Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

bug: fix accuracy metric and add test_with_y=True #41

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 20 additions & 12 deletions Finetune_GLUE.ipynb
Original file line number Diff line number Diff line change
Expand Up @@ -187,10 +187,11 @@
"metadata": {},
"outputs": [],
"source": [
"accuracy_metric = AvgMetric(lambda inp,targ: torch.eq(torch.tensor(inp.argmax(dim=-1)), torch.tensor(targ)).float().mean())\n",
"METRICS = {\n",
" **{ task:[MatthewsCorrCoef()] for task in ['cola']},\n",
" **{ task:[accuracy] for task in ['sst2', 'mnli', 'qnli', 'rte', 'wnli', 'snli','ax']},\n",
" **{ task:[F1Score(), accuracy] for task in ['mrpc', 'qqp']}, \n",
" **{ task:[accuracy_metric] for task in ['sst2', 'mnli', 'qnli', 'rte', 'wnli', 'snli','ax']},\n",
" **{ task:[F1Score(), accuracy_metric] for task in ['mrpc', 'qqp']}, \n",
" **{ task:[PearsonCorrCoef(), SpearmanCorrCoef()] for task in ['stsb']}\n",
"}\n",
"NUM_CLASS = {\n",
Expand Down Expand Up @@ -319,7 +320,7 @@
" glue_dsets[task]['train'] = datasets.concatenate_datasets([glue_dsets[task]['train'], swapped_train])\n",
"\n",
" # Load / Make dataloaders\n",
" hf_dsets = HF_Datasets(glue_dsets[task], hf_toker=hf_tokenizer, n_inp=3,\n",
" hf_dsets = HF_Datasets(glue_dsets[task], hf_toker=hf_tokenizer, n_inp=3, test_with_y=True,\n",
" cols={'inp_ids':TensorText, 'attn_mask':noop, 'token_type_ids':noop, 'label':TensorCategory})\n",
" if c.double_unordered and task in ['mrpc', 'stsb']:\n",
" dl_kwargs = {'train': {'cache_name': f\"double_dl_{c.max_length}_train.json\"}}\n",
Expand All @@ -342,7 +343,7 @@
" glue_dsets['wnli'] = wsc.my_map(partial(wsc_trick_process, hf_toker=hf_tokenizer),\n",
" cache_file_names=\"tricked_{split}.arrow\")\n",
" cols={'prefix':TensorText,'suffix':TensorText,'cands':TensorText,'cand_lens':noop,'label':TensorCategory}\n",
" glue_dls['wnli'] = HF_Datasets(glue_dsets['wnli'], hf_toker=hf_tokenizer, n_inp=4, \n",
" glue_dls['wnli'] = HF_Datasets(glue_dsets['wnli'], hf_toker=hf_tokenizer, n_inp=4, test_with_y=True,\n",
" cols=cols).dataloaders(bs=32, cache_name=\"dl_tricked_{split}.json\")"
]
},
Expand Down Expand Up @@ -718,17 +719,24 @@
},
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": "wnli\nSome weights of the model checkpoint at google/electra-large-discriminator were not used when initializing ElectraForPreTraining: ['electra.embeddings_project.weight', 'electra.embeddings_project.bias']\n- This IS expected if you are initializing ElectraForPreTraining from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPretraining model).\n- This IS NOT expected if you are initializing ElectraForPreTraining from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n"
"output_type": "stream",
"text": [
"wnli\n",
"Some weights of the model checkpoint at google/electra-large-discriminator were not used when initializing ElectraForPreTraining: ['electra.embeddings_project.weight', 'electra.embeddings_project.bias']\n",
"- This IS expected if you are initializing ElectraForPreTraining from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPretraining model).\n",
"- This IS NOT expected if you are initializing ElectraForPreTraining from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n"
]
},
{
"output_type": "display_data",
"data": {
"text/plain": "<IPython.core.display.HTML object>",
"text/html": ""
"text/html": [],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {}
"metadata": {},
"output_type": "display_data"
}
],
"source": [
Expand Down Expand Up @@ -778,7 +786,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.7.7-final"
"version": "3.10.11"
},
"widgets": {
"application/vnd.jupyter.widget-state+json": {
Expand Down Expand Up @@ -1277,4 +1285,4 @@
},
"nbformat": 4,
"nbformat_minor": 1
}
}
9 changes: 5 additions & 4 deletions finetune.py
Original file line number Diff line number Diff line change
Expand Up @@ -149,10 +149,11 @@ def after_fit(self):


# %%
accuracy_metric = AvgMetric(lambda inp,targ: torch.eq(torch.tensor(inp.argmax(dim=-1)), torch.tensor(targ)).float().mean())
METRICS = {
**{ task:[MatthewsCorrCoef()] for task in ['cola']},
**{ task:[accuracy] for task in ['sst2', 'mnli', 'qnli', 'rte', 'wnli', 'snli','ax']},
**{ task:[F1Score(), accuracy] for task in ['mrpc', 'qqp']},
**{ task:[accuracy_metric] for task in ['sst2', 'mnli', 'qnli', 'rte', 'wnli', 'snli','ax']},
**{ task:[F1Score(), accuracy_metric] for task in ['mrpc', 'qqp']},
**{ task:[PearsonCorrCoef(), SpearmanCorrCoef()] for task in ['stsb']}
}
NUM_CLASS = {
Expand Down Expand Up @@ -229,7 +230,7 @@ def tokenize_sents_max_len(example, cols, max_len, swap=False):
glue_dsets[task]['train'] = datasets.concatenate_datasets([glue_dsets[task]['train'], swapped_train])

# Load / Make dataloaders
hf_dsets = HF_Datasets(glue_dsets[task], hf_toker=hf_tokenizer, n_inp=3,
hf_dsets = HF_Datasets(glue_dsets[task], hf_toker=hf_tokenizer, n_inp=3, test_with_y=True,
cols={'inp_ids':TensorText, 'attn_mask':noop, 'token_type_ids':noop, 'label':TensorCategory})
if c.double_unordered and task in ['mrpc', 'stsb']:
dl_kwargs = {'train': {'cache_name': f"double_dl_{c.max_length}_train.json"}}
Expand All @@ -245,7 +246,7 @@ def tokenize_sents_max_len(example, cols, max_len, swap=False):
glue_dsets['wnli'] = wsc.my_map(partial(wsc_trick_process, hf_toker=hf_tokenizer),
cache_file_names="tricked_{split}.arrow")
cols={'prefix':TensorText,'suffix':TensorText,'cands':TensorText,'cand_lens':noop,'label':TensorCategory}
glue_dls['wnli'] = HF_Datasets(glue_dsets['wnli'], hf_toker=hf_tokenizer, n_inp=4,
glue_dls['wnli'] = HF_Datasets(glue_dsets['wnli'], hf_toker=hf_tokenizer, n_inp=4, test_with_y=True,
cols=cols).dataloaders(bs=32, cache_name="dl_tricked_{split}.json")

# %% [markdown]
Expand Down