Revert "Adding queries 25, 26 and 30 to be reviewed"

[VibhuJawa]

Reverts #241

We should revert the changes pushed in this PR as :

1. Q25 and Q26 changes are buggy and introduce the following error.

Traceback (most recent call last):
  File "/datasets/vjawa/miniconda3/envs/rapids-gpu-bdb-dask-sql-feb-11/lib/python3.8/site-packages/bdb_tools/utils.py", line 334, in run_sql_query
    results = benchmark(
  File "/datasets/vjawa/miniconda3/envs/rapids-gpu-bdb-dask-sql-feb-11/lib/python3.8/site-packages/bdb_tools/utils.py", line 57, in benchmark
    result = func(*args, **kwargs)
  File "gpu_bdb_query_25_dask_sql.py", line 188, in main
    results_dict = get_clusters(client=client, ml_input_df=cluster_input_ddf)
  File "gpu_bdb_query_25_dask_sql.py", line 58, in get_clusters
    pd.DataFrame(results_dict["cid_labels"]), npartitions=output.npartitions
  File "/datasets/vjawa/miniconda3/envs/rapids-gpu-bdb-dask-sql-feb-11/lib/python3.8/site-packages/pandas/core/frame.py", line 684, in __init__
    data = list(data)
  File "/home/nfs/vjawa/gpu_bdb_latest/dask-cuda/dask_cuda/proxy_object.py", line 560, in __iter__
    return iter(self._pxy_deserialize())
  File "/datasets/vjawa/miniconda3/envs/rapids-gpu-bdb-dask-sql-feb-11/lib/python3.8/site-packages/cudf/utils/utils.py", line 209, in __iter__
    raise TypeError(
TypeError: Series object is not iterable. Consider using `.to_arrow()`, `.to_pandas()` or `.values_host` if you wish to iterate over the values.

This is because the following should check for dask_cudf and not cudf.

gpu-bdb/gpu_bdb/queries/q25/gpu_bdb_query_25_dask_sql.py

Lines 52 to 55 in 9ae8a4d

	if isinstance(ml_input_df, cudf.DataFrame):
	labels_final = dask_cudf.from_cudf(
	results_dict["cid_labels"], npartitions=output.npartitions
	)

2. The ML component still uses cuML so uses GPUs so this not really a legit CPU implementation for these queries.

gpu-bdb/gpu_bdb/bdb_tools/utils.py

Lines 958 to 994 in 9ae8a4d

	def train_clustering_model(training_df, n_clusters, max_iter, n_init):
	"""Trains a KMeans clustering model on the
	given dataframe and returns the resulting
	labels and WSSSE"""
	from cuml.cluster.kmeans import KMeans
	best_sse = 0
	best_model = None
	# Optimizing by doing multiple seeding iterations.
	for i in range(n_init):
	model = KMeans(
	oversampling_factor=0,
	n_clusters=n_clusters,
	max_iter=max_iter,
	random_state=np.random.randint(0, 500),
	init="k-means++",
	)
	model.fit(training_df)
	score = model.inertia_
	if best_model is None:
	best_sse = score
	best_model = model
	elif abs(score) < abs(best_sse):
	best_sse = score
	best_model = model
	return {
	"cid_labels": best_model.labels_,
	"wssse": best_model.inertia_,
	"cluster_centers": best_model.cluster_centers_,
	"nclusters": n_clusters,
	}

CC: @DaceT , @ayushdg

[ayushdg]

lgtm! We should revisit enabling q30 separately if that doesn't break anything.

I'll open up another issue to discuss some sort of CI testing to catch these changes earlier