Add your own metrics¶

You nedd to blablabla ...

Add in checkatlas¶

Create a python file in checkatlas folder :

checkatlas/metrics

Example : checkatlas/metrics/annot/rand_index.py or checkatlas/metrics/cluster/silhouette.py

Add the following code for a clustering metric:

def run(count_repr, annotations):
    metric = silhouette_score(count_repr, annotations)
    return metric

Add the following code for a cell annotation metric:

def run(annotation, ref_annotation):
    metric = adjusted_rand_score(annotation, ref_annotation)
    return metric

Add the following code for a dimensionality reduction metric:

def run(high_dim_counts, low_dim_counts):
    low_dim_distances = euclidean_distances(low_dim_counts, low_dim_counts)
    high_dim_distances = euclidean_distances(high_dim_counts, high_dim_counts)
    metric = np.sqrt(
        np.square(high_dim_distances - low_dim_distances).sum()
        / np.square(low_dim_distances).sum()
    )
    return metric

Reference your metric in the init.py in the all variable. Example for checkatlas.metrics.cluster module:

all = ["silhouette", "davies_bouldin", "calinski_harabasz"]

It will be added automatically to the arguments of checkatlas workflow.

Arguments:

. --metric_cluster . --metric_annot . --metric_dimred

Documentation¶

Add a file my_metric.md in oone of the folder and the file : - cellannotation - cell_annotation.md - clustering - dimred