Update README.md
Browse files
README.md
CHANGED
|
@@ -20,6 +20,10 @@ license: apache-2.0
|
|
| 20 |
|
| 21 |

|
| 22 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
```python
|
| 24 |
import os
|
| 25 |
import numpy as np
|
|
@@ -53,6 +57,7 @@ dset = load_dataset(..., split="train")
|
|
| 53 |
|
| 54 |
dset = dset.map(batch_tokenize, batched=True, batch_size=64, num_proc=28)
|
| 55 |
|
|
|
|
| 56 |
max_shard_size = convert_file_size_to_int('300MB')
|
| 57 |
dataset_nbytes = dset.data.nbytes
|
| 58 |
num_shards = int(dataset_nbytes / max_shard_size) + 1
|
|
@@ -63,6 +68,11 @@ for shard_index in tqdm(range(num_shards)):
|
|
| 63 |
shard = dset.shard(num_shards=num_shards, index=shard_index, contiguous=True)
|
| 64 |
shard.to_parquet(f"{dset_name}/tokenized/tokenized-{shard_index:03d}.parquet")
|
| 65 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 66 |
client = Client() # To keep track of dask computation
|
| 67 |
client
|
| 68 |
|
|
@@ -90,7 +100,11 @@ tsne = TSNE(
|
|
| 90 |
)
|
| 91 |
|
| 92 |
tsne_embedding = tsne.fit(X)
|
|
|
|
|
|
|
|
|
|
| 93 |
|
|
|
|
| 94 |
df = pd.DataFrame(data=tsne_embedding, columns=['x','y'])
|
| 95 |
agg = ds.Canvas(plot_height=600, plot_width=600).points(df, 'x', 'y')
|
| 96 |
img = ds.tf.shade(agg, cmap=cc.fire, how='eq_hist')
|
|
|
|
| 20 |
|
| 21 |

|
| 22 |
|
| 23 |
+
What follows is research code. It is by no means optimized for speed, efficiency, or readability.
|
| 24 |
+
|
| 25 |
+
## Data loading, tokenizing and sharding
|
| 26 |
+
|
| 27 |
```python
|
| 28 |
import os
|
| 29 |
import numpy as np
|
|
|
|
| 57 |
|
| 58 |
dset = dset.map(batch_tokenize, batched=True, batch_size=64, num_proc=28)
|
| 59 |
|
| 60 |
+
dset_name = "roots_subset"
|
| 61 |
max_shard_size = convert_file_size_to_int('300MB')
|
| 62 |
dataset_nbytes = dset.data.nbytes
|
| 63 |
num_shards = int(dataset_nbytes / max_shard_size) + 1
|
|
|
|
| 68 |
shard = dset.shard(num_shards=num_shards, index=shard_index, contiguous=True)
|
| 69 |
shard.to_parquet(f"{dset_name}/tokenized/tokenized-{shard_index:03d}.parquet")
|
| 70 |
|
| 71 |
+
```
|
| 72 |
+
|
| 73 |
+
## Embedding
|
| 74 |
+
|
| 75 |
+
```python
|
| 76 |
client = Client() # To keep track of dask computation
|
| 77 |
client
|
| 78 |
|
|
|
|
| 100 |
)
|
| 101 |
|
| 102 |
tsne_embedding = tsne.fit(X)
|
| 103 |
+
```
|
| 104 |
+
|
| 105 |
+
## Plotting
|
| 106 |
|
| 107 |
+
```python
|
| 108 |
df = pd.DataFrame(data=tsne_embedding, columns=['x','y'])
|
| 109 |
agg = ds.Canvas(plot_height=600, plot_width=600).points(df, 'x', 'y')
|
| 110 |
img = ds.tf.shade(agg, cmap=cc.fire, how='eq_hist')
|