Upload train_t5_en_simple.py with huggingface_hub

train_t5_en_simple.py

@@ -0,0 +1,176 @@
+import pandas as pd
+import pickle as pkl
+import numpy as np
+
+def read_file_to_lines(path):
+    with open(path, "r") as f:
+        return pd.Series(f.readlines()).map(lambda x: x.replace("\n", ""))
+
+peil_en = read_file_to_lines("../peil_ja en-US.txt")
+
+with open("peil_ja.pkl", "rb") as f:
+    peil = pkl.load(f)
+
+assert len(peil) == len(peil_en)
+d = dict(zip(peil, peil_en))
+
+'''
+NEED_PREFIX = '次の出来事に必要な前提条件は何ですか: '
+EFFECT_PREFIX = '次の出来事の後に起こりうることは何ですか: '
+INTENT_PREFIX = '次の出来事が起こった動機は何ですか: '
+REACT_PREFIX = '次の出来事の後に感じることは何ですか: '
+
+NEED_PREFIX = '以下事件有哪些必要的先决条件：'
+EFFECT_PREFIX = '下面的事件发生后可能会发生什么：'
+INTENT_PREFIX = '以下事件的动机是什么：'
+REACT_PREFIX = '以下事件发生后，你有什么感觉：'
+'''
+
+NEED_PREFIX = 'What are the necessary preconditions for the next event?'
+EFFECT_PREFIX = 'What could happen after the next event?'
+INTENT_PREFIX = 'What is the motivation for the next event?'
+REACT_PREFIX = 'What are your feelings after the following event?'
+
+with open("graph_ja_zh_df.pkl", "rb") as f:
+    df = pkl.load(f)
+
+df["en_event"] = df["event"].map(lambda x: d[x])
+
+def one_infe_ele_map(inf_d):
+    req = {}
+    for k, v in inf_d.items():
+        req[k] = {}
+        for kk, vv in v.items():
+            assert type(vv) == type([])
+            req[k][kk] = list(map(lambda x: d[x], vv))
+    return req
+
+df["en_inference"] = df["inference"].map(one_infe_ele_map)
+
+with open("graph_ja_zh_en_df.pkl", "wb") as f:
+    pkl.dump(df, f)
+
+from huggingface_hub import HfApi
+api = HfApi()
+
+api.upload_file(
+    path_or_fileobj="graph_ja_zh_en_df.pkl",
+    path_in_repo="graph_ja_zh_en_df.pkl",
+    repo_id="svjack/comet-atomic-ja-zh",
+    repo_type="dataset",
+)
+
+examples = {
+    "event": df["en_event"].values.tolist(),
+    "inference": df["en_inference"].values.tolist(),
+}
+
+def preprocess_function_chunked(examples):
+    inputs = []
+    targets = []
+
+    for head_text, inf_type_dict in zip(examples['event'], examples['inference']):
+        for inf_type, inf_dirs in inf_type_dict.items():
+            if inf_dirs is None:
+                continue
+
+            for inf_dir, tail_texts in inf_dirs.items():
+                if tail_texts is None:
+                    continue
+
+                if inf_type == 'event':
+                    if inf_dir == 'before':
+                        prefix = NEED_PREFIX
+                    else:
+                        prefix = EFFECT_PREFIX
+                else:
+                    if inf_dir == 'before':
+                        prefix = INTENT_PREFIX
+                    else:
+                        prefix = REACT_PREFIX
+
+                for tail_text in tail_texts:
+                    inputs.append(prefix + head_text)
+                    targets.append(tail_text)
+    return inputs, targets
+
+src_tgt_df = pd.DataFrame(list(zip(*preprocess_function_chunked(examples))))
+src_tgt_df.columns = ["source_text", "target_text"]
+src_tgt_df.dropna().drop_duplicates().to_csv("comet-atomic-ja-en-simple.csv", index = False)
+
+from huggingface_hub import HfApi
+api = HfApi()
+
+api.upload_file(
+    path_or_fileobj="comet-atomic-ja-en-simple.csv",
+    path_in_repo="comet-atomic-ja-en-simple.csv",
+    repo_id="svjack/comet-atomic-ja-zh",
+    repo_type="dataset",
+)
+
+df = src_tgt_df.dropna().drop_duplicates().sample(frac = 1.0)
+df.iloc[:35300].to_csv("train_en.csv", index = False)
+df.iloc[35300:].to_csv("valid_en.csv", index = False)
+
+api.upload_file(
+    path_or_fileobj="train_en.csv",
+    path_in_repo="train_en.csv",
+    repo_id="svjack/comet-atomic-ja-zh",
+    repo_type="dataset",
+)
+
+api.upload_file(
+    path_or_fileobj="valid_en.csv",
+    path_in_repo="valid_en.csv",
+    repo_id="svjack/comet-atomic-ja-zh",
+    repo_type="dataset",
+)
+
+!tar -zcvf tt_en.tar.gz train_en.csv valid_en.csv
+
+from simplet5 import SimpleT5
+import pandas as pd
+model = SimpleT5()
+model.from_pretrained(model_type="t5",
+model_name="google/flan-t5-base",)
+
+train_df = pd.read_csv("train_en.csv")
+valid_df = pd.read_csv("valid_en.csv")
+
+train_df = train_df.dropna()
+valid_df = valid_df.dropna()
+
+model.train(train_df=train_df,
+            eval_df=valid_df,
+            source_max_token_len=128,
+            target_max_token_len=128,
+            batch_size=4, max_epochs=10, use_gpu=True)
+
+'''
+%matplotlib inline
+from time import sleep
+import matplotlib.pyplot as plt
+from IPython import display
+while True:
+    from tensorboard.backend.event_processing import event_accumulator
+    import pandas as pd
+    path = "lightning_logs/version_0/events.out.tfevents.1672423867.featurize.22967.0"
+    path = "lightning_logs/version_0/events.out.tfevents.1675325602.featurize.19734.0"
+    ea = event_accumulator.EventAccumulator(
+            path,
+            size_guidance={event_accumulator.SCALARS: 0},
+        )
+    _absorb_print = ea.Reload()
+    ea.Tags()
+    #pd.DataFrame(ea.Scalars("train_loss_step"))["value"].rolling(1000).mean().plot()
+    s = pd.DataFrame(ea.Scalars("train_loss_step"))["value"].rolling(100).mean()
+    plt.plot(s)
+    display.display(plt.gcf())
+    display.clear_output(wait=True)
+    sleep(5)
+'''
+
+model = SimpleT5()
+model.load_model(
+model_dir = "outputs/simplet5-epoch-0-train-loss-0.1042-val-loss-0.0814",
+use_gpu = False)