sketched setting

.gitignore

@@ -0,0 +1,3 @@
+/metrics
+.hyperopt
+.vscode

bin/git_merge_subfolder.py

@@ -0,0 +1,45 @@
+import argparse
+import subprocess
+
+class MyArg:
+  pass
+
+args = MyArg()
+args.metric = "score.txt"
+
+
+def check_output(*args, **kwargs):
+  return subprocess.check_output(*args, universal_newlines=True, **kwargs).strip()
+
+
+parser = argparse.ArgumentParser(description="Merge a complete subfolder by picking the best one according to the specified metric.")
+parser.add_argument("branch_subfolder")
+parser.add_argument("--min", dest='optimizer', action='store_const', const=min, help="minimize given metric")
+parser.add_argument("--max", dest='optimizer', action='store_const', const=max, help="maximize given metric")
+parser.add_argument("--metric")
+
+args = parser.parse_args()
+
+assert not args.branch_subfolder.endswith("/"), "assuming branch_subfolder is given without trailing slash"
+
+"""
+hyperopt/1:
+        score.txt: 0.9
+hyperopt/2:
+        score.txt: 0.73
+"""
+metrics = check_output(["dvc", "metrics", "show", "-a", args.metric]).splitlines()
+
+branchnames = metrics[0::2]  # every odd should be a branch name
+scores = [float(m[len(args.metric)+3:]) for m in metrics[1::2]]  # every even should be a metric value prepended by "metricname: "
+
+metrics_subset = [name, score for name, score in zip(branchnames, scores) if name.startswith(args.branch_subfolder)]
+best_branch = args.optimizer(metrics_subset, key=lambda t:t[1])[0]
+
+check_output(["git", "merge", "--no-ff", best_branch]).splitlines()
+
+branchnames_subset = [m[0] for m in metrics_subset]
+
+check_output(["git", "merge", "--strategy=ours", *branchnames_subset])
+
+

bin/git_push_set_upstream++.py

@@ -0,0 +1,66 @@
+import argparse
+import subprocess
+
+def check_output(*args, **kwargs):
+  return subprocess.check_output(*args, universal_newlines=True, **kwargs).strip()
+
+def git_current_branchname():
+  return check_output(["git", "rev-parse", "--abbrev-ref", "HEAD"])
+
+def git_tracking_branch():
+  return check_output(["git", "for-each-ref", "--format=%(upstream:short)", check_output(["git", "symbolic-ref", "-q", "HEAD"])])
+
+parser = argparse.ArgumentParser(description="hallo")
+parser.add_argument("branchname_without_incrementsuffix", nargs='?', default=None)
+parser.add_argument("--max-increments", type=int, default=3)
+args = parser.parse_args()
+
+if args.branchname_without_incrementsuffix is None:
+  args.branchname_without_incrementsuffix = git_current_branchname()
+
+
+def get_new_increment(branchname_without_incrementsuffix):
+  """ get suffix which does not exist yet, autoincrement if necessary """
+  result = check_output(["git", "ls-remote", "origin", f"refs/heads/{branchname_without_incrementsuffix}/*"]).splitlines()
+  print(result)
+  if not result:
+    return 1
+  else:
+    max_increment = 0
+    for line in result:
+      # the format is """commithash-tab-branchname"""
+      # hence we can simply compare from the right
+      row = line.split("/")
+      if row[-2] == branchname_without_incrementsuffix:
+        increment_string = row[-1]
+        try:
+          max_increment = max(max_increment, int(increment_string))
+        except ValueError:  # TODO parse errormessage to be sure
+          continue
+    return max_increment + 1
+
+
+def push_set_upstream_autoincrement(branchname_without_incrementsuffix, max_auto_increase=1):
+  # second call should not increment again, hence we raise an Error if we find a remote tracking branch
+  tracking_branch = git_tracking_branch()
+  if tracking_branch:
+    raise RuntimeError(f"The branch already tracks a remote branch ({tracking_branch}). "
+      "Either delete the reference in order to use `git_push_set_upstream++`, or fall back to use plain `git push`.")
+
+  increment = get_new_increment(branchname_without_incrementsuffix)
+
+  for i in range(max_auto_increase):
+    branchname_with_incrementsuffix = f"{branchname_without_incrementsuffix}/{increment + i}"
+    try:
+      command = ["git", "push", "--set-upstream", "origin", f"{branchname_without_incrementsuffix}:{branchname_with_incrementsuffix}"]
+      print(" ".join(command))
+      result = check_output(command)
+      print(" ".join(command))
+      return result
+    except subprocess.CalledProcessError:
+      continue
+
+  raise RuntimeError(f"Reached maximum number of auto-increase attempts {max_auto_increase}. Set command-line argument like `--max-increments 10` to increase the limit.")
+
+
+push_set_upstream_autoincrement(args.branchname_without_incrementsuffix, max_auto_increase=args.max_increments)

compute_score.py

@@ -0,0 +1,2 @@
+from random import random
+print(random())

hyperoptimization.py

@@ -0,0 +1,22 @@
+import subprocess
+import os
+from datetime import datetime
+from time import time
+def check_output(*args, **kwargs):
+  return subprocess.check_output(*args, universal_newlines=True, **kwargs).strip()
+
+origin = os.getcwd()
+
+# combine human-readable date suffix with utc timestamp in seconds
+hyperoptfolder = f'hyperopt_{datetime.now().strftime("%Y-%m-%d")}_{int(time())}'
+
+for i in range(2):
+  runfolder = f".hyperopt/{hyperoptfolder}/run{i+1}"
+  os.makedirs(runfolder)
+  os.chdir(runfolder)
+  check_output(["git", "clone", "../../../"])
+  check_output(["git", "checkout", "-b", hyperoptfolder])
+  check_output(["dvc", "repro", "metrics.dvc"])
+  check_output(["git", "commit", "-am", '"hyperopt run complete"'])
+  check_output(["python", "./bin/git_push_set_upstream++.py"])
+  os.chdir(origin)

metrics.dvc

@@ -0,0 +1,8 @@
+md5: 1149ef1e43c04e15d229f3366692f3dc
+cmd: python compute_score.py > metrics
+outs:
+- md5: 3529a08864f57bf05cf72d280c8dd415
+  path: metrics
+  cache: true
+  metric: true
+  persist: false