Watch videos with subtitles in your language, upload your videos, create your own subtitles! Click here to learn more on "how to Dotsub"


0 (0 Likes / 0 Dislikes)
So if you are around a bunch of models, let's say 100 models in the same data, this is a normal behavior of of space models in this XXXXX in which we plot the condition and condition error and we plot all the models that appear XXXXX ok? So from this information, we can say which models are good and which are bad let's see. this is the distribution area and these are the other points for each one of these models and each one of these points would represent one XXXXX of just one model in this part of the plot we have we have very high commission and low commission errors we know that we can trust in our presence data but our absence data as we mention is not that reliable, it's very difficult to confirm the absence of species so our strong XXXXXX here is our mission here more than our XXXXXX so if we have models with high XXXXXXX it means that many of the points validation points is not predicted by the model so this model is not reformed about here on the other extreme, we can do very well at predicting our precences if we predict the whole region we over predict, we will have severe omission error but also this represents a bad model because it's XXXXXXX predicting the distributions so we know that these parts of the plot and models aren't good there is also promise if we have more XXXX in these parts of the plot because this means they have XXXXXXXXX so this has the great promise that we can find in any model so we can set elite at the omission axis that we can upset, accept and XXXX this is the XXXXX that I was mentioning to you yesterday how much error are you going to accept based on how good is your input data OK well, the sample here is part of a 10% of error we can say that more or less above this we are not going to accept because they are not performing well so when we XXX with this set of models that have acceptable omission error also we know more or less that if you go to the etreme of the axis made me over feel and the model on the other side made me over predict so the region of the good model should fall somewhere here so these ideas have been of tremendous interest in the algorithm both in the stand alone software and the other model so right now we are going in open-modeller and see how to parameterize graphs in the XXXXX so if you open your own modeller and go to the algorithm profiles you see the list of your algorithms and you see that you have 4 implementations of GARP the first one says GARP single XXXXXXX best of GARP XXXXXX it means that XXXXXXX only one model this is single XXXX with GARP for the new XXXXXXXXXXXX when they recall the GARP for XXXXXXX modeller they make some subtle changes but those changes resulted that performed better than the original implementation? of GARP so we will recommend you to use the new open modeller implementation in this case then you have these other two options in which you can put this best XXXXXXXXX so for now we are going to run these GARP data sets in new open model XXXtations OK we are going to modify the fourth parameter, so you need to come out against it so you have here economy of XXXXXXXX best of set of the mutations??? and go to parameters and you would state the number of parameters that we can change the XXXXX train proportion which is the partition of the percentage of points you want for training let's go for 7.5., as we did in XXXXXXX then it says says in total rounds how many models have you wanted to use? because we are in affect XXXXXX save our models, in this case it says twenty GARP is like a very machine man software and it takes longer than XXXXXX or XXXXXX So the longer you are using that option it would take much longer to finish up your analysis so we are going to leave this XXXXX for now, but normally we use 50 or 100 models to have good creation of models then you have this option of hard omission phrase model?? this refers to the green line that we set here in this graph it's what percentage of omission pair you are accepting for a model to decide that this is a good one or not so we can say here 10%, we can say that models that have more than 10% of omission pair I want to discard if you read this first XXXXXXX it says set 100% used soft omission what is that? If we clear 100% we are not XXXXX any land in this omission this is going to run all the models and it's going to pick the number of models that just XXXXXX XXXXX with the lowest omission pair regardless of its value doesn't matter how big or small is that value, it will take certain number of models with the lowest omission when you use 100, so we are going to use 10% we are establishing a card omission XXXXXX and we are going to say here that initially it has to run 20 models that selects 10 models with these omission pairs so we have 20 models and this is going to select the 10 models that has 10 or less percent of omission pairs then it comes to the commision axis and you see here that we are selecting 50% of these 10 models that falls closest to the median this is kind of confusing I know, but remember this graph always for you to make decisions in these parameters we are saying here that for the 20 models that we are XXXXX we are going to keep 10 which I are these 10 right? from these 10 we are going to select 50% of them, 5 models that are closest to the median we are also getting these ones, OK and the other parameters are not relevant for what we are doing now some XXXXXXXXXXXXXXXXXXXXXXXX give you more than one XXXXXXXXX Now you apply the changes of your parameters, close XXXX, now you can draw your experiment with this part of the XXXXXX open a new experiment screen give a name, let's say XXXXXX so let only the line and in this case you are going to select the XXXXX one GARP as the algorithm to run the model then you could use the leadership Africa presence the prediction setting is going to be Africa presence your template is going to be one of the XXXXXXX in Africa data set you can use the same XXXXX as your output because you are running a different algorithm so it's going to be a differnt software XXXXXXX is fine and just put XXXX XXXXXXX Don't be suprised if so much XXXXXX in one, or if it's XXXXX XXXXX or your machine doesn't support it because it's much more intensive, this XXXXX so some things happen that is XXXXXXXXX yeah, put your models to run and you will continue and XXXXX the part of today keep running your own data with GARP XXXXX. you have some questions? any questions? One important thing when your own models are in softwares that are user friendly is you make an effort to document yourself about how each one of those algorithms work and don't be like blind users of these methods that's going to make a very important difference on the performance of the models OK so that model"s running and we are just going to leave it running some of these GARP models that in the old days when we used to use a desk top interface quite a bit some of them were run for a month... this one is going to run quick OK?

Video Details

Duration: 14 minutes and 18 seconds
Country: United States
Language: English
Views: 59
Posted by: townpeterson on Jul 12, 2013


Caption and Translate

    Sign In/Register for Dotsub above to caption this video.