I was working on this dataset, but there are some variables that I don't fully understand. I think It will be helpful to have a minimum knowledge of what is each variable to understand more the problem and design better models.
\nI know there are a lot of variables, but all are combinations of concepts in some numeric levels. For example, we have a lot of variables with this format: occurs_within_A_{number}_B_{number}_{number}, where A and B can be any feature of peaks, zone, cross_threshold_from_{above/below}, etc. Does anyone understand, or have some knowledge on features for detecting reversal points, and what are these variables telling us? For instance, I don't have any clue on what does the third number mean (in the example I gave) and the difference between occurs or happens.
I hope this question helps other participants and that we can understand all these variables.
\nGood luck with the competition!
I was working on this dataset, but there are some variables that I don't fully understand. I think It will be helpful to have a minimum knowledge of what is each variable to understand more the problem and design better models.
\nI know there are a lot of variables, but all are combinations of concepts in some numeric levels. For example, we have a lot of variables with this format: occurs_within_A_{number}_B_{number}_{number}, where A and B can be any feature of peaks, zone, cross_threshold_from_{above/below}, etc. Does anyone understand, or have some knowledge on features for detecting reversal points, and what are these variables telling us? For instance, I don't have any clue on what does the third number mean (in the example I gave) and the difference between occurs or happens.
I hope this question helps other participants and that we can understand all these variables.
\nGood luck with the competition!
Yes basically those are derived featured from base descriptors , we have initally ratio, momentum, sm_ratio and sm_momentum values as base descriptors, then from these we generate troughs , peaks, zones etc based on thresholds. Occurs with brings sequantial information too, happens has no order.
\n - occurs_within vs happens_within: These appear to be synonymous - both check if an event occurs within a time window after another event
Event Types:
\n - cross_threshold_from_above_{value}: Price crosses below a threshold from above
\n - cross_threshold_from_below_{value}: Price crosses above a threshold from below
\n - peaks_{value}: Local price peaks at specific levels
\n - zones_{value}: Price enters specific zones/ranges
Threshold Values:
\n - 100.0, 100.5, 101.0, etc.: These are normalized price levels (likely as percentages of some reference price)
The Third Number (Time Window):
\n - _1, _2, _3, …, _10: This represents the number of periods/bars to look ahead
\n - It answers: \"Does event B happen within X periods after event A?\"
Example Interpretation:
\nhappens_within_cross_threshold_from_above_100.0_peaks_100.5_3
\nTranslation: \"Does a peak at 100.5 level occur within 3 periods after the price crosses below the 100.0 threshold from above?\"
\nThis is measuring sequential event patterns that are characteristic of reversal points:
\nFirst event: Price drops below 100.0 (potential support break)
Second event: Price makes a peak at 100.5 within 3 bars (failed breakdown, potential reversal)
\nReversal Detection Logic:
\nThese features capture common reversal patterns:
Failed breakdowns: Price breaks support but quickly reverses
Failed breakouts: Price breaks resistance but fails to sustain
Double bottoms/tops: Sequential peaks/troughs at similar levels
Support/resistance retests: Price returns to test previous levels
\nThe time windows (1-10 periods) capture both immediate reversals and slightly delayed ones, which is crucial since reversal patterns can unfold over different timeframes.
\n\noccurs_within vs happens_within: These appear to be synonymous
\n
Then why must you add both of them to dataset as features? What is the diffirence? Or if they were exactly the same?
", "rawMarkdown": "ozzymet \n>occurs_within vs happens_within: These appear to be synonymous\n \nThen why must you add both of them to dataset as features? What is the diffirence? Or if they were exactly the same?", "votes": null }, { "id": "3379813", "postDate": "12/20/2025 20:18:38", "content": "Exactly correct. The columns are events of a variable, in a discretized range. It is my chore to build a transformer to recover the real variables and values.
", "rawMarkdown": "Exactly correct. The columns are events of a variable, in a discretized range. It is my chore to build a transformer to recover the real variables and values.", "votes": null } ], "comments": [ { "id": 3292795, "author_name": "ozzymet", "author_url": "", "post_date": "09/22/2025 15:08:05", "content": "Yes basically those are derived featured from base descriptors , we have initally ratio, momentum, sm_ratio and sm_momentum values as base descriptors, then from these we generate troughs , peaks, zones etc based on thresholds. Occurs with brings sequantial information too, happens has no order.
\n - occurs_within vs happens_within: These appear to be synonymous - both check if an event occurs within a time window after another event
Event Types:
\n - cross_threshold_from_above_{value}: Price crosses below a threshold from above
\n - cross_threshold_from_below_{value}: Price crosses above a threshold from below
\n - peaks_{value}: Local price peaks at specific levels
\n - zones_{value}: Price enters specific zones/ranges
Threshold Values:
\n - 100.0, 100.5, 101.0, etc.: These are normalized price levels (likely as percentages of some reference price)
The Third Number (Time Window):
\n - _1, _2, _3, …, _10: This represents the number of periods/bars to look ahead
\n - It answers: \"Does event B happen within X periods after event A?\"
Example Interpretation:
\nhappens_within_cross_threshold_from_above_100.0_peaks_100.5_3
\nTranslation: \"Does a peak at 100.5 level occur within 3 periods after the price crosses below the 100.0 threshold from above?\"
\nThis is measuring sequential event patterns that are characteristic of reversal points:
\nFirst event: Price drops below 100.0 (potential support break)
Second event: Price makes a peak at 100.5 within 3 bars (failed breakdown, potential reversal)
\nReversal Detection Logic:
\nThese features capture common reversal patterns:
Failed breakdowns: Price breaks support but quickly reverses
Failed breakouts: Price breaks resistance but fails to sustain
Double bottoms/tops: Sequential peaks/troughs at similar levels
Support/resistance retests: Price returns to test previous levels
\nThe time windows (1-10 periods) capture both immediate reversals and slightly delayed ones, which is crucial since reversal patterns can unfold over different timeframes.
\n\noccurs_within vs happens_within: These appear to be synonymous
\n
Then why must you add both of them to dataset as features? What is the diffirence? Or if they were exactly the same?
", "votes": null, "replies": [] } ] }, { "id": 3379813, "author_name": "johnoliverjones", "author_url": "", "post_date": "12/20/2025 20:18:38", "content": "Exactly correct. The columns are events of a variable, in a discretized range. It is my chore to build a transformer to recover the real variables and values.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3289149": "I was working on this dataset, but there are some variables that I don't fully understand. I think It will be helpful to have a minimum knowledge of what is each variable to understand more the problem and design better models. \n\nI know there are a lot of variables, but all are combinations of concepts in some numeric levels. For example, we have a lot of variables with this format: `occurs_within_A_{number}_B_{number}_{number}`, where A and B can be any feature of `peaks`, `zone`, `cross_threshold_from_{above/below}`, etc. Does anyone understand, or have some knowledge on features for detecting reversal points, and what are these variables telling us? For instance, I don't have any clue on what does the third number mean (in the example I gave) and the difference between occurs or happens. \n\nI hope this question helps other participants and that we can understand all these variables. \nGood luck with the competition!", "3292795": "Yes basically those are derived featured from base descriptors , we have initally ratio, momentum, sm_ratio and sm_momentum values as base descriptors, then from these we generate troughs , peaks, zones etc based on thresholds. Occurs with brings sequantial information too, happens has no order.\n - occurs_within vs happens_within: These appear to be synonymous - both check if an event occurs within a time window after another event\n \nEvent Types:\n - cross_threshold_from_above_{value}: Price crosses below a threshold from above\n - cross_threshold_from_below_{value}: Price crosses above a threshold from below\n - peaks_{value}: Local price peaks at specific levels\n - zones_{value}: Price enters specific zones/ranges\n \nThreshold Values:\n - 100.0, 100.5, 101.0, etc.: These are normalized price levels (likely as percentages of some reference price)\n\nThe Third Number (Time Window):\n - _1, _2, _3, ..., _10: This represents the number of periods/bars to look ahead\n - It answers: \"Does event B happen within X periods after event A?\"\n\n Example Interpretation:\n\n happens_within_cross_threshold_from_above_100.0_peaks_100.5_3\n\n Translation: \"Does a peak at 100.5 level occur within 3 periods after the price crosses below the 100.0 threshold from above?\"\n\n This is measuring sequential event patterns that are characteristic of reversal points:\n - First event: Price drops below 100.0 (potential support break)\n - Second event: Price makes a peak at 100.5 within 3 bars (failed breakdown, potential reversal)\n\n Reversal Detection Logic:\n\n These features capture common reversal patterns:\n 1. Failed breakdowns: Price breaks support but quickly reverses\n 2. Failed breakouts: Price breaks resistance but fails to sustain\n 3. Double bottoms/tops: Sequential peaks/troughs at similar levels\n 4. Support/resistance retests: Price returns to test previous levels\n\n The time windows (1-10 periods) capture both immediate reversals and slightly delayed ones, which is crucial since reversal patterns can unfold over different timeframes.", "3295036": "ozzymet \n>occurs_within vs happens_within: These appear to be synonymous\n \nThen why must you add both of them to dataset as features? What is the diffirence? Or if they were exactly the same?", "3379813": "Exactly correct. The columns are events of a variable, in a discretized range. It is my chore to build a transformer to recover the real variables and values." }, "source": "meta" }