{ "id": 7590, "title": "Last minute starter", "url": "/competitions/galaxy-zoo-the-galaxy-challenge/discussion/7590", "author_name": "", "post_date": "2014-04-02T10:55:47.520Z", "votes": null, "comment_count": 7, "views": 3376, "content": "

Hi, I am trying to understand the values given in the training solutions.

\n

In the link Decision Tree, it gives an example which says that class1.1 has probability of 0.8, class 1.2 has a probability of 0.15 and class1.3 has a probability of 0.05. Then it says that 80% of users which have identified the galaxy as smooth , values of class 7.1 , 7.2 and 7.3 are given. There seems to be a dependence here between class 1.1 and class7.1.What is the value of 7.1, 7.2 or 7.3 if class1.2 is chosen. How do I find it in the training sample csv file.

", "messages": [ { "id": "41388", "postDate": "04/02/2014 10:55:47", "content": "

Hi, I am trying to understand the values given in the training solutions.

\n

In the link Decision Tree, it gives an example which says that class1.1 has probability of 0.8, class 1.2 has a probability of 0.15 and class1.3 has a probability of 0.05. Then it says that 80% of users which have identified the galaxy as smooth , values of class 7.1 , 7.2 and 7.3 are given. There seems to be a dependence here between class 1.1 and class7.1.What is the value of 7.1, 7.2 or 7.3 if class1.2 is chosen. How do I find it in the training sample csv file.

", "rawMarkdown": "", "votes": null }, { "id": "41393", "postDate": "04/02/2014 12:53:25", "content": "

If I understood the decision tree correctly, it goes as follows (I'll denote the probability of a class by P(class)):

\n

P(1.1) + P(1.2) + P(1.3) = 1.0
P(7.1) + P(7.2) + P(7.3) = 1.0 * P(1.1)

\n

And so on. Unless explicitly stated that the probabilities should sum to 1, the probabilities of a class should sum to the probability of the previous node in the decision tree.

", "rawMarkdown": "", "votes": null }, { "id": "41395", "postDate": "04/02/2014 12:58:33", "content": "

[quote=George Oblapenko;41393]

\n

If I understood the decision tree correctly, it goes as follows (I'll denote the probability of a class by P(class)):

\n

P(1.1) + P(1.2) + P(1.3) = 1.0
P(7.1) + P(7.2) + P(7.3) = 1.0 * P(1.1)

\n

And so on. Unless explicitly stated that the probabilities should sum to 1, the probabilities of a class should sum to the probability of the previous node in the decision tree.

\n

[/quote]

\n\n

What does the last statement mean? Can you elaborate further with one more example please?

", "rawMarkdown": "", "votes": null }, { "id": "41412", "postDate": "04/02/2014 18:35:26", "content": "

Abhishek, have you seen 'The Galaxy Zoo Decision Tree' page?

", "rawMarkdown": "", "votes": null }, { "id": "41421", "postDate": "04/03/2014 00:11:50", "content": "

One another question, why is this tree not considered as a "proper" decision tree. I have seen the post by user @joycenv who suggests not to use to KL algorithm, as it was not a proper decision tree. Can anyone suggest reason behind it.

", "rawMarkdown": "", "votes": null }, { "id": "41425", "postDate": "04/03/2014 05:24:16", "content": "

@Abhishek

\n

As I understood

\n

If Q1 (Round / Bar / else) is answered bij 60% (0.6) as 'Round'
Only this 60% (0.6) will go to Q7.
So Q7 can maximally score 0.6.

\n

I guess one might exploit this a bit by placing low importance on Questions with a low max score (ie Q11).

Q6: see sedielem's answer....

", "rawMarkdown": "", "votes": null }, { "id": "41426", "postDate": "04/03/2014 05:24:17", "content": "

For example, a galaxy can be of class 8 of and only if it is of class 6.1.

\n

So P(8.1) + P(8.2) + ... + P(8.7) = P(6.1)

", "rawMarkdown": "", "votes": null }, { "id": "41440", "postDate": "04/03/2014 11:58:19", "content": "

[quote=Julian de Wit;41425]One tricky Question is Q6 since it has multiple paths leading to it (Q5 and Q9).So you have to add up those scores coming from Q5 and Q9..[/quote]

\n

Actually, Q6 is the one exception this rule, its answers always sum to 1, even if Q5 and Q9 don't. See this post for an explanation.

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 41393, "author_name": "kunstmord", "author_url": "", "post_date": "04/02/2014 12:53:25", "content": "

If I understood the decision tree correctly, it goes as follows (I'll denote the probability of a class by P(class)):

\n

P(1.1) + P(1.2) + P(1.3) = 1.0
P(7.1) + P(7.2) + P(7.3) = 1.0 * P(1.1)

\n

And so on. Unless explicitly stated that the probabilities should sum to 1, the probabilities of a class should sum to the probability of the previous node in the decision tree.

", "votes": null, "replies": [] }, { "id": 41395, "author_name": "abhishek", "author_url": "", "post_date": "04/02/2014 12:58:33", "content": "

[quote=George Oblapenko;41393]

\n

If I understood the decision tree correctly, it goes as follows (I'll denote the probability of a class by P(class)):

\n

P(1.1) + P(1.2) + P(1.3) = 1.0
P(7.1) + P(7.2) + P(7.3) = 1.0 * P(1.1)

\n

And so on. Unless explicitly stated that the probabilities should sum to 1, the probabilities of a class should sum to the probability of the previous node in the decision tree.

\n

[/quote]

\n\n

What does the last statement mean? Can you elaborate further with one more example please?

", "votes": null, "replies": [] }, { "id": 41412, "author_name": "sbachish", "author_url": "", "post_date": "04/02/2014 18:35:26", "content": "

Abhishek, have you seen 'The Galaxy Zoo Decision Tree' page?

", "votes": null, "replies": [] }, { "id": 41421, "author_name": "mohammadmotiurrahman", "author_url": "", "post_date": "04/03/2014 00:11:50", "content": "

One another question, why is this tree not considered as a "proper" decision tree. I have seen the post by user @joycenv who suggests not to use to KL algorithm, as it was not a proper decision tree. Can anyone suggest reason behind it.

", "votes": null, "replies": [] }, { "id": 41425, "author_name": "juliandewit", "author_url": "", "post_date": "04/03/2014 05:24:16", "content": "

@Abhishek

\n

As I understood

\n

If Q1 (Round / Bar / else) is answered bij 60% (0.6) as 'Round'
Only this 60% (0.6) will go to Q7.
So Q7 can maximally score 0.6.

\n

I guess one might exploit this a bit by placing low importance on Questions with a low max score (ie Q11).

Q6: see sedielem's answer....

", "votes": null, "replies": [] }, { "id": 41426, "author_name": "kunstmord", "author_url": "", "post_date": "04/03/2014 05:24:17", "content": "

For example, a galaxy can be of class 8 of and only if it is of class 6.1.

\n

So P(8.1) + P(8.2) + ... + P(8.7) = P(6.1)

", "votes": null, "replies": [] }, { "id": 41440, "author_name": "sedielem", "author_url": "", "post_date": "04/03/2014 11:58:19", "content": "

[quote=Julian de Wit;41425]One tricky Question is Q6 since it has multiple paths leading to it (Q5 and Q9).So you have to add up those scores coming from Q5 and Q9..[/quote]

\n

Actually, Q6 is the one exception this rule, its answers always sum to 1, even if Q5 and Q9 don't. See this post for an explanation.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "41388": "", "41393": "", "41395": "", "41412": "", "41421": "", "41425": "", "41426": "", "41440": "" }, "source": "meta" }