strings.py

@@ -1,8 +1,11 @@
-import json
+import time
+
+from pyspark.sql.types import StringType, ArrayType, StructType, StructField, DoubleType
+from threading import Thread
 from affro_cluster import *
-import os
+
 from pyspark.sql import SparkSession
-from pyspark.sql.functions import col
+from pyspark.sql.functions import col, explode, first, collect_list, udf, collect_set
 import sys
 
 spark = SparkSession.builder.appName("JSONProcessing").getOrCreate()
@@ -10,31 +13,85 @@ spark = SparkSession.builder.appName("JSONProcessing").getOrCreate()
 folder_path = sys.argv[1]
 hdfs_output_path = sys.argv[2]
 
-# folder_path = '/user/zeppelin/affiliations/raw_aff_string/2024-08'
-# hdfs_output_path = 'tmp/affro/results_strings'
+matchings_schema = ArrayType(
+    StructType([
+        StructField("Provenance", StringType(), nullable=True),
+        StructField("PID", StringType(), nullable=True),
+        StructField("Value", StringType(), nullable=True),
+        StructField("Confidence", DoubleType(), nullable=True),
+        StructField("Status", StringType(), nullable=True)
+    ])
+)
 
+operation_counter = spark.sparkContext.accumulator(0)
 
-def oalex_affro(record):
-    doi = record['doi'][16:]
-    oalex = record['rors']
-    try: 
-        matchings =  [item for sublist in [affro(x) for x in record['raw_aff_string']] for item in (sublist if isinstance(sublist, list) else [sublist])]  
+#Version of affro application on a single raw_aff_string and returns just the Matchins set
+def oalex_affro(aff_string):
+    global operation_counter
+    try:
+        matchings = affro(aff_string)
+        operation_counter += 1
+        # Ensure matchings is a list, even if affro returns a single dict
+        if not isinstance(matchings, list):
+            matchings = [matchings]
+
+        # Create the result as a tuple that matches matchings_schema
+        result = []
+        for matching in matchings:
+            # Assuming 'matching' is a dictionary that contains 'Provenance', 'PID', 'Value', 'Confidence', 'Status'
+            result.append((
+                matching.get("Provenance", None),
+                matching.get("PID", None),
+                matching.get("Value", None),
+                float(matching.get("Confidence", None)),
+                matching.get("Status", None)
+            ))
 
-        result = {'DOI' : doi, 'OAlex' : oalex, 'Matchings': matchings}
-      
         return result
-   
+
     except Exception as e:
-        print(f"Error processing record with id {record.get('doi')}: {str(e)}")
-        return None
+        print(f"Error processing affiliation string {aff_string}: {str(e)}")
+        return ()
 
+oalex_affro_udf = udf(oalex_affro, matchings_schema)
+monitor_done = False
 
-df = spark.read.json(folder_path)
-filtered_df = df.filter(col("doi").isNotNull())
-updated_rdd = filtered_df.rdd.map(lambda row: oalex_affro(row.asDict()))
+def monitor_counter(interval):
+    while True:
+        print(f"Number of calls to AffRo: {operation_counter.value}")
+        time.sleep(interval)
+        if monitor_done:
+            break
 
-json_rdd = updated_rdd.map(lambda record: json.dumps(record))
+exploded = spark.read.json(folder_path) \
+    .filter(col("doi").isNotNull()) \
+    .select(
+    col("doi").alias("DOI"),
+    col("rors").alias("OAlex"),
+    explode(col("raw_aff_string")).alias("affiliation") #this allows to split all the raw_aff_string and to parallelize better
+)
 
+monitor_thread = Thread(target=monitor_counter, args=(600,), daemon=True)
+monitor_thread.start()
 
-json_rdd.saveAsTextFile(hdfs_output_path)
+affs = exploded \
+    .select("affiliation") \
+    .distinct() \
+    .withColumn("Matchings", oalex_affro_udf(col("affiliation")))
 
+affs.join(exploded, on="affiliation") \
+    .select(col("DOI"),
+            col("OAlex"),
+            explode(col("Matchings")).alias("match")
+            ) \
+    .groupBy("DOI") \
+    .agg(first("OAlex").alias("OAlex"), #for each DOI it says what are the other columns Since OALEX is equal for each doi just select the first, while use the collect_list function to aggregate the Matchings
+         collect_set("match").alias("Matchings") #each exploded match is collected again
+         ) \
+    .write \
+    .mode("overwrite") \
+    .option("compression","gzip") \
+    .json(hdfs_output_path)
+
+monitor_done = True
+monitor_thread.join()