Perkzi commited on
Commit
464a7d8
·
1 Parent(s): 15aef6f
Files changed (2) hide show
  1. scripts/to_parquet.py +13 -5
  2. violin-test.parquet +2 -2
scripts/to_parquet.py CHANGED
@@ -1,19 +1,27 @@
 
1
  import pandas as pd
2
  from datasets import Dataset, Image
3
- import os
4
 
5
 
6
  df = pd.read_json("data/metadata.jsonl", lines=True)
7
 
8
 
9
- def full_path(path):
10
- return os.path.join(os.getcwd(),"data", path)
11
 
12
- df['image'] = df['file_name'].apply(full_path)
 
 
 
 
 
 
13
 
14
 
15
- ds = Dataset.from_pandas(df)
16
  ds = ds.cast_column("image", Image())
17
 
 
 
 
 
18
 
19
  ds.to_parquet("violin-test.parquet", batch_size=10000)
 
1
+ import os
2
  import pandas as pd
3
  from datasets import Dataset, Image
 
4
 
5
 
6
  df = pd.read_json("data/metadata.jsonl", lines=True)
7
 
8
 
9
+ df['image'] = df['file_name'].apply(lambda x: os.path.join(os.getcwd(), "data", x))
 
10
 
11
+ ds = Dataset.from_pandas(df)
12
+
13
+ # 3. Define the transformation and trigger the read.
14
+ def load_image_bytes(example):
15
+ # No operation is required here, but the map process will trigger encoding and decoding of the Image() type.
16
+ # Force the path to be converted to in-memory image data.
17
+ return {"image": example["image"]}
18
 
19
 
 
20
  ds = ds.cast_column("image", Image())
21
 
22
+ ds = ds.map(load_image_bytes, num_proc=1)
23
+
24
+
25
+ print(f"first entry: {ds[0]}")
26
 
27
  ds.to_parquet("violin-test.parquet", batch_size=10000)
violin-test.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2157037818cb9b4f0292113da2e93f8c66d0fc4ca4a6c0b74b9d65e74e761a5d
3
- size 1981815
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:399d0a9956f159293e8027e28d76e873df3e7091839c5cb94b1d54aee0f5bd99
3
+ size 1981175