Added script to convert Numberbatch to a Parquet file

convert_numberbatch_to_parquet.py

@@ -0,0 +1,23 @@
+# Converts the text file version of Numberbatch to a Parquet file
+# The resulting Parquet file can be loaded by Pandas significantly faster than
+# the text file or the HDF5 file 
+
+import pandas as pd
+
+if __name__=='__main__':
+    # Note that the first line (the line indicating the size of the matrix)
+    # MUST BE MANUALLY DELETED BEFORE RUNNING THIS SCRIPT
+    # in order to simplify loading into a DataFrame
+    print('Loading text file into DataFrame...')
+    df = pd.read_csv('./numberbatch-en-19.08.txt', sep=' ', header=None, index_col=0)
+    print('Done.')
+
+    # Changes to make to_parquet happy
+    df.columns = df.columns.astype(str) # Convert column names to string
+    df.index.name = 'concept'           # Provide an index name
+
+    # index=True because we need the indices (the concept which corresponds to the embedding)
+    print('Writing to Parquet file...')
+    df.to_parquet('./numberbatch-en-19.08.parquet', index=True)
+    print('Done.')
+