tech/a
No Ordinary Duck
- Joined
- 14 October 2004
- Posts
- 20,417
- Reactions
- 6,356
Howard
Thank you for your rsponse and I like the analogy.
Can I pose the following to add perhaps something which is possibly not considered in the above.
The database which you are applying your conditions,arguements and variables to---your system.
We choose a database as this is from where we will extract our model to best take advantage of the characteristic we know is within that dataset.
Lets say that this is our dataset and since the beginning of time it has looked like this.
We also have our own objectives which we require in the results which we see when applying our system and its constituents to the data.
The out of sample data needs to be long enough to have any meaning full result on the out of sample testing.
To select an out of data sample which looks like this
Will have obviously negative results as the system wasnt designed for these conditions.
You will find similar conditions across most datasets.
So while your system has been designed across a large set of data your out of sample dataset is much less (Generally) if taken in isolation it can be misleading.You could throw away a perfectly viable method which suits your objectives.Hence the question of how many out of sample test results give rise to confidence.
My arguement is that in the end you want to outperform the dataset your designing your system to take part in.Wether it outperforms it at the very highest level,you will never determine as the data will be dynamic.Wether it fits perfectly your criteria ALL the time is also a pointless exercise. At times it will at times it wont. The destination is in my view more important than the journey as each time its applied through time the journey (Catastrophic events avoided as best we can) will be different to the in sample AND the out of sample testing.
Lets say this is my system with all the objectives,variables and conditions applicable to my own requirements and those that work well against the data base its designed for.
Here is my system--I like it!
I've selected all of my objectives and this is what I have come up with.
Continued in next post due to number of images.
Thank you for your rsponse and I like the analogy.
Can I pose the following to add perhaps something which is possibly not considered in the above.
The database which you are applying your conditions,arguements and variables to---your system.
We choose a database as this is from where we will extract our model to best take advantage of the characteristic we know is within that dataset.
Lets say that this is our dataset and since the beginning of time it has looked like this.
We also have our own objectives which we require in the results which we see when applying our system and its constituents to the data.
The out of sample data needs to be long enough to have any meaning full result on the out of sample testing.
To select an out of data sample which looks like this
Will have obviously negative results as the system wasnt designed for these conditions.
You will find similar conditions across most datasets.
So while your system has been designed across a large set of data your out of sample dataset is much less (Generally) if taken in isolation it can be misleading.You could throw away a perfectly viable method which suits your objectives.Hence the question of how many out of sample test results give rise to confidence.
My arguement is that in the end you want to outperform the dataset your designing your system to take part in.Wether it outperforms it at the very highest level,you will never determine as the data will be dynamic.Wether it fits perfectly your criteria ALL the time is also a pointless exercise. At times it will at times it wont. The destination is in my view more important than the journey as each time its applied through time the journey (Catastrophic events avoided as best we can) will be different to the in sample AND the out of sample testing.
Lets say this is my system with all the objectives,variables and conditions applicable to my own requirements and those that work well against the data base its designed for.
Here is my system--I like it!
I've selected all of my objectives and this is what I have come up with.
Continued in next post due to number of images.