Add initial implementation of training and generation

generate.py

@@ -0,0 +1,10 @@
+#!/usr/bin/env python3
+import gpt_2_simple as gpt2
+
+sess = gpt2.start_tf_sess()
+gpt2.load_gpt2(sess)
+
+petition = gpt2.generate(sess, prefix="<|startofpetition|>", truncate="<|endofpetition|>", include_prefix=False, return_as_list=True)[0]
+petition = petition.replace("<|startofpetition|>", "") # Sometimes this gets duplicated
+
+print(petition)

train.py

@@ -0,0 +1,49 @@
+#!/usr/bin/env python3
+import gpt_2_simple as gpt2
+import os
+import requests
+import json
+
+TRAINING_STEPS = 1000
+MODEL_NAME = "124M"
+
+if not os.path.isdir(os.path.join("models", MODEL_NAME)):
+ print(f"Downloading {MODEL_NAME} model...")
+ gpt2.download_gpt2(model_name=MODEL_NAME)
+
+output_filename = "petitions.txt"
+
+output = open(output_filename, 'w')
+
+for dirname, subdirs, files in os.walk(os.path.join('uk_petitions_data', 'petitions')):
+ print(dirname)
+ for filename in files:
+ if filename[-5:] == ".json":
+ print(filename)
+ petition_json = json.load(open(os.path.join(dirname, filename), 'r'))
+ if 'data' in petition_json:
+ petition = petition_json['data']['attributes']
+ if 'attributes' in petition_json:
+ petition = petition_json['attributes']
+ else:
+ continue
+ output.write("<|startofpetition|>\n")
+ output.write(petition['action'] + "\n")
+ output.write("=================\n\n")
+ if 'background' in petition and petition['background'] is not None:
+ output.write(petition['background'].replace("\r", "") + "\n")
+ if 'additional_details' in petition and petition['additional_details'] is not None:
+ output.write(petition['additional_details'].replace("\r", "") + "\n")
+ if petition['rejected_at'] is not None and petition['rejection']['details'] is not None and petition['rejection']['details'][:8] != "https://":
+ output.write("\n\nReason for rejection:\n\n")
+ output.write(petition['rejection']['details'].replace("\r", "") + "\n")
+ output.write("<|endofpetition|>\n\n")
+
+output.close()
+
+sess = gpt2.start_tf_sess()
+gpt2.finetune(sess,
+ output_filename,
+ model_name=MODEL_NAME,
+ steps=TRAINING_STEPS,
+ save_every=100)