70bc96d7f82	### RMSLE computed with raw `log()` on unclipped model output - Kind: operational - Applies when: `code` -- validation code evaluates a log-based error metric on predictions from a model family that can emit negatives (linear/ridge, boosting on raw targets). - Pattern: `np.log(pred + 1)` / `mean_squared_log...	RMSLE computed with raw `log()` on unclipped model output -- `code` -- validation code evaluates a log-based error metric on predictions from a model family that can emit negatives (linear/ridge, boosting on raw targets).	code	mle_clean	da-pool-clean/dsbench-bike-sharing-demand@s3#0	dsbench	clean:operational	mle_clean	52	RMSLE computed with raw `log()` on unclipped model output	### RMSLE computed with raw `log()` on unclipped model output - Kind: operational - Applies when: `code` -- validation code evaluates a log-based error metric on predictions from a model family that can emit negatives (linear/ridge, boosting on raw targets). - Pattern: `np.log(pred + 1)` / `mean_squared_log...
00e99e96cb58	### Fitting on the raw target when the metric is logarithmic - Kind: methodology - Applies when: `task` -- evaluation is RMSLE (or any log-scale error) on a skewed count/positive target. - Pattern: Regressors are fit directly on the raw target with squared-error loss and predictions are written as-is; no `l...	Fitting on the raw target when the metric is logarithmic -- `task` -- evaluation is RMSLE (or any log-scale error) on a skewed count/positive target.	task	mle_clean	da-pool-clean/dsbench-bike-sharing-demand@s3#1	dsbench	clean:methodology	mle_clean	53	Fitting on the raw target when the metric is logarithmic	### Fitting on the raw target when the metric is logarithmic - Kind: methodology - Applies when: `task` -- evaluation is RMSLE (or any log-scale error) on a skewed count/positive target. - Pattern: Regressors are fit directly on the raw target with squared-error loss and predictions are written as-is; no `l...
f398a9314d32	### Shipping a blend the validation said was worse - Kind: methodology - Applies when: `code` -- a held-out evaluation compares individual models and an ensemble, and a separate final script produces the submission. - Pattern: Validation ranks a single model best, but the submission script hard-codes an arb...	Shipping a blend the validation said was worse -- `code` -- a held-out evaluation compares individual models and an ensemble, and a separate final script produces the submission.	code	mle_clean	da-pool-clean/dsbench-bike-sharing-demand@s3#2	dsbench	clean:methodology	mle_clean	54	Shipping a blend the validation said was worse	### Shipping a blend the validation said was worse - Kind: methodology - Applies when: `code` -- a held-out evaluation compares individual models and an ensemble, and a separate final script produces the submission. - Pattern: Validation ranks a single model best, but the submission script hard-codes an arb...
1661b0925b09	### Hyperparameters changed between validation and the final fit - Kind: methodology - Applies when: `code` -- a final/production script is a copy of an earlier validated script with edited settings. - Pattern: Tree depth, estimator counts, regularization strength or blend weights differ from the configurat...	Hyperparameters changed between validation and the final fit -- `code` -- a final/production script is a copy of an earlier validated script with edited settings.	code	mle_clean	da-pool-clean/dsbench-bike-sharing-demand@s3#3	dsbench	clean:methodology	mle_clean	55	Hyperparameters changed between validation and the final fit	### Hyperparameters changed between validation and the final fit - Kind: methodology - Applies when: `code` -- a final/production script is a copy of an earlier validated script with edited settings. - Pattern: Tree depth, estimator counts, regularization strength or blend weights differ from the configurat...
a67dcf8288b3	### Random holdout split when train/test are split by a calendar rule - Kind: methodology - Applies when: `task` -- the task states the train/test partition follows a temporal or calendar rule (e.g. early part of each period trains, later part tests). - Pattern: Validation uses `train_test_split`/random shu...	Random holdout split when train/test are split by a calendar rule -- `task` -- the task states the train/test partition follows a temporal or calendar rule (e.g. early part of each period trains, later part tests).	task	mle_clean	da-pool-clean/dsbench-bike-sharing-demand@s3#4	dsbench	clean:methodology	mle_clean	56	Random holdout split when train/test are split by a calendar rule	### Random holdout split when train/test are split by a calendar rule - Kind: methodology - Applies when: `task` -- the task states the train/test partition follows a temporal or calendar rule (e.g. early part of each period trains, later part tests). - Pattern: Validation uses `train_test_split`/random shu...
7a660cc0e0c6	### Feature whose train support is disjoint from test support - Kind: methodology - Applies when: `code` -- derived calendar/index features are generated from a timestamp while the task says train and test occupy disjoint ranges of that component. - Pattern: A feature such as day-of-period is added to the f...	Feature whose train support is disjoint from test support -- `code` -- derived calendar/index features are generated from a timestamp while the task says train and test occupy disjoint ranges of that component.	code	mle_clean	da-pool-clean/dsbench-bike-sharing-demand@s3#5	dsbench	clean:methodology	mle_clean	57	Feature whose train support is disjoint from test support	### Feature whose train support is disjoint from test support - Kind: methodology - Applies when: `code` -- derived calendar/index features are generated from a timestamp while the task says train and test occupy disjoint ranges of that component. - Pattern: A feature such as day-of-period is added to the f...
d4795d9b6ef1	### Date-difference converted to integers via `.days` / `.dt` on the wrong object type - Kind: operational - Applies when: `code` -- the script turns a column of dates into an elapsed-day feature or forecast horizon - Pattern: Subtracting two date columns (or a date column and an anchor date) and then calli...	Date-difference converted to integers via `.days` / `.dt` on the wrong object type -- `code` -- the script turns a column of dates into an elapsed-day feature or forecast horizon	code	mle_clean	da-pool-clean/dsbench-covid19-global-forecasting-week-2@s2#0	dsbench	clean:operational	mle_clean	58	Date-difference converted to integers via `.days` / `.dt` on the wrong object type	### Date-difference converted to integers via `.days` / `.dt` on the wrong object type - Kind: operational - Applies when: `code` -- the script turns a column of dates into an elapsed-day feature or forecast horizon - Pattern: Subtracting two date columns (or a date column and an anchor date) and then calli...
9e76279108fe	### Mixing datetime64 and plain numbers in one comparison or arithmetic expression - Kind: operational - Applies when: `code` -- the script filters, masks or thresholds rows using a date column - Pattern: Comparing or combining a `datetime64` column with an integer/float literal (e.g. `df['date'] > cutoff_i...	Mixing datetime64 and plain numbers in one comparison or arithmetic expression -- `code` -- the script filters, masks or thresholds rows using a date column	code	mle_clean	da-pool-clean/dsbench-covid19-global-forecasting-week-2@s2#1	dsbench	clean:operational	mle_clean	59	Mixing datetime64 and plain numbers in one comparison or arithmetic expression	### Mixing datetime64 and plain numbers in one comparison or arithmetic expression - Kind: operational - Applies when: `code` -- the script filters, masks or thresholds rows using a date column - Pattern: Comparing or combining a `datetime64` column with an integer/float literal (e.g. `df['date'] > cutoff_i...
03617d0a5304	### Test rows that fall inside the training window are predicted instead of looked up - Kind: methodology - Applies when: `task` -- the evaluation file overlaps the historical period covered by the training file for the same entities - Pattern: The script builds one forecast per entity (last value, or a fit...	Test rows that fall inside the training window are predicted instead of looked up -- `task` -- the evaluation file overlaps the historical period covered by the training file for the same entities	task	mle_clean	da-pool-clean/dsbench-covid19-global-forecasting-week-2@s2#2	dsbench	clean:methodology	mle_clean	60	Test rows that fall inside the training window are predicted instead of looked up	### Test rows that fall inside the training window are predicted instead of looked up - Kind: methodology - Applies when: `task` -- the evaluation file overlaps the historical period covered by the training file for the same entities - Pattern: The script builds one forecast per entity (last value, or a fit...
215b70383608	### No holdout backtest and no comparison against a trivial baseline - Kind: methodology - Applies when: `code` -- a forecasting/prediction script produces a submission file from a model with no reported error metric - Pattern: The only validation performed is structural (row count, no NaN, non-negative, ID...	No holdout backtest and no comparison against a trivial baseline -- `code` -- a forecasting/prediction script produces a submission file from a model with no reported error metric	code	mle_clean	da-pool-clean/dsbench-covid19-global-forecasting-week-2@s2#3	dsbench	clean:methodology	mle_clean	61	No holdout backtest and no comparison against a trivial baseline	### No holdout backtest and no comparison against a trivial baseline - Kind: methodology - Applies when: `code` -- a forecasting/prediction script produces a submission file from a model with no reported error metric - Pattern: The only validation performed is structural (row count, no NaN, non-negative, ID...
ca73e995c21f	### Unbounded extrapolation of an exponential/growth fit with no sanity cap - Kind: methodology - Applies when: `code` -- per-entity growth rates are estimated from a short recent window and projected over many future steps - Pattern: Fitting a log-linear or exponential trend on a handful of recent points a...	Unbounded extrapolation of an exponential/growth fit with no sanity cap -- `code` -- per-entity growth rates are estimated from a short recent window and projected over many future steps	code	mle_clean	da-pool-clean/dsbench-covid19-global-forecasting-week-2@s2#4	dsbench	clean:methodology	mle_clean	62	Unbounded extrapolation of an exponential/growth fit with no sanity cap	### Unbounded extrapolation of an exponential/growth fit with no sanity cap - Kind: methodology - Applies when: `code` -- per-entity growth rates are estimated from a short recent window and projected over many future steps - Pattern: Fitting a log-linear or exponential trend on a handful of recent points a...
e491a3b7417f	### Repeated crashes in date handling collapse the method into an unvalidated constant fill - Kind: causal - Applies when: `code` -- successive rewrites of the same modelling block progressively delete features after runtime errors - Pattern: After several aborted attempts caused by datetime/timedelta type ...	Repeated crashes in date handling collapse the method into an unvalidated constant fill -- `code` -- successive rewrites of the same modelling block progressively delete features after runtime errors	code	mle_clean	da-pool-clean/dsbench-covid19-global-forecasting-week-2@s2#5	dsbench	clean:causal	mle_clean	63	Repeated crashes in date handling collapse the method into an unvalidated constant fill	### Repeated crashes in date handling collapse the method into an unvalidated constant fill - Kind: causal - Applies when: `code` -- successive rewrites of the same modelling block progressively delete features after runtime errors - Pattern: After several aborted attempts caused by datetime/timedelta type ...
67ef2401848d	### Feature selection fit on all labeled rows before cross-validation - Kind: methodology - Applies when: `code` -- a univariate/model-based selector or scaler is fit once on the full training matrix and the reduced matrix is then passed to `cross_val_score`/`cross_val_predict`. - Pattern: `sel = SelectKBes...	Feature selection fit on all labeled rows before cross-validation -- `code` -- a univariate/model-based selector or scaler is fit once on the full training matrix and the reduced matrix is then passed to `cross_val_score`/`cross_val_predict`.	code	mle_clean	da-pool-clean/dsbench-dont-overfit-ii@s3#0	dsbench	clean:methodology	mle_clean	64	Feature selection fit on all labeled rows before cross-validation	### Feature selection fit on all labeled rows before cross-validation - Kind: methodology - Applies when: `code` -- a univariate/model-based selector or scaler is fit once on the full training matrix and the reduced matrix is then passed to `cross_val_score`/`cross_val_predict`. - Pattern: `sel = SelectKBes...

End of preview. Expand in Data Studio

README.md exists but content is empty.

Downloads last month: 24

Collection including EdwardoSunny/mle-rubrics-clean

Coding world model: rubric libraries

Collection

Rubric libraries for a coding world model, by curation method; see https://codingwm.github.io/rubric_distribution.html • 12 items • Updated 26 days ago