Borda

#!/usr/bin/env bash
# ─────────────────────────────────────────────────────────────────────────────
# archive-repos.sh
#
# Archive GitHub repos listed in one or more CSV files. Reads the first
# column (repo slug) of each CSV, deduplicates across files, and calls
# `gh repo archive --yes` on each.
#
# Defaults to DRY RUN — prints what would happen but doesn't touch anything.
# Set APPLY=1 to actually archive. Note: GitHub's API does NOT support

Borda

#!/usr/bin/env bash
# ─────────────────────────────────────────────────────────────────────────────
# list-inactive.sh
#
# Find repos in an org/user account with no push activity since a cutoff date,
# so they're candidates for archiving.
#
# Uses the pushedAt field from a single GitHub API call — fast (one request),
# no per-repo lookup. Computes days since last push and sorts the output
# from least stale (recent) to most stale (oldest).

Borda

#!/usr/bin/env bash
# ─────────────────────────────────────────────────────────────────────────────
# list-clean-forks.sh
#
# Find forks in an org/user account that never diverged from their upstream,
# so they're safe to archive or delete.
#
# For each non-archived fork, compares the fork's default branch against the
# upstream's default branch via the GitHub compare API. A fork is "clean" if
# ahead_by == 0 — i.e., zero commits on top of upstream.

Borda

import glob
import torch
import scipy
import random
import matplotlib.pyplot as plt
from PIL import Image
from torchvision import transforms
from train import LitClassification, ClassificationData

# load test association to the numerical labels

Borda

import torch
import torch.optim as optim
from torch.utils.data import DataLoader
from torchvision import models, datasets, ops
from torchvision.transforms import v2 as transforms
import pytorch_lightning as pl


# Step 1a: Define the transform
transform = transforms.Compose([

Borda

name: Update git submodules

on:
  pull_request:
    branches: ["main"]
    paths:
      - ".github/workflows/ci-update-submodules.yml"
  schedule:
    # on Sundays
    - cron: "0 0 * * 0"

Borda

name: Sample build
on:
  - push
  - pull_request
  - workflow_dispatch

jobs:
  build:
    runs-on: ubuntu-latest
    steps:

Borda

preds = []
# move model to GPU for faster inference and set evaluation
model.cuda().eval()
for imgs, names in dm.test_dataloader():
    # for the prediction we do not need gradients
    with torch.no_grad():
        onehots = model(imgs.cuda()).cpu()
    # aggregate particular preditions
    for oh, name in zip(onehots, names):
        lbs = dm.onehot_to_labels(oh)

Borda

from pytorch_lightning import Trainer
from kaggle_plantpatho.data import PlantPathologyDM
from kaggle_plantpatho.models import LitResnet, MultiPlantPathology

# create DataModule with training/validation split
dm = PlantPathologyDM(batch_size=98)
# initialize ResNet50 network
net = LitResnet(arch='resnet50', num_classes=dm.num_classes)
# initialize PL module with ResNet50
model = MultiPlantPathology(model=net, lr=6e-4)

Borda

fig = plt.figure(figsize=(3, 7))
for imgs, lbs in dm.val_dataloader():
    # some stats about the batch - label distribution
    print(f'batch labels: {torch.sum(lbs, axis=0)}')
    print(f'image size: {imgs[0].shape}')
    # similar as above show just first images from the batch
    for i in range(3):
        ax = fig.add_subplot(3, 1, i + 1, xticks=[], yticks=[])
        ax.imshow(np.rollaxis(imgs[i].numpy(), 0, 3))
        ax.set_title(lbs[i])