Add minhash deduplicator based on RAY and Redis

configs/config_all.yaml

@@ -643,6 +643,17 @@ process:
       redis_port: 6380                                      # the port of redis instance, please note that the default port of redis is 6379 which is the same as default port for ray, so we need to modify the default redis config to use it in other port
       lowercase: false                                        # whether to convert text to lower case
       ignore_non_character: false                             # whether to ignore non-alphabet characters, including whitespaces, digits, and punctuations
+  - ray_redis_minhash_deduplicator:                           # the document deduplicator that can run on multi-nodes using minhashLSH algorithm
+      redis_address: 'redis://localhost:6379'                # the address of the redis instance
+      tokenization: space                                     # tokenization method for text. One of [space, punctuation, character, sentencepiece]
+      window_size: 5                                          # window size of shingling
+      num_permutations: 256                                   # number of permutations in minhash computing
+      jaccard_threshold: 0.7                                  # the min jaccard similarity threshold in near-duplicate detection. When the jaccard similarity of two sample texts is >= this threshold, they are regarded as similar samples and this op will only keep one of them after deduplication
+      num_bands: null                                         # number of bands in LSH. Default it's None, and it will be determined by an optimal params computation algorithm by minimize the weighted sum of probs of False Positives and False Negatives
+      num_rows_per_band: null                                 # number of rows in each band in LSH. Default it's None, and it will be determined by an optimal params computation algorithm
+      lowercase: true                                         # whether to convert text to lower case
+      ignore_pattern: null                                    # whether to ignore sub-strings with specific pattern when computing simhash.
+      tokenizer_model: null                                   # path for the sentencepiece model, used for sentencepiece tokenization.
 
   # Selector ops
   - frequency_specified_field_selector:                     # selector to select samples based on the sorted frequency of specified field value

data_juicer/core/ray_data.py

@@ -5,7 +5,7 @@
 
 from data_juicer import cuda_device_count
 from data_juicer.core.data import DJDataset
-from data_juicer.ops import Filter, Mapper
+from data_juicer.ops import Deduplicator, Filter, Mapper
 from data_juicer.utils.constant import Fields
 from data_juicer.utils.lazy_loader import LazyLoader
 from data_juicer.utils.process_utils import calculate_np
@@ -123,6 +123,8 @@ def _run_single_op(self, op):
                     self.data.write_json(op.stats_export_path,
                                          force_ascii=False)
                 self.data = self.data.filter(op.process)
+            elif isinstance(op, Deduplicator):
+                self.data = op.run(self.data)
             else:
                 logger.error(
                     'Ray executor only support Filter and Mapper OPs for now')

data_juicer/ops/deduplicator/__init__.py

@@ -5,12 +5,13 @@
 from .ray_basic_deduplicator import RayBasicDeduplicator
 from .ray_document_deduplicator import RayDocumentDeduplicator
 from .ray_image_deduplicator import RayImageDeduplicator
+from .ray_redis_minhash_deduplicator import RayRedisMinhashDeduplicator
 from .ray_video_deduplicator import RayVideoDeduplicator
 from .video_deduplicator import VideoDeduplicator
 
 __all__ = [
     'DocumentDeduplicator', 'DocumentMinhashDeduplicator',
     'DocumentSimhashDeduplicator', 'ImageDeduplicator', 'RayBasicDeduplicator',
     'RayDocumentDeduplicator', 'RayImageDeduplicator', 'RayVideoDeduplicator',
-    'VideoDeduplicator'
+    'RayRedisMinhashDeduplicator', 'VideoDeduplicator'
 ]