The data column says mean(measurement value over probed positions)/mean( statistical error in measurement value over probed positions) as the calculation method of signal_to_noise.
However, the following calculation gives a close result, but I can't reproduce it.
\ndf_2A3 = df.loc[df.experiment_type == ]\nreact_2A3 = df_2A3[\n [c c df_2A3.columns c]\n].values\nreact_err_2A3 = df_2A3[\n [c c df_2A3.columns c]\n].values\nreact_mean = np.nanmean(react_2A3, axis=)\nreact_err_mean = np.nanmean(react_err_2A3, axis=)\n(react_mean / react_err_mean)\n\n> [ ... ]\n\n(df_2A3[].values)\n\n> [ ... ]\nHow to reproduce it completely?
", "messages": [ { "id": "2530922", "postDate": "11/19/2023 16:48:13", "content": "The data column says mean(measurement value over probed positions)/mean( statistical error in measurement value over probed positions) as the calculation method of signal_to_noise.
However, the following calculation gives a close result, but I can't reproduce it.
\ndf_2A3 = df.loc[df.experiment_type == ]\nreact_2A3 = df_2A3[\n [c c df_2A3.columns c]\n].values\nreact_err_2A3 = df_2A3[\n [c c df_2A3.columns c]\n].values\nreact_mean = np.nanmean(react_2A3, axis=)\nreact_err_mean = np.nanmean(react_err_2A3, axis=)\n(react_mean / react_err_mean)\n\n> [ ... ]\n\n(df_2A3[].values)\n\n> [ ... ]\nHow to reproduce it completely?
", "rawMarkdown": "The data column says `mean(measurement value over probed positions)/mean( statistical error in measurement value over probed positions)` as the calculation method of `signal_to_noise`. \n\nHowever, the following calculation gives a close result, but I can't reproduce it.\n```python\ndf_2A3 = df.loc[df.experiment_type == \"2A3_MaP\"]\nreact_2A3 = df_2A3[\n [c for c in df_2A3.columns if \"reactivity_0\" in c]\n].values\nreact_err_2A3 = df_2A3[\n [c for c in df_2A3.columns if \"reactivity_error_0\" in c]\n].values\nreact_mean = np.nanmean(react_2A3, axis=1)\nreact_err_mean = np.nanmean(react_err_2A3, axis=1)\nprint(react_mean / react_err_mean)\n\n> [ 0.986319 1.9689928 2.07819 ... 9.472149 15.853026 13.717828 ]\n\nprint(df_2A3[\"signal_to_noise\"].values)\n\n> [ 0.944 1.933 2.347 ... 9.554 15.995 13.799]\n```\n\nHow to reproduce it completely?", "votes": null }, { "id": "2531132", "postDate": "11/19/2023 22:41:15", "content": "I guess they are, but have you checked if the NaN positions on reactivity are the same than the ones on errors? Your calculation suggest just a small change.
", "rawMarkdown": "I guess they are, but have you checked if the NaN positions on reactivity are the same than the ones on errors? Your calculation suggest just a small change.", "votes": null }, { "id": "2531134", "postDate": "11/19/2023 22:49:35", "content": "The host explained here. There are several intricacies, like ignoring the first and last no-nan positions and ignoring sequences with less than four valid positions. See the exact function here.
", "rawMarkdown": "The host explained [here](https://www.kaggle.com/competitions/stanford-ribonanza-rna-folding/discussion/451158). There are several intricacies, like ignoring the first and last no-nan positions and ignoring sequences with less than four valid positions. See the exact function [here](https://github.com/DasLab/ubr/blob/main/matlab/data/ubr_estimate_signal_to_noise_ratio.m).", "votes": null }, { "id": "2531186", "postDate": "11/20/2023 02:24:09", "content": "Thank you, I didn't take care of that.
", "rawMarkdown": "Thank you, I didn't take care of that.", "votes": null }, { "id": "2531187", "postDate": "11/20/2023 02:24:50", "content": "I missed that post, will check it out.
", "rawMarkdown": "I missed that post, will check it out.", "votes": null }, { "id": "2532388", "postDate": "11/21/2023 03:00:52", "content": "According to the code. a signal is a positive value and a noise is the reactivity error. However these values are computed before normalization which is a value computed in a batch. We are unsure of what and where they sit in the batch. Some of these values may be from a large batch that were not given to us. But I think you got a good approximation
", "rawMarkdown": "According to the code. a signal is a positive value and a noise is the reactivity error. However these values are computed before normalization which is a value computed in a batch. We are unsure of what and where they sit in the batch. Some of these values may be from a large batch that were not given to us. But I think you got a good approximation", "votes": null } ], "comments": [ { "id": 2531132, "author_name": "sacuscreed", "author_url": "", "post_date": "11/19/2023 22:41:15", "content": "I guess they are, but have you checked if the NaN positions on reactivity are the same than the ones on errors? Your calculation suggest just a small change.
", "votes": null, "replies": [ { "id": 2531186, "author_name": "tattaka", "author_url": "", "post_date": "11/20/2023 02:24:09", "content": "Thank you, I didn't take care of that.
", "votes": null, "replies": [] } ] }, { "id": 2531134, "author_name": "shlomoron", "author_url": "", "post_date": "11/19/2023 22:49:35", "content": "The host explained here. There are several intricacies, like ignoring the first and last no-nan positions and ignoring sequences with less than four valid positions. See the exact function here.
", "votes": null, "replies": [ { "id": 2531187, "author_name": "tattaka", "author_url": "", "post_date": "11/20/2023 02:24:50", "content": "I missed that post, will check it out.
", "votes": null, "replies": [] } ] }, { "id": 2532388, "author_name": "tuttlen", "author_url": "", "post_date": "11/21/2023 03:00:52", "content": "According to the code. a signal is a positive value and a noise is the reactivity error. However these values are computed before normalization which is a value computed in a batch. We are unsure of what and where they sit in the batch. Some of these values may be from a large batch that were not given to us. But I think you got a good approximation
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2530922": "The data column says `mean(measurement value over probed positions)/mean( statistical error in measurement value over probed positions)` as the calculation method of `signal_to_noise`. \n\nHowever, the following calculation gives a close result, but I can't reproduce it.\n```python\ndf_2A3 = df.loc[df.experiment_type == \"2A3_MaP\"]\nreact_2A3 = df_2A3[\n [c for c in df_2A3.columns if \"reactivity_0\" in c]\n].values\nreact_err_2A3 = df_2A3[\n [c for c in df_2A3.columns if \"reactivity_error_0\" in c]\n].values\nreact_mean = np.nanmean(react_2A3, axis=1)\nreact_err_mean = np.nanmean(react_err_2A3, axis=1)\nprint(react_mean / react_err_mean)\n\n> [ 0.986319 1.9689928 2.07819 ... 9.472149 15.853026 13.717828 ]\n\nprint(df_2A3[\"signal_to_noise\"].values)\n\n> [ 0.944 1.933 2.347 ... 9.554 15.995 13.799]\n```\n\nHow to reproduce it completely?", "2531132": "I guess they are, but have you checked if the NaN positions on reactivity are the same than the ones on errors? Your calculation suggest just a small change.", "2531134": "The host explained [here](https://www.kaggle.com/competitions/stanford-ribonanza-rna-folding/discussion/451158). There are several intricacies, like ignoring the first and last no-nan positions and ignoring sequences with less than four valid positions. See the exact function [here](https://github.com/DasLab/ubr/blob/main/matlab/data/ubr_estimate_signal_to_noise_ratio.m).", "2531186": "Thank you, I didn't take care of that.", "2531187": "I missed that post, will check it out.", "2532388": "According to the code. a signal is a positive value and a noise is the reactivity error. However these values are computed before normalization which is a value computed in a batch. We are unsure of what and where they sit in the batch. Some of these values may be from a large batch that were not given to us. But I think you got a good approximation" }, "source": "meta" }