-
Notifications
You must be signed in to change notification settings - Fork 8
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Errors #1
Comments
But I can call the following code
|
please install the following : https://github.com/msmbuilder/vde vde_metadynamics is built on top of that library |
how many features do you have? that looks like 2 tics wth 2 features |
Yes I have two tICS and 2 features. But in wwdomain case it seems like you provide the tIC data as a training set
What are these 1 and 2 in tica_data[i][::1,:2] are they mean tIC1 and tIC 2? Why it is 1 and 3 in the next line tica_data[i][::1,:3]? Whereas in case of alanine dipeptide you provided the sin_cos features as training set? Why so `: #model = VDE(4,100,hidden_layer_depth=2,hidden_size=16,n_epochs=50,learning_rate=1e-3) model.fit(sin_cos_feat)torch.save(model,"trained_ala.pkl")` |
The 1 is the stride (i.e. every frame), and 2 is going up to column 2 (so 0th and 1st tic). I played around with what stride to use, and how many tICs but 1 was fast enough
That 3 should be a 2. It doesnt matter though because python is smart enough to automatically only go up to the 2nd column.
I would recommend reading the manuscript for more details. But basically, i argued that training these models on large proteins takes too long and the models get very large very fast. That makes them terrible CVs since the compute time could easily exceed plain unbiased sampling. Using tICA as a pre-processing step allows us to have faster convergence since we are limited to a few tICs and we can build smaller networks so they are easier to sample.
yep 4 is the features and 100 is the lagtime |
When running this getting the following error
Error AttributeError Traceback (most recent call last) AttributeError: 'tICA' object has no attribute 'keys' |
Oh thats my fault. I use an older/different variant of the tica_data object. Basically, i use dictionaries where as the default msmbuilder output is list of lists now. To use my code for that, replace all of:
with
|
Gives an error like TypeError Traceback (most recent call last) TypeError: 'tICA' object is not iterable |
tica_data is the output of the tica_object and not the object itself. I.e its the projection of each of your simulation frames in this reduced dimension.
|
Okay I got this crazily huge Plumed input file but there are some questions There is existence of ARG=l0_0,l0_1 but where does in the script it is defined? I think it should be l0_0 instead of f0_0. RESTART DISTANCE ATOMS=3337,4559 LABEL=ca_215_293 MATHEVAL ARG=ca_35_79 FUNC=x LABEL=f0_0 PERIODIC=NO MATHEVAL ARG=ca_215_293 FUNC=x LABEL=f0_1 PERIODIC=NO COMBINE LABEL=l1_0nb ARG=l0_0,l0_1 COEFFICIENTS=0.06554490327835083,0.2746332287788391 PERIODIC=NO |
Also I think for a big protein driving on all these networks will be craziness. I guess driving on the first network is a good try? What do you think? |
oh weird, i thought the code automatically changed that
to
note the updated LABEL. I think i was using f0_0 at somepoint but realized that i should make it more generic so features became the 0th layer of the network (hence l0_0 is the 0th feature, l0_1 is the first feature ...) I will fix this in the code. Thanks for the bug report. |
Yea, you could try and reduce the complexity by changing the hidden layer depth. alternatively number of hidden nodes. |
Also make sure that the output of the network matches the output from pytorch. They should match up to 4 sig figs at least |
Hi,
So I am trying to plot my distance on the top of a tICA model
I used the following code
# lets visualize the data, and the dominant tICA solution from matplotlib import pyplot as plt plt.scatter(raw_feat[0][:,0],raw_feat[0][:,1],c=tica_features[0],cmap='viridis')
I get an error like
`---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
in ()
1 # lets visualize the data, and the dominant tICA solution
2 from matplotlib import pyplot as plt
----> 3 plt.scatter(raw_feat[0][:,0],raw_feat[0][:,1],c=tica_features[0],cmap='viridis')
TypeError: 'ContactFeaturizer' object does not support indexing`
I also installed vde however it seems like I can't call it
from vde import VDE
It produces an error like
`---------------------------------------------------------------------------
ModuleNotFoundError Traceback (most recent call last)
in ()
----> 1 from vde import VDE
ModuleNotFoundError: No module named 'vde'`
The text was updated successfully, but these errors were encountered: