![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F113660%2F3665f7d701b62b803c0102b51f47d533%2FSelection_043.png?generation=1604946037906933&alt=media\")
I am kind of lost understanding of the scoring calculations. Can anyone help me understand them with a beginner-friendly perspective?
", "messages": [ { "id": "1073303", "postDate": "11/09/2020 12:20:22", "content": "I am kind of lost understanding of the scoring calculations. Can anyone help me understand them with a beginner-friendly perspective?
", "rawMarkdown": "I am kind of lost understanding of the scoring calculations. Can anyone help me understand them with a beginner-friendly perspective?", "votes": null }, { "id": "1073637", "postDate": "11/09/2020 18:20:39", "content": "
in summary,
\nThank you so much! I guess I got lost understanding it step by step. I will remember to directly see the final form of the metric. I was actually trying to understand the line with the arrow.![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5433316%2F9c8766107152044c2728f52205aa4620%2Floss.jpg?generation=1604956729943288&alt=media\")
But, I totally understand what you said. How does one understand those in-between derivations without knowing what each term means? Like, what are those N's . And, its mentioned that \" We assume the ground truth positions to be modelled by a mixture of multi-dimensional independent Normal distributions over time, yielding the likelihood\" . Here, I was confused about why they have put x1…t and y1…t in the numerator.
\nAfter thinking about it now, I guess what they mean is.. We will be predicting our trajectory. And, our trajectory will be an n-dimensional independent normal distribution over time. In our pdf, what's the likelihood of the actual trajectory!
\nAm I thinking it right?
", "rawMarkdown": "Thank you so much! I guess I got lost understanding it step by step. I will remember to directly see the final form of the metric. I was actually trying to understand the line with the arrow. \n\nBut, I totally understand what you said. How does one understand those in-between derivations without knowing what each term means? Like, what are those N's . And, its mentioned that \" We assume the ground truth positions to be modelled by a mixture of multi-dimensional independent Normal distributions over time, yielding the likelihood\" . Here, I was confused about why they have put x1...t and y1...t in the numerator.\n\nAfter thinking about it now, I guess what they mean is.. We will be predicting our trajectory. And, our trajectory will be an n-dimensional independent normal distribution over time. In our pdf, what's the likelihood of the actual trajectory! \n\nAm I thinking it right?", "votes": null }, { "id": "1078405", "postDate": "11/14/2020 18:32:37", "content": "N is the normal distribution. N(x_pred|x_true) = exp(-(x_pred - x_true)^2/2) at single time step
\nAnd same for y. This assumption is how you get mean square error as the loss function for most of the machine learning problem.
\n(Though I think there is some (unimportant) normalization factor that we miss from this formula.)
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F113660%2Fdab91ba68edc714d8f10a44f4de36e16%2FSelection_187.png?generation=1605432500208528&alt=media\")
Lecture 10.2 — Mixtures of Experts [Neural Networks for Machine Learning]
\nLecture from the course Neural Networks for Machine Learning, as taught by Geoffrey Hinton (University of Toronto) on Coursera in 2012.
", "rawMarkdown": "\n\nLecture 10.2 — Mixtures of Experts [Neural Networks for Machine Learning]\n\nLecture from the course Neural Networks for Machine Learning, as taught by Geoffrey Hinton (University of Toronto) on Coursera in 2012.", "votes": null }, { "id": "1078772", "postDate": "11/15/2020 09:34:27", "content": "Thank you!
", "rawMarkdown": "Thank you!", "votes": null } ], "comments": [ { "id": 1073637, "author_name": "hengck23", "author_url": "", "post_date": "11/09/2020 18:20:39", "content": "
in summary,
\nThank you so much! I guess I got lost understanding it step by step. I will remember to directly see the final form of the metric. I was actually trying to understand the line with the arrow.
But, I totally understand what you said. How does one understand those in-between derivations without knowing what each term means? Like, what are those N's . And, its mentioned that \" We assume the ground truth positions to be modelled by a mixture of multi-dimensional independent Normal distributions over time, yielding the likelihood\" . Here, I was confused about why they have put x1…t and y1…t in the numerator.
\nAfter thinking about it now, I guess what they mean is.. We will be predicting our trajectory. And, our trajectory will be an n-dimensional independent normal distribution over time. In our pdf, what's the likelihood of the actual trajectory!
\nAm I thinking it right?
", "votes": null, "replies": [] }, { "id": 1078405, "author_name": "louis925", "author_url": "", "post_date": "11/14/2020 18:32:37", "content": "N is the normal distribution. N(x_pred|x_true) = exp(-(x_pred - x_true)^2/2) at single time step
\nAnd same for y. This assumption is how you get mean square error as the loss function for most of the machine learning problem.
\n(Though I think there is some (unimportant) normalization factor that we miss from this formula.)
Lecture 10.2 — Mixtures of Experts [Neural Networks for Machine Learning]
\nLecture from the course Neural Networks for Machine Learning, as taught by Geoffrey Hinton (University of Toronto) on Coursera in 2012.
", "votes": null, "replies": [ { "id": 1078772, "author_name": "rawwar", "author_url": "", "post_date": "11/15/2020 09:34:27", "content": "Thank you!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1073303": "I am kind of lost understanding of the scoring calculations. Can anyone help me understand them with a beginner-friendly perspective?", "1073637": "\n\nin summary,\n- when there is multi-mode, gives highest confidence to the prediction with lowest l2-loss trajectory", "1073727": "Thank you so much! I guess I got lost understanding it step by step. I will remember to directly see the final form of the metric. I was actually trying to understand the line with the arrow. \n\nBut, I totally understand what you said. How does one understand those in-between derivations without knowing what each term means? Like, what are those N's . And, its mentioned that \" We assume the ground truth positions to be modelled by a mixture of multi-dimensional independent Normal distributions over time, yielding the likelihood\" . Here, I was confused about why they have put x1...t and y1...t in the numerator.\n\nAfter thinking about it now, I guess what they mean is.. We will be predicting our trajectory. And, our trajectory will be an n-dimensional independent normal distribution over time. In our pdf, what's the likelihood of the actual trajectory! \n\nAm I thinking it right?", "1078405": "N is the normal distribution. N(x_pred|x_true) = exp(-(x_pred - x_true)^2/2) at single time step\nAnd same for y. This assumption is how you get mean square error as the loss function for most of the machine learning problem.\n(Though I think there is some (unimportant) normalization factor that we miss from this formula.)", "1078765": "\n\nLecture 10.2 — Mixtures of Experts [Neural Networks for Machine Learning]\n\nLecture from the course Neural Networks for Machine Learning, as taught by Geoffrey Hinton (University of Toronto) on Coursera in 2012.", "1078772": "Thank you!" }, "source": "meta" }