blythed

padl.build my_experiment/params.json

blythed

{
  "source": "config.py"
}

blythed

{
  "source": "config.py",
  "params": {
    "rnn": {
        "input_size": 64,
        "hidden_size": 512
    },
    "classifier": {
        "n_tokens": 16,
    }

blythed

my_model.padl
|__13.pt
|__14.pt
|__requirements.txt
|__transform.py

blythed

from padl import run
m = run('config.py').m
print(m)

blythed

Compose - "m":

      │
      ▼ text
   0: clean                                                  
      │
      ▼ args
   1: lower()                                                
      │
      ▼ x

blythed

from padl.transforms import Batchify
from padl.transforms import identity
from padl.transforms import Unbatchify
from padl import batch
import padl
from padl import params
from padl import unbatch
import re
import torch

blythed

import padl
from my_codebase.transforms import (
    clean, lower_case, tokenize, to_tensor, get_output, post_process_annotation,
)


def build_string_processor():
    return (
        clean
        >> lower_case

blythed

import padl
import re
import torch


@padl.transform
def clean(text):
    return re.sub('[^A-Za-z\\ ]', ' ', text)

blythed

def build_preprocessor(dictionary, default='<unk>'):
    
    UNK = dictionary[default]
    
    @padl.transform
    def pad(x):
        return list(x) + [UNK for _ in range(MIN_LEN - len(x))]
    
    return (
        clean