Ezward
Minor typos and suggestions in Week 6 Precision and Recall notes
On page 1:
What is our classifier has 90% accuracy?
What if our classifier has 90% accuracy?
In this particular example, we would never find a positive review, which what we are trying to do in our scenario.
In this particular example, we would never find a positive review, which is what we are trying to do in our scenario.
On page 3:
So we used the probability = 0.5 that the data is classified as positive to make the decision about the classification. We can simply choose a threshold other that 0.5 to make the tradeoff between precision and recall.
So we used as a threshold probability = 0.5 that the data is classified as positive to make the decision about the classification. We can simply choose a threshold other than 0.5 to make the tradeoff between precision and recall.
On page 5:
It is not always this clear. In some cases, the precision-recall curves for two classifiers my cross. This creates regions in which one curve is better (has higher precision for the same recall) than the other.
It is not always this clear. In some cases, the precision-recall curves for two classifiers may cross. This creates regions in which one classifier is better (has higher precision for the same recall) than the other.
It is common to use area-under-the-curve (AUC) measures to choose one curve over another. Area under the curve calculates a measure for a range of t, so that we can pick a curve that works best over a range.
It is common to use area-under-the-curve (AUC) measures to choose one classifier over another. Area under the curve calculates a measure for a range of t, so that we can pick a classifier that works best over a range of thresholds.
Also, the hyperlink to the image matching classifier research (great idea to link to current research, by the way!) is not clickable in the PDFs. Maybe it would be better to replace the word "here" with the URL?
Got it. Thank you Daniel. I will make the suggested changes.
On Thu, Apr 21, 2016 at 7:21 AM, Daniel Trebbien [email protected] wrote:
On page 1:
What is our classifier has 90% accuracy?
What if our classifier has 90% accuracy?
In this particular example, we would never find a positive review, which what we are trying to do in our scenario.
In this particular example, we would never find a positive review, which is what we are trying to do in our scenario.
On page 3:
So we used the probability = 0.5 that the data is classified as positive to make the decision about the classification. We can simply choose a threshold other that 0.5 to make the tradeoff between precision and recall.
So we used as a threshold probability = 0.5 that the data is classified as positive to make the decision about the classification. We can simply choose a threshold other than 0.5 to make the tradeoff between precision and recall.
On page 5:
It is not always this clear. In some cases, the precision-recall curves for two classifiers my cross. This creates regions in which one curve is better (has higher precision for the same recall) than the other.
It is not always this clear. In some cases, the precision-recall curves for two classifiers may cross. This creates regions in which one classifier is better (has higher precision for the same recall) than the other.
It is common to use area-under-the-curve (AUC) measures to choose one curve over another. Area under the curve calculates a measure for a range of t, so that we can pick a curve that works best over a range.
It is common to use area-under-the-curve (AUC) measures to choose one classifier over another. Area under the curve calculates a measure for a range of t, so that we can pick a classifier that works best over a range of thresholds.
Also, the hyperlink to the image matching classifier research (great idea to link to current research, by the way!) is not clickable in the PDFs. Maybe it would be better to replace the word "here" with the URL?
— You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub https://github.com/Ezward/machinelearningnotes/issues/2
Thanks again Daniel. I made changes and posted. I went back and fixed the hyperlinks in other documents as well based on your discovery that links don't work in the pdfs.
Ed
On Sun, Apr 24, 2016 at 10:23 AM, Ed Murphy [email protected] wrote:
Got it. Thank you Daniel. I will make the suggested changes.
On Thu, Apr 21, 2016 at 7:21 AM, Daniel Trebbien <[email protected]
wrote:
On page 1:
What is our classifier has 90% accuracy?
What if our classifier has 90% accuracy?
In this particular example, we would never find a positive review, which what we are trying to do in our scenario.
In this particular example, we would never find a positive review, which is what we are trying to do in our scenario.
On page 3:
So we used the probability = 0.5 that the data is classified as positive to make the decision about the classification. We can simply choose a threshold other that 0.5 to make the tradeoff between precision and recall.
So we used as a threshold probability = 0.5 that the data is classified as positive to make the decision about the classification. We can simply choose a threshold other than 0.5 to make the tradeoff between precision and recall.
On page 5:
It is not always this clear. In some cases, the precision-recall curves for two classifiers my cross. This creates regions in which one curve is better (has higher precision for the same recall) than the other.
It is not always this clear. In some cases, the precision-recall curves for two classifiers may cross. This creates regions in which one classifier is better (has higher precision for the same recall) than the other.
It is common to use area-under-the-curve (AUC) measures to choose one curve over another. Area under the curve calculates a measure for a range of t, so that we can pick a curve that works best over a range.
It is common to use area-under-the-curve (AUC) measures to choose one classifier over another. Area under the curve calculates a measure for a range of t, so that we can pick a classifier that works best over a range of thresholds.
Also, the hyperlink to the image matching classifier research (great idea to link to current research, by the way!) is not clickable in the PDFs. Maybe it would be better to replace the word "here" with the URL?
— You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub https://github.com/Ezward/machinelearningnotes/issues/2
Thanks Ed. I have downloaded the latest update and verified that the typos are fixed except for one on Page 3:
We can simply choose a threshold other that 0.5 to make the tradeoff between precision and recall.
We can simply choose a threshold other than 0.5 to make the tradeoff between precision and recall.