Adapt to a new data with different dimension

[snehashis-roy]

Hi,
I'd like to train on one dataset but apply to new dataset with different number of columns, so that the number of neurons differ.
Here is my template training code,

X=np.loadtxt('load/a/7200/335/matrix.csv', delimiter=',')
max_iterations = 3000 #default is 10000.
cebra_time_model = CEBRA(model_architecture='offset10-model',
                        batch_size=512,
                        learning_rate=3e-4,
                        temperature=1.12,
                        output_dimension=3,
                        max_iterations=max_iterations,
                        distance='cosine',
                        conditional='time',
                        device='cuda',  
                        verbose=True,
                        time_offsets=10)

cebra_time_model.fit(X, adapt=True)
Y=np.loadtxt('load/a/7200/362/matrix.csv', delimiter=',')
cebra_time1 = cebra_time_model.transform(Y)

I am getting a following error:

RuntimeError                              Traceback (most recent call last)
Cell In[199], line 20
     17 #print(outputmodel)
     18 #print(A[0])
     19 cebra_time_model = cebra.CEBRA.load(outputmodel)
---> 20 cebra_time1 = cebra_time_model.transform(X1)
     21 cebra_time_model = cebra.CEBRA.load(outputmodel)
     22 cebra_time2 = cebra_time_model.transform(X2)

File ~/Programs/anaconda3.10/lib/python3.10/site-packages/sklearn/utils/_set_output.py:142, in _wrap_method_output.<locals>.wrapped(self, X, *args, **kwargs)
    140 @wraps(f)
    141 def wrapped(self, X, *args, **kwargs):
--> 142     data_to_wrap = f(self, X, *args, **kwargs)
    143     if isinstance(data_to_wrap, tuple):
    144         # only wrap the first output for cross decomposition
    145         return (
    146             _wrap_data_with_container(method, data_to_wrap[0], X, self),
    147             *data_to_wrap[1:],
    148         )

File ~/Programs/anaconda3.10/lib/python3.10/site-packages/cebra/integrations/sklearn/cebra.py:1244, in CEBRA.transform(self, X, session_id)
   1241 if isinstance(model, cebra.models.ConvolutionalModelMixin):
   1242     # Fully convolutional evaluation, switch (T, C) -> (1, C, T)
   1243     X = X.transpose(1, 0).unsqueeze(0)
-> 1244     output = model(X).cpu().numpy().squeeze(0).transpose(1, 0)
   1245 else:
   1246     # Standard evaluation, (T, C, dt)
   1247     output = model(X).cpu().numpy()

File ~/Programs/anaconda3.10/lib/python3.10/site-packages/torch/nn/modules/module.py:1532, in Module._wrapped_call_impl(self, *args, **kwargs)
   1530     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   1531 else:
-> 1532     return self._call_impl(*args, **kwargs)

File ~/Programs/anaconda3.10/lib/python3.10/site-packages/torch/nn/modules/module.py:1541, in Module._call_impl(self, *args, **kwargs)
   1536 # If we don't have any hooks, we want to skip the rest of the logic in
   1537 # this function, and just call forward.
   1538 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   1539         or _global_backward_pre_hooks or _global_backward_hooks
   1540         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1541     return forward_call(*args, **kwargs)
   1543 try:
   1544     result = None

File ~/Programs/anaconda3.10/lib/python3.10/site-packages/cebra/models/model.py:240, in _OffsetModel.forward(self, inp)
    228 def forward(self, inp):
    229     """Compute the embedding given the input signal.
    230 
    231     Args:
   (...)
    238     is normalized to the hypersphere (`normalize = True`).
    239     """
--> 240     return self.net(inp)

File ~/Programs/anaconda3.10/lib/python3.10/site-packages/torch/nn/modules/module.py:1532, in Module._wrapped_call_impl(self, *args, **kwargs)
   1530     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   1531 else:
-> 1532     return self._call_impl(*args, **kwargs)

File ~/Programs/anaconda3.10/lib/python3.10/site-packages/torch/nn/modules/module.py:1541, in Module._call_impl(self, *args, **kwargs)
   1536 # If we don't have any hooks, we want to skip the rest of the logic in
   1537 # this function, and just call forward.
   1538 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   1539         or _global_backward_pre_hooks or _global_backward_hooks
   1540         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1541     return forward_call(*args, **kwargs)
   1543 try:
   1544     result = None

File ~/Programs/anaconda3.10/lib/python3.10/site-packages/torch/nn/modules/container.py:217, in Sequential.forward(self, input)
    215 def forward(self, input):
    216     for module in self:
--> 217         input = module(input)
    218     return input

File ~/Programs/anaconda3.10/lib/python3.10/site-packages/torch/nn/modules/module.py:1532, in Module._wrapped_call_impl(self, *args, **kwargs)
   1530     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   1531 else:
-> 1532     return self._call_impl(*args, **kwargs)

File ~/Programs/anaconda3.10/lib/python3.10/site-packages/torch/nn/modules/module.py:1541, in Module._call_impl(self, *args, **kwargs)
   1536 # If we don't have any hooks, we want to skip the rest of the logic in
   1537 # this function, and just call forward.
   1538 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   1539         or _global_backward_pre_hooks or _global_backward_hooks
   1540         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1541     return forward_call(*args, **kwargs)
   1543 try:
   1544     result = None

File ~/Programs/anaconda3.10/lib/python3.10/site-packages/torch/nn/modules/conv.py:310, in Conv1d.forward(self, input)
    309 def forward(self, input: Tensor) -> Tensor:
--> 310     return self._conv_forward(input, self.weight, self.bias)

File ~/Programs/anaconda3.10/lib/python3.10/site-packages/torch/nn/modules/conv.py:306, in Conv1d._conv_forward(self, input, weight, bias)
    302 if self.padding_mode != 'zeros':
    303     return F.conv1d(F.pad(input, self._reversed_padding_repeated_twice, mode=self.padding_mode),
    304                     weight, bias, self.stride,
    305                     _single(0), self.dilation, self.groups)
--> 306 return F.conv1d(input, weight, bias, self.stride,
    307                 self.padding, self.dilation, self.groups)

RuntimeError: Given groups=1, weight of size [32, 335, 2], expected input[1, 362, 7207] to have 335 channels, but got 362 channels instead

Here is the trained model trained on a 7200x335 matrix, while the test data is 7198x362.

https://hpc.nih.gov/~roys5/template_trained_on_7200x335.pt

I appreciate any help to fix my code to adapt to a new data.
Thank you.

[MMathisLab]

As @stes stated in this answer here, you need to use "adapt = True"- #160

I suggest checking out the user documentation on this feature.

[snehashis-roy]

I believe I am using the adapt=True option. Here is a reproducible code.

from cebra import CEBRA
import numpy as np
X=np.random.normal(loc=0,scale=1,size=(7200,200))
cebra_time_model = CEBRA(model_architecture='offset10-model',
                        batch_size=512,
                        learning_rate=3e-4,
                        temperature=1.12,
                        output_dimension=3,
                        max_iterations=100,
                        distance='cosine',
                        conditional='time',
                        device='cuda', 
                        verbose=True,
                        time_offsets=10)
cebra_time_model.fit(X, adapt=True)

Y=np.random.normal(loc=0,scale=1,size=(7200,150))
cebra_time1 = cebra_time_model.transform(Y)

I used cebra_time_model.fit(X, adapt=True)
The error persists,

RuntimeError: Given groups=1, weight of size [32, 200, 2], expected input[1, 150, 7209] to have 200 channels, but got 150 channels instead

Could you please take a look the code and let me know what I am missing? Thank you.