Download tests/test_third_party_dataset.py from Tinghong-Ye22/code-HFD: direct link, hf CLI and curl.
- Browser
- Download file 4.49 kB
-
https://huggingface.co/Tinghong-Ye22/code-HFD/resolve/main/tests/test_third_party_dataset.py
- Command line
-
hf download hf://Tinghong-Ye22/code-HFD/tests/test_third_party_dataset.py
-
curl -L -o test_third_party_dataset.py https://huggingface.co/Tinghong-Ye22/code-HFD/resolve/main/tests/test_third_party_dataset.py
4.49 kB
| import copy | |
| import json | |
| import random | |
| import wave | |
| from pathlib import Path | |
| import numpy as np | |
| import pytest | |
| from psi.data.seed.third_party_dataset import contained, file_hash, replace_microphone | |
| from psi.data.seed.live_multimodal_overlay import continuous_audio, interaction_task | |
| def fixture(tmp_path): | |
| pcm = np.arange(16000, dtype=np.int16) | |
| audio = tmp_path / 'input.wav' | |
| with wave.open(str(audio), 'wb') as f: | |
| f.setnchannels(1); f.setsampwidth(2); f.setframerate(16000) | |
| f.writeframes(pcm.tobytes()) | |
| steps = [dict(step_index=i, states=np.zeros(36), control_label=1, | |
| update_semantic_cache=i in (0,20), semantic_window_index=i//20) | |
| for i in range(21)] | |
| item = dict(steps=steps, microphone_speech_intervals=[dict(start_seconds=.2,end_seconds=.4)], | |
| interaction_mode='turn-taking') | |
| path = tmp_path/'timeline.npz' | |
| np.savez(path, states=np.zeros((21,36)), control_label=np.ones(21,dtype=np.uint8)) | |
| recipe = dict(session_id='example', session_type='mixed_short', background_language='en', | |
| interference_intervals=[dict(start_seconds=.1,end_seconds=.7)], | |
| timeline=dict(path=path.name,file_sha256=file_hash(path),action_steps=21), | |
| audio=dict(path=audio.name,sha256=file_hash(audio),sample_rate=16000,channels=1, | |
| sample_width_bytes=2,window_samples=9600,stride_samples=6400, | |
| window_count=2,sample_count=16000), | |
| supervision_targets=[], target_speech_events=[dict(start_sample=3200,end_sample=6400)]) | |
| return item,recipe,pcm | |
| def test_exact_pcm_overlap_and_current_pose_preserved(tmp_path): | |
| item,recipe,pcm=fixture(tmp_path) | |
| item['steps'][0]['states'][15]=.8 # Latest raised-arm pose need not match old archive. | |
| actual=replace_microphone(item,recipe,tmp_path) | |
| np.testing.assert_array_equal(continuous_audio(actual),pcm.astype(np.float32)/32768.) | |
| assert actual['steps'][0]['states'][15]==.8 | |
| assert interaction_task(actual)=='third-party' | |
| assert actual['interaction_mode']=='turn-taking' | |
| assert 'audio_waveform' not in actual['steps'][1] | |
| def test_reject_timing_drift(tmp_path): | |
| item,recipe,_=fixture(tmp_path) | |
| item['steps'][3]['control_label']=0 | |
| with pytest.raises(ValueError,match='timing/supervision drift'): | |
| replace_microphone(item,recipe,tmp_path) | |
| def test_reject_changed_speech_and_audio(tmp_path): | |
| item,recipe,_=fixture(tmp_path) | |
| item['microphone_speech_intervals'][0]['start_seconds']=.3 | |
| with pytest.raises(ValueError,match='speech shifted'): | |
| replace_microphone(item,recipe,tmp_path) | |
| item['microphone_speech_intervals'][0]['start_seconds']=.2 | |
| recipe['audio']['sha256']='bad' | |
| with pytest.raises(ValueError,match='audio corrupted'): | |
| replace_microphone(item,recipe,tmp_path) | |
| def test_reject_escape(tmp_path): | |
| with pytest.raises(ValueError,match='escapes'): | |
| contained(tmp_path,'../outside.wav') | |
| def test_appended_source_keeps_rng_and_base_membership(monkeypatch,tmp_path): | |
| import psi.data.seed.third_party_dataset as module | |
| recipe=dict(session_id='x',session_type='single_short',timeline_seed=123, | |
| composition_variant_id=4,composition_salt=5, | |
| turns=[dict(row_id=1,motion_filename='motion',audio_dir='audio',voice_name='voice', | |
| motion_dir=str(tmp_path),motion_sha256='unused')], | |
| timeline=dict(action_steps=31)) | |
| class Raw: | |
| records=[dict(row_id=1,filename='motion',audio_dir='audio',motion_dir=str(tmp_path))] | |
| sessions=((records[0],),) | |
| def __len__(self):return 1 | |
| def __getitem__(self,index):return {'original':index} | |
| def estimated_stream_steps(self,index):return 10 | |
| def _build_source(self,index):return {'random':random.random()} | |
| monkeypatch.setattr(module,'read_recipes',lambda *a:(recipe,)) | |
| monkeypatch.setattr(module,'replace_microphone',lambda item,*a:item) | |
| monkeypatch.setattr(module,'verify_primitive',lambda *a:None) | |
| for name in ('joint_pos.csv','body_pos.csv','body_quat.csv'): | |
| (tmp_path/name).touch() | |
| raw=Raw(); ds=module.ThirdPartyAugmentedDataset(raw,tmp_path,'unused','train') | |
| state=random.getstate() | |
| assert ds[1]==ds[1] | |
| assert random.getstate()==state | |
| assert ds[0]==raw[0] | |
| assert len(raw.sessions)==1 and len(ds)==2 | |
| assert ds.estimated_stream_steps(1)==31 | |