Closed GreenGilad closed 2 years ago
The second commit contains multiple changes of the following types:
the following:
if visual == False:
data = self.data
else:
data = data_visual
changed to :
data = data_visual if visual else self.data
if too_big == False
to if not too_big
getting array's dimensions. Instead of
num_dims = data.shape[1]
n_elements = data.shape[0]
then n_elements, num_dims = data.shape
In addition, pop_list
in run_VIA
was defined but never used so was removed
I also split the examples of all the different datasets to a separate file - this way it is easier to understand which part is the VIA algorithm and which is the examples.
There are still many lines in run_subPARC
that have code specific for all of these examples (2610-2648, 2716-2802 and the functions that are called there fore finding the root). Maybe it is worth having classes that derive from the VIA class for these cases?
Thank you so very much for all your really neat contributions and modifications. I really appreciate your efforts to polish the code and make it neater. A big thank you :)
@ShobiStassen if course :) I think this is a nice algorithm and it seems to be working nicely for my problem
@GreenGilad glad to hear it is working for your problem. what kind of data set are you using Via on, if i might ask? I find the parameter for graph pruning, cluster_graph_pruning_std, and the number of visualized outgoing edges, max_visual_outgoing_edges, to be quite useful to play around with to observe different levels of complexity of the trajectories. there are actually lots of things I want to improve/add for Via, just been busy/distracted with other projects cropping up!
@ShobiStassen I have sent you an email to explain a bit more about the data
In the current version if
true_label
is not passed in VIA's ctor one get the following exception:This happens because in the
run_VIA
functionself.true_label
is used in line 3615 but is only initialized in line 3618. In the suggested fix I simply moved the initialization ofself.true_label
to the ctor