implemented greedy clustering

multitask/Cluster.ipynb

@@ -145,6 +145,28 @@
  " return clusters"
  ]
  },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "def greedy_clustering(df, bucket_size, clusters=[], symmetric=False):\n",
+ " if not symmetric:\n",
+ " mat = df.to_numpy()\n",
+ " df[:] = (mat+mat.T)/2\n",
+ " used = [t for c in clusters for t in c]\n",
+ " available = [t for t in df.columns if t not in used]\n",
+ " if len(available) <= bucket_size:\n",
+ " clusters.append(available)\n",
+ " return clusters\n",
+ " def score(bucket):\n",
+ " return sum([df.loc[a, b] for a, b in itertools.combinations(bucket, 2)])\n",
+ " _, bucket = min([(score(b), b) for b in itertools.combinations(available, bucket_size)])\n",
+ " clusters.append(list(bucket))\n",
+ " return greedy_clustering(df, bucket_size, clusters=clusters, symmetric=True)"
+ ]
+ },
  {
  "cell_type": "code",
  "execution_count": null,