dnnbrain.dnn.base module

class dnnbrain.dnn.base.ImageProcessor

Bases: object

Metrics for pre-processing pictures to further DNN operations.

crop(image, box)

Crop image with a rectangular region.

Parameters

  • image (ndarray, Tensor, PIL.Image) – Image data.

  • box (tuple) – The crop rectangle as a (left, upper, right, lower)-tuple.

Returns

image – Image data after crop.

Return type

ndarray, Tensor, PIL.Image

norm(image, ord)

Calculate norms of the image by the following formula:

sum(abs(image)**ord)**(1./ord)

Parameters

  • image (ndarray, Tensor, PIL.Image) – Image data.

  • ord (int) – The order of the norm.

Returns

norm – The norm of the image.

Return type

float

resize(image, size, interpolation='nearest')

Resize image.

Parameters

  • image (ndarray, Tensor, PIL.Image) – Image data.

  • size (tuple) – The target size as a 2-tuple: (height, width).

  • interpolation (str) – Interpolation method for resize, check self.str2pil_interp and self.str2cv2_interp to find the available interpolation.

Returns

image – Image data after resizing.

Return type

ndarray, Tensor, PIL.Image

to_array(image)

Convert image to array.

Parameters

image (ndarray, Tensor, PIL.Image) – Image data.

Returns

arr – Image array.

Return type

ndarray

to_pil(image, normalization=False)

Convert image to PIL.Image

Parameters

  • image (ndarray, Tensor, PIL.Image) – Image data.

  • normalization (bool) – Normalization operation. If is True, normalize image data to integers in [0, 255].

Returns

image – Output image with type of PIL.Image.

Return type

PIL.Image

to_tensor(image)

Convert image to tensor

Parameters

image (ndarray, Tensor, PIL.Image) – Image data.

Returns

tensor – Image tensor.

Return type

Tensor

total_variation(image)

Calculate total variation of the image.

Parameters

image (ndarray, Tensor, PIL.Image) – Image data.

Returns

tv – Total variation.

Return type

float

translate(image, bkg, startpoint, endpoint, stride)

Translate image on a background based on given startpoint and stride, only support one axis now.

Parameters

  • image (ndarray, Tensor, PIL.Image) – Image data.

  • bkg (ndarray, Tensor, PIL.Image) – Same type with image.Remember bkg must bigger than image.

  • startpoint (tuple) – The start point of translating in upper left position as a (x_axis,y_axis)-tuple, horizontal:x_axis, vertical:y_axis.

  • endpoint (tuple) – The end point of translating in upper left position as a (x_axis,y_axis)-tuple (horizontal: x_axis; vertical: y_axis).

  • stride (int) – Stride of each translation.

Returns

image_tran – Image data after translating, which add a dim in the first axis meaning n_stim.

Return type

ndarray, Tensor, PIL.Image

class dnnbrain.dnn.base.ImageSet(img_dir, img_ids, labels=None, transform=None)

Bases: object

Build a dataset to load image.

class dnnbrain.dnn.base.MultivariateMapping(estimator=None, cv=5, scoring=None)

Bases: object

map(X, Y)

Use all columns of X to predict each column of Y.

Parameters

  • X (ndarray) – shape=(n_sample, n_feature)

  • Y (ndarray) – shape=(n_sample, n_target)

Returns

map_dict – It depends on estimator type.

estimator type

key

value

classifier

score

An array with shape as (n_target, cv).
Each row contains accuracies of each cross
validation folds, when using all features to
predict the corresponding target.

model

An array with shape as (n_target,).
Each element is a model fitted by all features
and the corresponding target.

conf_m

An array with shape as (n_target, cv).
Each row contains confusion matrices
(n_label, n_label) of each cross validation
folds, when using all features to predict the
corresponding target.

regressor

score

An array with shape as (n_target, cv).
Each row contains scores of each cross
validation folds, when using all features to
predict the corresponding target.

model

An array with shape as (n_target,).
Each element is a model fitted by all features
and the corresponding target.

Return type

dict

set_cv(cv)
Parameters

cv (int) – The number of cross validation folds.

set_estimator(estimator)
Parameters

estimator (str | sklearn estimator or pipeline) –

If is str, it is a name of a estimator used to do mapping.
There are some optional names at below:

name

type

description

lrc

classifier

Logistic Regression Classifier

svc

classifier

C-Support Vector Classification

glm

regressor

Ordinary least squares Linear Regression

lasso

regressor

Linear Model trained with L1 prior as regularizer

set_scoring(scoring)
Parameters

scoring (str or callable) –

The method to evaluate the predictions on the test set. It depends on estimator type.

estimator type

Scoring description

classifier

The evaluation method is fixed as accuracy and confusion matrix.

regressor

Scoring parameters or strategies supported by sklearn in addition to ‘correlation’.

class dnnbrain.dnn.base.UnivariateMapping(estimator=None, cv=5, scoring=None)

Bases: object

For each target samples, evaluate every feature samples’ predictive ability and record location, prediction score and model of the feature with maximal predictive ability.

map(X, Y)

Use all columns of X (one-by-one) to predict each column of Y;

For each column of Y: Find the location of the column of X which has the maximal prediction score; Record the location, and corresponding score and model.

Parameters

  • X (ndarray) – shape=(n_sample, n_feature)

  • Y (ndarray) – shape=(n_sample, n_target)

Returns

map_dict – It depends on estimator type.

estimator type

key

value

classifier

score

An array with shape as (n_target, cv).
Each row contains accuracies of each cross
validation folds, when using the feature at the
maximal location to predict the corresponding target.

location

An array with shape as (n_target,).
Each element is a location of the feature
which makes the maximal score.

model

An array with shape as (n_target,).
Each element is a model fitted by the feature
at the maximal location and the corresponding target.

conf_m

An array with shape as (n_target, cv).
Each row contains confusion matrices
(n_label, n_label) of each cross validation folds,
when using the feature at the maximal location to
predict the corresponding target.

regressor

score

An array with shape as (n_target, cv).
Each row contains scores of each cross
validation folds, when using the feature at
the maximal location to predict the corresponding target.

location

An array with shape as (n_target,).
Each element is a location of the feature
which makes the maximal score.

model

An array with shape as (n_target,).
Each element is a model fitted by the feature
at the maximal location and the corresponding target.

correlation

score

An array with shape as (n_target,).
Each element is the maximal pearson r among all
features correlating to the corresponding target.

location

An array with shape as (n_target,).
Each element is a location of the feature
which makes the maximal score.

Return type

dict

set_cv(cv)
Parameters

cv (int) – The number of cross validation folds.

set_estimator(estimator)
Parameters

estimator (str | sklearn estimator or pipeline) –

If is str, it is a name of a estimator used to do mapping.
There are some optional names at below:

name

type

description

lrc

classifier

Logistic Regression Classifier

svc

classifier

C-Support Vector Classification

glm

regressor

Ordinary least squares Linear Regression

lasso

regressor

Linear Model trained with L1 prior as regularizer

Note: If name is ‘corr’, it just uses correlation rather than prediction.

set_scoring(scoring)
Parameters

scoring (str or callable) –

The method to evaluate the predictions on the test set. It depends on estimator type.

estimator type

Scoring description

classifier

The evaluation method is fixed as accuracy and confusion matrix.

regressor

Scoring parameters or strategies supported by sklearn in addition to ‘correlation’.

correlation

No evaluation method is needed.

class dnnbrain.dnn.base.VideoClipSet(clip_files, transform=None)

Bases: object

class dnnbrain.dnn.base.VideoSet(vid_file, frame_nums, labels=None, transform=None)

Bases: object

Dataset for video data

dnnbrain.dnn.base.array_statistic(arr, method, axis=None, keepdims=False)

extract statistic of an array

Parameters

  • arr (ndarray) – A numpy array.

  • method (str) – Feature extraction method

  • axis (int, tuple) – None or int or tuple of ints. Axis or axes along which to operate. If it’s None, operate on the whole array.

  • keepdims (bool) – Keep the axis which is reduced.

Returns

arr – Extracted statistic.

Return type

ndarray

dnnbrain.dnn.base.correlation_score(y_true, y_pred, multioutput='uniform_average')
Parameters

  • y_true (ndarray with shape as (n_samples,) or (n_samples, n_outputs)) – Ground truth target values.

  • y_pred (ndarray with shape as (n_samples,) or (n_samples, n_outputs)) – Estimated target values.

  • multioutput (string in ['raw_values', 'uniform_average']) –

    ‘raw_values’ :

    Returns a full set of scores in case of multioutput input.

    ’uniform_average’ :

    Scores of all outputs are averaged with uniform weight.

Returns

score – a single value if ‘multioutput’ is ‘uniform_average’ a ndarray if ‘multioutput’ is ‘raw_values’

Return type

float or ndarray of floats

dnnbrain.dnn.base.correlation_scorer(regressor, X, y)
dnnbrain.dnn.base.cross_val_confusion(classifier, X, y, cv=None)

Evaluate confusion matrix and score from each fold of cross validation

Parameters

  • classifier (classifier object) – The object used to fit the data.

  • X (ndarray) – Shape=(n_sample, n_feature).

  • y (ndarray) – Shape=(n_sample,).

  • cv (int) – The number of folds of the cross validation.

Returns

  • conf_ms (list) – Confusion matrices of the folds.

  • accuracies (list) – Accuracies of the folds.

dnnbrain.dnn.base.cross_val_scores(regressor, X, Y, scoring, cv=None, multi_trg_flag=True)

Evaluate scores for each run of cross validation

Parameters

  • regressor (regressor object) – The estimator used to fit the data.

  • X (ndarray) – Shape=(n_sample, n_feature).

  • Y (ndarray) – Shape=(n_sample, n_target).

  • scoring (str | callable) –

    see scoring of Regression at https://scikit-learn.org/stable/modules/model_evaluation.html#common-cases-predefined-values If is str, choices=(explained_variance, r2, correlation). If is callable, the inputs and outputs should imitate scoring metrics in sklearn.

    The input parameters should include y_true, y_pred, and multioutput at least.

  • cv (int) – The number of runs of the cross validation.

  • multi_trg_flag (bool) – Whether the regressor is able to fit multi-targets at onces or not.

Returns

scores – shape=(cv, n_target)

Return type

ndarray

dnnbrain.dnn.base.dnn_fe(dnn_acts, method, n_feat, axis=None)

Extract features of DNN activation

Parameters

  • dnn_acts (ndarray) – DNN activation. A 4D array with its shape as (n_stim, n_chn, n_row, n_col).

  • method (str) –

    Feature extraction method, choices: (‘pca’, ‘hist’, ‘psd’)

    method name

    description

    pca

    use n_feat principal components as features

    hist

    use histogram of activation as features

    psd

    use power spectral density as features

    Note: In ‘hist’, n_feat equal-width bins in the given range will be used!

  • n_feat (int, float) – The number of features to extract. Note: It can be a float only when the method is pca.

  • axis (str) –

    axis for feature extraction, choices=(chn, row_col)

    axis

    array shape

    chn

    (n_stim, n_feat, n_row, n_col)

    row_col

    (n_stim, n_chn, n_feat, 1)

    None

    (n_stim, n_feat, 1, 1)

    Note: We always regard the shape of the result as (n_stim, n_chn, n_row, n_col)

Returns

dnn_acts_new – DNN activation. A 4D array with its shape as (n_stim, n_chn, n_row, n_col).

Return type

ndarray

dnnbrain.dnn.base.dnn_mask(dnn_acts, channels='all', rows='all', columns='all')

Extract DNN activation

Parameters

  • dnn_acts (ndarray) – DNN activation A 4D array with its shape as (n_stim, n_chn, n_row, n_col).

  • channels (str, list) – Channels of interest. If is str, it must be ‘all’ which means all channels. If is list, its elements are serial numbers of channels.

  • rows (str, list) – rows of interest. If is str, it must be ‘all’ which means all rows. If is list, its elements are serial numbers of rows.

  • columns (str, list) – Columns of interest. If is str, it must be ‘all’ which means all columns. If is list, its elements are serial numbers of columns.

Returns

dnn_acts – DNN activation after mask. A 4D array with its shape as (n_stim, n_chn, n_row, n_col).

Return type

ndarray

dnnbrain.dnn.base.gen_estimator_from_name(name)

Generate sklearn estimator from name

Parameters

name (str) – name of estimator

Returns

estimator

Return type

sklearn estimator

dnnbrain.dnn.base.normalize(array)

Normalize an array’s value domain to [0, 1] Note: the original normalize function is at dnnbrain/utils/util.py but ‘from dnnbrain.dnn.core import Mask’ in the file causes import conflicts. Fix the conflicts in future.

Parameters

array (ndarray) – A numpy array waiting normalize.

Returns

array – A numpy array after normalization.

Return type

ndarray