exchange master branch

.gitignore

@@ -0,0 +1,3 @@
+__pycache__
+.ipynb_checkpoints
+*.egg-info

nnvision/datasets/monkey_loaders.py

@@ -6,9 +6,8 @@
 from collections import namedtuple, Iterable
 import os
 from pathlib import Path
-from mlutils.data.samplers import RepeatsBatchSampler
 from .utility import get_validation_split, ImageCache, get_cached_loader, get_fraction_of_training_images
-from nnfabrik.utility.nn_helpers import get_module_output, set_random_seed, get_dims_for_loader_dict
+from nnfabrik.utility.nn_helpers import set_random_seed, get_dims_for_loader_dict
 from nnfabrik.utility.dj_helpers import make_hash
 
 
@@ -177,7 +176,6 @@ def monkey_static_loader(dataset,
         dataloaders["validation"][data_key] = val_loader
         dataloaders["test"][data_key] = test_loader
 
-
     if store_data_info and not os.path.exists(stats_path):
 
         in_name, out_name = next(iter(list(dataloaders["train"].values())[0]))._fields

nnvision/datasets/mouse_loaders.py

@@ -3,9 +3,9 @@
 import warnings
 
 import numpy as np
-from mlutils.data.datasets import StaticImageSet, FileTreeDataset
-from mlutils.data.transforms import Subsample, ToTensor, NeuroNormalizer, AddBehaviorAsChannels, SelectInputChannel
-from mlutils.data.samplers import SubsetSequentialSampler
+from neuralpredictors.data.datasets import StaticImageSet, FileTreeDataset
+from neuralpredictors.data.transforms import Subsample, ToTensor, NeuroNormalizer, AddBehaviorAsChannels, SelectInputChannel
+from neuralpredictors.data.samplers import SubsetSequentialSampler
 from nnfabrik.utility.nn_helpers import set_random_seed
 from .utility import get_oracle_dataloader
 from torch.utils.data import DataLoader

nnvision/datasets/utility.py

@@ -6,7 +6,7 @@
 from skimage.transform import rescale
 from collections import namedtuple, Iterable
 import os
-from mlutils.data.samplers import RepeatsBatchSampler
+from neuralpredictors.data.samplers import RepeatsBatchSampler
 
 
 def get_oracle_dataloader(dat,

nnvision/legacy/featurevis/main.py

@@ -7,7 +7,7 @@
 from nnfabrik.main import Dataset, schema
 from nnfabrik.utility.dj_helpers import make_hash
 from . import integration
-from mlutils.data.datasets import StaticImageSet
+from neuralpredictors.data.datasets import StaticImageSet
 
 
 class TrainedEnsembleModelTemplate(dj.Manual):

nnvision/legacy/nnfabrik/datasets/mouse.py

@@ -6,9 +6,9 @@
 from torch.utils.data import DataLoader
 from torch.utils.data.sampler import SubsetRandomSampler
 
-from mlutils.data.datasets import StaticImageSet
-from mlutils.data.transforms import Subsample, ToTensor, NeuroNormalizer, AddBehaviorAsChannels
-from mlutils.data.samplers import SubsetSequentialSampler
+from neuralpredictors.data.datasets import StaticImageSet
+from neuralpredictors.data.transforms import Subsample, ToTensor, NeuroNormalizer, AddBehaviorAsChannels
+from neuralpredictors.data.samplers import SubsetSequentialSampler
 
 from ..utility.nn_helpers import set_random_seed
 

nnvision/legacy/nnfabrik/datasets/movies.py

@@ -1,7 +1,7 @@
 # Mouse Movie Datasets
 import torch
-from mlutils.data.datasets import MovieSet
-from mlutils.data.transforms import Subsequence, Subsample, Normalizer, ToTensor
+from neuralpredictors.data.datasets import MovieSet
+from neuralpredictors.data.transforms import Subsequence, Subsample, Normalizer, ToTensor
 from torch.utils.data.sampler import SubsetRandomSampler
 from torch.utils.data import DataLoader
 import numpy as np

nnvision/legacy/nnfabrik/datasets/sysident_v1_dataset.py

@@ -5,7 +5,7 @@
 #from retina.retina import warp_image
 from collections import namedtuple, Iterable
 import os
-from mlutils.data.samplers import RepeatsBatchSampler
+from neuralpredictors.data.samplers import RepeatsBatchSampler
 
 
 class ImageCache:

nnvision/legacy/nnfabrik/measures/measures.py

@@ -1,8 +1,8 @@
 import warnings
 import numpy as np
 import torch
-from mlutils.measures import corr
-from mlutils.training import eval_state, device_state
+from neuralpredictors.measures import corr
+from neuralpredictors.training import eval_state, device_state
 import types
 import contextlib
 import warnings

nnvision/legacy/nnfabrik/models/dynamic_models/cores.py

@@ -7,7 +7,7 @@
 from torch.nn import functional as F, Parameter
 
 from .misc import DepthSeparableConv2d
-from mlutils.regularizers import LaplaceL2, LaplaceL23d
+from neuralpredictors.regularizers import LaplaceL2, LaplaceL23d
 from . import logger as log
 
 

nnvision/legacy/nnfabrik/models/dynamic_models/readouts.py

@@ -1,6 +1,6 @@
 from pprint import pformat
 
-from mlutils.layers.readouts import SpatialTransformerPooled3d
+from neuralpredictors.layers.readouts import SpatialTransformerPooled3d
 from torch.nn import ModuleDict
 
 from . import logger as log

nnvision/legacy/nnfabrik/models/gaussian_readout_models.py

@@ -6,12 +6,12 @@
 from torch.nn import Parameter
 from torch.nn import functional as F
 from torch.nn import ModuleDict
-from mlutils.constraints import positive
-from mlutils.layers.cores import DepthSeparableConv2d, Core2d, Stacked2dCore
+from neuralpredictors.constraints import positive
+from neuralpredictors.layers.cores import DepthSeparableConv2d, Core2d, Stacked2dCore
 from ..utility.nn_helpers import get_io_dims, get_module_output, set_random_seed, get_dims_for_loader_dict
-from mlutils import regularizers
-from mlutils.layers.readouts import PointPooled2d
-from mlutils.layers.legacy import Gaussian2d
+from neuralpredictors import regularizers
+from neuralpredictors.layers.readouts import PointPooled2d
+from neuralpredictors.layers.legacy import Gaussian2d
 from .pretrained_models import TransferLearningCore
 
 # Squeeze and Excitation Block
@@ -69,7 +69,7 @@ def __init__(
             input_kern:     kernel size of the first layer (i.e. the input layer)
             hidden_kern:    kernel size of each hidden layer's kernel
             layers:         number of layers
-            gamma_input:    regularizer factor for the input weights (default: LaplaceL2, see mlutils.regularizers)
+            gamma_input:    regularizer factor for the input weights (default: LaplaceL2, see neuralpredictors.regularizers)
             skip:           Adds a skip connection
             final_nonlinearity: Boolean, if true, appends an ELU layer after the last BatchNorm (if BN=True)
             bias:           Adds a bias layer. Note: bias and batch_norm can not both be true
@@ -81,7 +81,7 @@ def __init__(
                 the kernel size (recommended). Setting Padding to 0 is not recommended and leads to artefacts,
                 zero is the default however to recreate backwards compatibility.
             normalize_laplace_regularizer: Boolean, if set to True, will use the LaplaceL2norm function from
-                mlutils.regularizers, which returns the regularizer as |laplace(filters)| / |filters|
+                neuralpredictors.regularizers, which returns the regularizer as |laplace(filters)| / |filters|
             input_regularizer:  String that must match one of the regularizers in ..regularizers
             stack:        Int or iterable. Selects which layers of the core should be stacked for the readout.
                             default value will stack all layers on top of each other.
@@ -204,7 +204,7 @@ def __init__(
             input_kern:     kernel size of the first layer (i.e. the input layer)
             hidden_kern:    kernel size of each hidden layer's kernel
             layers:         number of layers
-            gamma_input:    regularizer factor for the input weights (default: LaplaceL2, see mlutils.regularizers)
+            gamma_input:    regularizer factor for the input weights (default: LaplaceL2, see neuralpredictors.regularizers)
             skip:           Adds a skip connection
             final_nonlinearity: Boolean, if true, appends an ELU layer after the last BatchNorm (if BN=True)
             bias:           Adds a bias layer. Note: bias and batch_norm can not both be true
@@ -216,7 +216,7 @@ def __init__(
                 the kernel size (recommended). Setting Padding to 0 is not recommended and leads to artefacts,
                 zero is the default however to recreate backwards compatibility.
             normalize_laplace_regularizer: Boolean, if set to True, will use the LaplaceL2norm function from
-                mlutils.regularizers, which returns the regularizer as |laplace(filters)| / |filters|
+                neuralpredictors.regularizers, which returns the regularizer as |laplace(filters)| / |filters|
             input_regularizer:  String that must match one of the regularizers in ..regularizers
             stack:        Int or iterable. Selects which layers of the core should be stacked for the readout.
                             default value will stack all layers on top of each other.
@@ -359,16 +359,16 @@ def se_core_gauss_readout(dataloaders, seed, hidden_channels=32, input_kern=13,
                           depth_separable=False,
                           ):
     """
-    Model class of a stacked2dCore (from mlutils) and a pointpooled (spatial transformer) readout
+    Model class of a stacked2dCore (from neuralpredictors) and a pointpooled (spatial transformer) readout
 
     Args:
         dataloaders: a dictionary of dataloaders, one loader per session
             in the format {'data_key': dataloader object, .. }
         seed: random seed
         elu_offset: Offset for the output non-linearity [F.elu(x + self.offset)]
 
-        all other args: See Documentation of Stacked2dCore in mlutils.layers.cores and
-            PointPooled2D in mlutils.layers.readouts
+        all other args: See Documentation of Stacked2dCore in neuralpredictors.layers.cores and
+            PointPooled2D in neuralpredictors.layers.readouts
 
     Returns: An initialized model which consists of model.core and model.readout
     """
@@ -404,7 +404,7 @@ def regularizer(self, data_key):
 
     set_random_seed(seed)
 
-    # get a stacked2D core from mlutils
+    # get a stacked2D core from neuralpredictors
     core = SE2dCore(input_channels=input_channels[0],
                     hidden_channels=hidden_channels,
                     input_kern=input_kern,
@@ -451,16 +451,16 @@ def ds_core_gauss_readout(dataloaders, seed, hidden_channels=32, input_kern=13,
                                  gamma_readout=4,  elu_offset=0, stack=None,
                                  ):
     """
-    Model class of a stacked2dCore (from mlutils) and a pointpooled (spatial transformer) readout
+    Model class of a stacked2dCore (from neuralpredictors) and a pointpooled (spatial transformer) readout
 
     Args:
         dataloaders: a dictionary of dataloaders, one loader per session
             in the format {'data_key': dataloader object, .. }
         seed: random seed
         elu_offset: Offset for the output non-linearity [F.elu(x + self.offset)]
 
-        all other args: See Documentation of Stacked2dCore in mlutils.layers.cores and
-            PointPooled2D in mlutils.layers.readouts
+        all other args: See Documentation of Stacked2dCore in neuralpredictors.layers.cores and
+            PointPooled2D in neuralpredictors.layers.readouts
 
     Returns: An initialized model which consists of model.core and model.readout
     """
@@ -496,7 +496,7 @@ def regularizer(self, data_key):
 
     set_random_seed(seed)
 
-    # get a stacked2D core from mlutils
+    # get a stacked2D core from neuralpredictors
     core = DepthSeparableCore(input_channels=input_channels[0],
                          hidden_channels=hidden_channels,
                          input_kern=input_kern,
@@ -540,16 +540,16 @@ def ds_core_point_readout(dataloaders, seed, hidden_channels=32, input_kern=13,
                                  init_range=0.2, gamma_readout=0.1,  elu_offset=0, stack=None,
                                  ):
     """
-    Model class of a stacked2dCore (from mlutils) and a pointpooled (spatial transformer) readout
+    Model class of a stacked2dCore (from neuralpredictors) and a pointpooled (spatial transformer) readout
 
     Args:
         dataloaders: a dictionary of dataloaders, one loader per session
             in the format {'data_key': dataloader object, .. }
         seed: random seed
         elu_offset: Offset for the output non-linearity [F.elu(x + self.offset)]
 
-        all other args: See Documentation of Stacked2dCore in mlutils.layers.cores and
-            PointPooled2D in mlutils.layers.readouts
+        all other args: See Documentation of Stacked2dCore in neuralpredictors.layers.cores and
+            PointPooled2D in neuralpredictors.layers.readouts
 
     Returns: An initialized model which consists of model.core and model.readout
     """
@@ -583,7 +583,7 @@ def regularizer(self, data_key):
 
     set_random_seed(seed)
 
-    # get a stacked2D core from mlutils
+    # get a stacked2D core from neuralpredictors
     core = DepthSeparableCore(input_channels=input_channels[0],
                          hidden_channels=hidden_channels,
                          input_kern=input_kern,
@@ -627,16 +627,16 @@ def stacked2d_core_gaussian_readout(dataloaders, seed, hidden_channels=32, input
                                  gamma_readout=0.1,  elu_offset=0, stack=None,
                                  ):
     """
-    Model class of a stacked2dCore (from mlutils) and a pointpooled (spatial transformer) readout
+    Model class of a stacked2dCore (from neuralpredictors) and a pointpooled (spatial transformer) readout
 
     Args:
         dataloaders: a dictionary of dataloaders, one loader per session
             in the format {'data_key': dataloader object, .. }
         seed: random seed
         elu_offset: Offset for the output non-linearity [F.elu(x + self.offset)]
 
-        all other args: See Documentation of Stacked2dCore in mlutils.layers.cores and
-            PointPooled2D in mlutils.layers.readouts
+        all other args: See Documentation of Stacked2dCore in neuralpredictors.layers.cores and
+            PointPooled2D in neuralpredictors.layers.readouts
 
     Returns: An initialized model which consists of model.core and model.readout
     """
@@ -670,7 +670,7 @@ def regularizer(self, data_key):
 
     set_random_seed(seed)
 
-    # get a stacked2D core from mlutils
+    # get a stacked2D core from neuralpredictors
     core = Stacked2dCore(input_channels=input_channels[0],
                          hidden_channels=hidden_channels,
                          input_kern=input_kern,

nnvision/legacy/nnfabrik/models/pretrained_models.py

@@ -1,5 +1,5 @@
-from mlutils.layers.readouts import PointPooled2d
-from mlutils.layers.cores import Core2d, Core
+from neuralpredictors.layers.readouts import PointPooled2d
+from neuralpredictors.layers.cores import Core2d, Core
 from ..utility.nn_helpers import get_io_dims, get_module_output, set_random_seed, get_dims_for_loader_dict
 
 from itertools import count

nnvision/legacy/nnfabrik/models/v1_models.py

@@ -2,9 +2,9 @@
 from torch import nn as nn
 from torch.nn import functional as F
 
-from mlutils.layers.readouts import PointPooled2d
-from mlutils.layers.cores import Stacked2dCore
-from mlutils.training import eval_state
+from neuralpredictors.layers.readouts import PointPooled2d
+from neuralpredictors.layers.cores import Stacked2dCore
+from neuralpredictors.training import eval_state
 
 from .pretrained_models import TransferLearningCore
 from ..utility.nn_helpers import get_io_dims, get_module_output, set_random_seed, get_dims_for_loader_dict
@@ -47,16 +47,16 @@ def stacked2d_core_point_readout(dataloaders, seed, hidden_channels=32, input_ke
                                  gamma_readout=0.1,  elu_offset=0, stack=None, readout_reg_avg=False,
                                  use_avg_reg=False):
     """
-    Model class of a stacked2dCore (from mlutils) and a pointpooled (spatial transformer) readout
+    Model class of a stacked2dCore (from neuralpredictors) and a pointpooled (spatial transformer) readout
 
     Args:
         dataloaders: a dictionary of dataloaders, one loader per session
             in the format {'data_key': dataloader object, .. }
         seed: random seed
         elu_offset: Offset for the output non-linearity [F.elu(x + self.offset)]
 
-        all other args: See Documentation of Stacked2dCore in mlutils.layers.cores and
-            PointPooled2D in mlutils.layers.readouts
+        all other args: See Documentation of Stacked2dCore in neuralpredictors.layers.cores and
+            PointPooled2D in neuralpredictors.layers.readouts
 
     Returns: An initialized model which consists of model.core and model.readout
     """
@@ -109,7 +109,7 @@ def tracked_values(self):
 
     set_random_seed(seed)
 
-    # get a stacked2D core from mlutils
+    # get a stacked2D core from neuralpredictors
     core = Stacked2dCore(input_channels=input_channels[0],
                          hidden_channels=hidden_channels,
                          input_kern=input_kern,

nnvision/legacy/nnfabrik/training/dynamic_trainers.py

@@ -1,6 +1,6 @@
 from collections import namedtuple
 
-from mlutils.measures import PoissonLoss3d
+from neuralpredictors.measures import PoissonLoss3d
 
 
 def slice_iter(n, step):

nnvision/legacy/nnfabrik/training/dynamic_training/__init__.py

@@ -1,4 +1,4 @@
-from mlutils.measures import PoissonLoss3d
+from neuralpredictors.measures import PoissonLoss3d
 from .utils import (
     corr,
     ptcorr,
@@ -8,7 +8,7 @@
     correlation_closure,
     compute_scores,
 )
-from mlutils.training import cycle_datasets, early_stopping
+from neuralpredictors.training import cycle_datasets, early_stopping
 from tqdm import tqdm
 from logging import getLogger
 from itertools import product, repeat

nnvision/legacy/nnfabrik/training/trainers.py

@@ -6,9 +6,9 @@
 from scipy import stats
 from tqdm import tqdm
 
-from mlutils import measures
-from mlutils.measures import *
-from mlutils.training import early_stopping, MultipleObjectiveTracker, eval_state, cycle_datasets, Exhauster, LongCycler
+from neuralpredictors import measures
+from neuralpredictors.measures import *
+from neuralpredictors.training import early_stopping, MultipleObjectiveTracker, eval_state, cycle_datasets, Exhauster, LongCycler
 from ..utility.nn_helpers import set_random_seed
 
 from ..utility import metrics
@@ -32,7 +32,7 @@ def early_stop_trainer(model, seed, stop_function='corr_stop',
                 'exp_stop'
                 'poisson_stop'
             loss_function: has to be a string that gets evaluated with eval()
-                Loss functions that are built in at mlutils that can
+                Loss functions that are built in at neuralpredictors that can
                 be selected in the trainer config are:
                     'PoissonLoss'
                     'GammaLoss'

nnvision/legacy/nnfabrik/utility/metrics.py

@@ -4,8 +4,8 @@
 import torch
 from scipy import stats
 
-from mlutils.measures import PoissonLoss, corr
-from mlutils.training import eval_state
+from neuralpredictors.measures import PoissonLoss, corr
+from neuralpredictors.training import eval_state
 
 
 def model_predictions(loader, model, data_key, device):

nnvision/legacy/nnfabrik/utility/nn_helpers.py

@@ -1,7 +1,7 @@
 # helper functions concerning the ANN architecture
 
 import torch
-from mlutils.training import eval_state
+from neuralpredictors.training import eval_state
 import numpy as np
 import random