updated paddding from int to tuple

[adamklie]

See https://github.com/jmschrei/bpnet-lite/issues/4

Simple change of two lines to change padding of iconv and rconv from int type to tuple type.

Tested using tangermeme predict but that's about it.

[jmschrei]

Good catch. I think I saw this elsewhere too. When you say "tested" you mean you got the same predictions before/after?

[adamklie]

No sorry that was vague. I just meant it runs without error.

[jmschrei]

Would you mind checking that the predictions are the same for your model and posting the results here? I'd be happy to incorporate this change once confirmed.

[adamklie]

Generated 2056 random sequences and predicted counts and profile before and after the change. Predictions are the same and are here: https://drive.google.com/drive/folders/1pWyznsOnPwQU-uduAzts3bxax-KyTYk0?usp=sharing.

import matplotlib.pyplot as plt
import numpy as np

test_dir = "/cellar/users/aklie/data/datasets/mo_EndoC-bH1_ATAC-seq/test"

# Load em up
profile_preds_before = np.squeeze(np.load(f"{test_dir}/profile_preds_before.npz")['arr_0'], axis=1)
counts_preds_before = np.squeeze(np.load(f"{test_dir}/counts_preds_before.npz")['arr_0'])
profile_preds_after = np.squeeze(np.load(f"{test_dir}/profile_preds_after.npz")['arr_0'], axis=1)
counts_preds_after = np.squeeze(np.load(f"{test_dir}/counts_preds_after.npz")['arr_0'])

# Scatterplot of counts before and after
plt.scatter(counts_preds_before, counts_preds_after)
plt.xlabel("Counts before change")
plt.ylabel("Counts after change")
same = np.allclose(counts_preds_before, counts_preds_after, atol=1e-6)
print("Counts are the same:", same)  # this is true
plt.show()

# Plot profile for arbitrary sequence
plt.plot(profile_preds_before[10], color='blue', label='Before', alpha=0.5)
plt.plot(profile_preds_after[10], color='red', label='After', alpha=0.5)
plt.legend()
plt.xlabel("Position")
plt.ylabel("Profile")
same = np.allclose(profile_preds_before, profile_preds_after, atol=1e-6)
print("Profile is the same:", same)  # this is true
plt.show()

[jmschrei]

Only 2056? What if it's different on the 2057th sequence?

[adamklie]

I was gonna do all possible sequences, but I got bored after 2056