rescue size_factor if median is 0

pinellolab · May 7, 2024 · 255e5f6 · 255e5f6
1 parent 509ad9e
commit 255e5f6
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 3 deletions.
diff --git a/bean/preprocessing/data_class.py b/bean/preprocessing/data_class.py
@@ -232,6 +232,8 @@ def get_size_factor(self, X: np.array):
         assert geom_mean_x.shape == (n_guides,)
         norm_count = X / geom_mean_x[:, None]
         size_factor = np.median(norm_count, axis=0)
+        if any(size_factor == 0):
+            size_factor = np.mean(norm_count, axis=0)
         assert size_factor.shape == (n_samples,)
         return size_factor
 

diff --git a/bean/preprocessing/get_alpha0.py b/bean/preprocessing/get_alpha0.py
@@ -89,7 +89,6 @@ def get_fitted_alpha0(
         sample_mask = torch.ones((n_reps, n_condits), device="cpu")
     elif (sample_mask.sum(axis=0) == 0).any():
         raise ValueError("Some bins have no data.")
-    print(sample_size_factors)
     w = get_w(X + 1, sample_size_factors, sample_mask=sample_mask)
     q = get_q(X + 1, sample_size_factors, sample_mask=sample_mask)
     n = (
@@ -107,8 +106,6 @@ def get_fitted_alpha0(
         print(
             f"Cannot fit log(a0) ~ log(q): data too sparse ({len(y)} valid values)! Using pre-fitted values [b0, b1]={popt}"
         )
-        print(n)
-        print(a0)
     else:
         popt, pcov = curve_fit(linear, x, y)
         print("Linear fit of log(a0) ~ log(q): [b0, b1]={}, cov={}".format(popt, pcov))