\nCompetition Goal: detect GW (Gravitational Wave)このコンテストでは、ブラックホール連星の合併によるGW信号の検出を目指します\n
","metadata":{"execution":{"iopub.status.busy":"2021-07-04T00:10:20.053695Z","iopub.execute_input":"2021-07-04T00:10:20.054235Z","iopub.status.idle":"2021-07-04T00:10:20.060253Z","shell.execute_reply.started":"2021-07-04T00:10:20.054187Z","shell.execute_reply":"2021-07-04T00:10:20.058883Z"}}},{"cell_type":"markdown","source":"### 😁よくわからないから中身をみていこう","metadata":{}},{"cell_type":"code","source":"!pip install -q nnAudio\n!pip install timm","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:28.195790Z","iopub.execute_input":"2021-08-31T12:09:28.196390Z","iopub.status.idle":"2021-08-31T12:09:42.220625Z","shell.execute_reply.started":"2021-08-31T12:09:28.196351Z","shell.execute_reply":"2021-08-31T12:09:42.219256Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import sys\nsys.path.append('../input/pytorch-image-models/pytorch-image-models-master')\n\nimport os\nimport math\nimport time\nimport random\nimport shutil\nfrom pathlib import Path\nfrom contextlib import contextmanager\nfrom collections import defaultdict, Counter\nfrom nnAudio.Spectrogram import CQT\nimport scipy as sp\nimport numpy as np\nimport pandas as pd\n\nfrom sklearn import preprocessing\nfrom sklearn.metrics import roc_auc_score\nfrom sklearn.model_selection import StratifiedKFold, GroupKFold, KFold\n\nfrom tqdm.auto import tqdm\nfrom functools import partial\n\nimport cv2\nfrom PIL import Image\n\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nfrom torch.optim import Adam, SGD\nimport torchvision.models as models\nfrom torch.nn.parameter import Parameter\nfrom torch.utils.data import DataLoader, Dataset\nfrom torch.optim.lr_scheduler import CosineAnnealingWarmRestarts, CosineAnnealingLR, ReduceLROnPlateau\n\nimport albumentations as A\nfrom albumentations.pytorch import ToTensorV2\nfrom albumentations import ImageOnlyTransform\n\nimport timm\n\nfrom torch.cuda.amp import autocast, GradScaler\n\nfrom nnAudio.Spectrogram import CQT1992v2\n\nimport warnings\nwarnings.filterwarnings('ignore')\nimport librosa\nfrom librosa.feature import melspectrogram\ndevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\nfrom pathlib import Path\nimport pandas as pd\nimport glob\nimport numpy as np\nimport matplotlib as mpl\nimport seaborn as sns\nimport random\nimport joblib\nfrom tqdm import tqdm_notebook as tqdm\nimport librosa\nimport librosa.display\nfrom scipy import signal\nfrom scipy.interpolate import interp1d\nfrom scipy.signal import butter, filtfilt, iirdesign, zpk2tf, freqz\nimport matplotlib.gridspec as gridspec\nimport matplotlib.pyplot as plt\nimport matplotlib.mlab as mlab\n%matplotlib inline\n\nimport warnings\nwarnings.filterwarnings('ignore')","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:42.223120Z","iopub.execute_input":"2021-08-31T12:09:42.223503Z","iopub.status.idle":"2021-08-31T12:09:42.249771Z","shell.execute_reply.started":"2021-08-31T12:09:42.223466Z","shell.execute_reply":"2021-08-31T12:09:42.248404Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# データ作成","metadata":{}},{"cell_type":"code","source":"#list = glob.glob('./data/*.csv')\n#data = pd.read_csv(list[0])\n#for i in tqdm(range(1,len(list))):\n# data = pd.concat([data,pd.read_csv(list[i])])\n#data.to_csv('g2net_train.csv')","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:42.251651Z","iopub.execute_input":"2021-08-31T12:09:42.252000Z","iopub.status.idle":"2021-08-31T12:09:42.269205Z","shell.execute_reply.started":"2021-08-31T12:09:42.251955Z","shell.execute_reply":"2021-08-31T12:09:42.268206Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#stop","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:42.270491Z","iopub.execute_input":"2021-08-31T12:09:42.270853Z","iopub.status.idle":"2021-08-31T12:09:42.292859Z","shell.execute_reply.started":"2021-08-31T12:09:42.270816Z","shell.execute_reply":"2021-08-31T12:09:42.290928Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## 📌training_labels.csv-関連する信号に重力波が含まれているかどうかの目標値","metadata":{}},{"cell_type":"code","source":"pd.read_csv('../input/g2net-gravitational-wave-detection/training_labels.csv')","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:42.294589Z","iopub.execute_input":"2021-08-31T12:09:42.295087Z","iopub.status.idle":"2021-08-31T12:09:42.701028Z","shell.execute_reply.started":"2021-08-31T12:09:42.295021Z","shell.execute_reply":"2021-08-31T12:09:42.699758Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### ファイル構成はこんな感じになっている。","metadata":{}},{"cell_type":"markdown","source":"","metadata":{},"attachments":{"9d705140-6e40-483b-a40e-fc7b16aa52bb.png":{"image/png":"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"}}},{"cell_type":"markdown","source":"## 📌train /-トレーニングセットファイル、観測ごとに1つのnpyファイル。 \n","metadata":{}},{"cell_type":"markdown","source":"### target=1でブラックホールあるパターン","metadata":{}},{"cell_type":"code","source":"input = np.load('../input/g2net-gravitational-wave-detection/train/0/0/0/00000e74ad.npy')\nfig, ax = plt.subplots(1,3,figsize=(15,5))\n\nax[0].hist(pd.DataFrame(input[0]))\nax[0].set_title(\"channel 0\")\nax[1].hist(pd.DataFrame(input[1]))\nax[1].set_title(\"channel 1\")\nax[2].hist(pd.DataFrame(input[2]))\nax[2].set_title(\"channel 2\")","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:42.702811Z","iopub.execute_input":"2021-08-31T12:09:42.703278Z","iopub.status.idle":"2021-08-31T12:09:43.177826Z","shell.execute_reply.started":"2021-08-31T12:09:42.703226Z","shell.execute_reply":"2021-08-31T12:09:43.176786Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### target=0でブラックホールないパターン","metadata":{}},{"cell_type":"code","source":"input = np.load('../input/g2net-gravitational-wave-detection/train/0/0/0/00001f4945.npy')\nfig, ax = plt.subplots(1,3,figsize=(15,5))\n\nax[0].hist(pd.DataFrame(input[0]))\nax[0].set_title(\"channel 0\")\nax[1].hist(pd.DataFrame(input[1]))\nax[0].set_title(\"channel 1\")\nax[2].hist(pd.DataFrame(input[2]))\nax[0].set_title(\"channel 2\")","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:43.181942Z","iopub.execute_input":"2021-08-31T12:09:43.182313Z","iopub.status.idle":"2021-08-31T12:09:43.628298Z","shell.execute_reply.started":"2021-08-31T12:09:43.182275Z","shell.execute_reply":"2021-08-31T12:09:43.627161Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### target=1でブラックホールあるパターン","metadata":{}},{"cell_type":"code","source":"input = np.load('../input/g2net-gravitational-wave-detection/train/0/0/0/00000e74ad.npy')\nsns.distplot(pd.DataFrame(input[0]))\nsns.distplot(pd.DataFrame(input[1]))\nsns.distplot(pd.DataFrame(input[2]))","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:43.630166Z","iopub.execute_input":"2021-08-31T12:09:43.630520Z","iopub.status.idle":"2021-08-31T12:09:44.073402Z","shell.execute_reply.started":"2021-08-31T12:09:43.630486Z","shell.execute_reply":"2021-08-31T12:09:44.072048Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"input = np.load('../input/g2net-gravitational-wave-detection/train/0/0/0/0000a38978.npy')\nsns.distplot(pd.DataFrame(input[0]))\nsns.distplot(pd.DataFrame(input[1]))\nsns.distplot(pd.DataFrame(input[2]))","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:44.075363Z","iopub.execute_input":"2021-08-31T12:09:44.075833Z","iopub.status.idle":"2021-08-31T12:09:44.501289Z","shell.execute_reply.started":"2021-08-31T12:09:44.075783Z","shell.execute_reply":"2021-08-31T12:09:44.500123Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### target=0でブラックホールないパターン","metadata":{}},{"cell_type":"code","source":"input = np.load('../input/g2net-gravitational-wave-detection/train/0/0/0/00001f4945.npy')\nsns.distplot(pd.DataFrame(input[0]))\nsns.distplot(pd.DataFrame(input[1]))\nsns.distplot(pd.DataFrame(input[2]))","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:44.502562Z","iopub.execute_input":"2021-08-31T12:09:44.502887Z","iopub.status.idle":"2021-08-31T12:09:44.942411Z","shell.execute_reply.started":"2021-08-31T12:09:44.502854Z","shell.execute_reply":"2021-08-31T12:09:44.941403Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"input = np.load('../input/g2net-gravitational-wave-detection/train/0/0/0/0000661522.npy')\nsns.distplot(pd.DataFrame(input[0]))\nsns.distplot(pd.DataFrame(input[1]))\nsns.distplot(pd.DataFrame(input[2]))","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:44.943899Z","iopub.execute_input":"2021-08-31T12:09:44.944231Z","iopub.status.idle":"2021-08-31T12:09:45.381777Z","shell.execute_reply.started":"2021-08-31T12:09:44.944196Z","shell.execute_reply":"2021-08-31T12:09:45.380798Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### なにか共通して違うというわけではなさそう","metadata":{}},{"cell_type":"code","source":"","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### なにを予測するのか","metadata":{}},{"cell_type":"code","source":"pd.read_csv('../input/g2net-gravitational-wave-detection/sample_submission.csv')","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:45.383097Z","iopub.execute_input":"2021-08-31T12:09:45.383418Z","iopub.status.idle":"2021-08-31T12:09:45.555141Z","shell.execute_reply.started":"2021-08-31T12:09:45.383386Z","shell.execute_reply":"2021-08-31T12:09:45.554210Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"id 0は、テストフォルダの１個目の名前ですね。","metadata":{}},{"cell_type":"markdown","source":"","metadata":{},"attachments":{"1517e3d4-fca5-4b14-b5be-6a6345a7586b.png":{"image/png":"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"}}},{"cell_type":"markdown","source":"提出物は、予測された確率と観察されたターゲットの間のROC曲線の下の領域で評価されます。とあります。ターゲットはこれを意識する必要があるようです。
\n","metadata":{}},{"cell_type":"code","source":"#list = glob.glob('../input/g2net-gravitational-wave-detection/train/*/*/*/*')\n#len(list)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:45.556462Z","iopub.execute_input":"2021-08-31T12:09:45.556807Z","iopub.status.idle":"2021-08-31T12:09:45.560192Z","shell.execute_reply.started":"2021-08-31T12:09:45.556765Z","shell.execute_reply":"2021-08-31T12:09:45.559344Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#list = sorted(list)\n#list[:10]","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:45.561380Z","iopub.execute_input":"2021-08-31T12:09:45.561667Z","iopub.status.idle":"2021-08-31T12:09:45.573661Z","shell.execute_reply.started":"2021-08-31T12:09:45.561638Z","shell.execute_reply":"2021-08-31T12:09:45.572458Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"https://www.kaggle.com/yasufuminakama/g2net-efficientnet-b7-baseline-training","metadata":{}},{"cell_type":"code","source":"train = pd.read_csv('../input/g2net-gravitational-wave-detection/training_labels.csv')\ntest = pd.read_csv('../input/g2net-gravitational-wave-detection/sample_submission.csv')\n\ndef get_train_file_path(image_id):\n return \"../input/g2net-gravitational-wave-detection/train/{}/{}/{}/{}.npy\".format(\n image_id[0], image_id[1], image_id[2], image_id)\n\ndef get_test_file_path(image_id):\n return \"../input/g2net-gravitational-wave-detection/test/{}/{}/{}/{}.npy\".format(\n image_id[0], image_id[1], image_id[2], image_id)\n\ntrain['file_path'] = train['id'].apply(get_train_file_path)\ntest['file_path'] = test['id'].apply(get_test_file_path)\nprint(train.shape)\ndisplay(train.head())\ndisplay(test.head())","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:45.575268Z","iopub.execute_input":"2021-08-31T12:09:45.575662Z","iopub.status.idle":"2021-08-31T12:09:47.029604Z","shell.execute_reply.started":"2021-08-31T12:09:45.575620Z","shell.execute_reply":"2021-08-31T12:09:47.028285Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def make_spectrogram(path, prints=False):\n '''Creates a MEL spectrogram.'''\n \n # Get the waves from the 3 sites\n waves = np.load(path).astype(np.float32)\n if prints:\n print(color.S+\"Waves Shape:\"+color.E, waves.shape)\n \n # Loop and make spectrogram\n spectrograms = []\n \n for i in range(3):\n # Compute a mel-scaled spectrogram.\n spec = melspectrogram(waves[i] / max(waves[i]), sr=4096, \n n_mels=128, fmin=20, fmax=2048)\n # Convert a power spectrogram (amplitude squared) to decibel (dB) units\n spec = librosa.power_to_db(spec).transpose((1, 0))\n spectrograms.append(spec)\n \n return spectrograms","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:47.030935Z","iopub.execute_input":"2021-08-31T12:09:47.031253Z","iopub.status.idle":"2021-08-31T12:09:47.039278Z","shell.execute_reply.started":"2021-08-31T12:09:47.031219Z","shell.execute_reply":"2021-08-31T12:09:47.038014Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"credit https://www.kaggle.com/andradaolteanu/g2net-searching-the-sky-pytorch-effnet-w-meta","metadata":{}},{"cell_type":"markdown","source":"
Melspectrogram","metadata":{}},{"cell_type":"markdown","source":"### I can't see the difference","metadata":{}},{"cell_type":"code","source":"# Samples per category\nn=3\n\n# Sample 6 paths with target and no target available\npaths_no_target = train[train[\"target\"] == 0][\"file_path\"].sample(n, random_state=23).values\npaths_with_target = train[train[\"target\"] == 1][\"file_path\"].sample(n, random_state=23).values\n\nall_paths = np.append(paths_no_target, paths_with_target)\n\n# Plot\nfig, axes = plt.subplots(nrows=2, ncols=n, figsize=(21,5))\nwandb_logs = []\n\n# Enumerate & plot\nfor i, path in enumerate(all_paths):\n if i < n: title = \"No Target\" \n else: title=\"With Target\"\n \n spec = make_spectrogram(path, prints=False)\n img = np.vstack(spec)\n \n x = i // n\n y = i % n\n \n axes[x, y].imshow(img, cmap=\"cool\")\n axes[x, y].set_title(title)\n axes[x, y].axis('off');\n \n \n \nplt.subplots_adjust(left=None, bottom=None, right=None, top=None, wspace=0.07, hspace=0.0)","metadata":{"_kg_hide-output":true,"execution":{"iopub.status.busy":"2021-08-31T12:09:47.040712Z","iopub.execute_input":"2021-08-31T12:09:47.041055Z","iopub.status.idle":"2021-08-31T12:09:47.893243Z","shell.execute_reply.started":"2021-08-31T12:09:47.041022Z","shell.execute_reply":"2021-08-31T12:09:47.892061Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def visualize_spectogram(\n id_, \n target):\n \n signal_names=(\"channel 0\", \"channel 1\", \"channel 2\")\n path = train[train.id==id_]\n path = str(path.file_path.values)[2:-2]\n x = np.load(path)\n plt.figure(figsize=(14, 4))\n for i in range(3):\n X = librosa.stft(x[i] / x[i].max())\n Xdb = librosa.amplitude_to_db(abs(X))\n plt.subplot(1, 3, i + 1)\n librosa.display.specshow(Xdb, sr=2048, x_axis=\"time\", y_axis=\"hz\", vmin=-30, vmax=50) \n plt.colorbar()\n plt.title(signal_names[i], fontsize=14)\n\n plt.suptitle(f\"id: {id_} target: {target}\", fontsize=16)\n plt.show()","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2021-08-31T12:09:47.895241Z","iopub.execute_input":"2021-08-31T12:09:47.895746Z","iopub.status.idle":"2021-08-31T12:09:47.906535Z","shell.execute_reply.started":"2021-08-31T12:09:47.895689Z","shell.execute_reply":"2021-08-31T12:09:47.905333Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"credit https://www.kaggle.com/ihelon/g2net-eda-and-modeling","metadata":{}},{"cell_type":"markdown","source":"
librosa.amplitude_to_db","metadata":{}},{"cell_type":"markdown","source":"### I can't see the difference","metadata":{}},{"cell_type":"markdown","source":"## Target =1","metadata":{}},{"cell_type":"code","source":" id_ = train.iloc[0][\"id\"]\n target = train.iloc[0][\"target\"]\n visualize_spectogram(id_, target)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:47.908328Z","iopub.execute_input":"2021-08-31T12:09:47.908666Z","iopub.status.idle":"2021-08-31T12:09:48.638528Z","shell.execute_reply.started":"2021-08-31T12:09:47.908633Z","shell.execute_reply":"2021-08-31T12:09:48.637319Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":" id_ = train.iloc[4][\"id\"]\n target = train.iloc[4][\"target\"]\n visualize_spectogram(id_, target)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:48.640509Z","iopub.execute_input":"2021-08-31T12:09:48.640980Z","iopub.status.idle":"2021-08-31T12:09:49.363420Z","shell.execute_reply.started":"2021-08-31T12:09:48.640928Z","shell.execute_reply":"2021-08-31T12:09:49.362297Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Target =0","metadata":{}},{"cell_type":"code","source":" id_ = train.iloc[1][\"id\"]\n target = train.iloc[1][\"target\"]\n visualize_spectogram(id_, target)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:49.365214Z","iopub.execute_input":"2021-08-31T12:09:49.365673Z","iopub.status.idle":"2021-08-31T12:09:50.080840Z","shell.execute_reply.started":"2021-08-31T12:09:49.365624Z","shell.execute_reply":"2021-08-31T12:09:50.079575Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":" id_ = train.iloc[2][\"id\"]\n target = train.iloc[2][\"target\"]\n visualize_spectogram(id_, target)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:09:50.085669Z","iopub.execute_input":"2021-08-31T12:09:50.086025Z","iopub.status.idle":"2021-08-31T12:09:50.799747Z","shell.execute_reply.started":"2021-08-31T12:09:50.085989Z","shell.execute_reply":"2021-08-31T12:09:50.798668Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{"_kg_hide-input":true,"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{"_kg_hide-input":true,"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"credit https://www.kaggle.com/allunia/signal-where-are-you","metadata":{}},{"cell_type":"markdown","source":"
signal.tukey","metadata":{}},{"cell_type":"code","source":"def bandpass(strain, fband, fs):\n \"\"\"Bandpasses strain data using a butterworth filter.\n \n Args:\n strain (ndarray): strain data to bandpass\n fband (ndarray): low and high-pass filter values to use\n fs (float): sample rate of data\n \n Returns:\n ndarray: array of bandpassed strain data\n \"\"\"\n bb, ab = butter(4, [fband[0]*2./fs, fband[1]*2./fs], btype='band')\n normalization = np.sqrt((fband[1]-fband[0])/(fs/2))\n strain_bp = filtfilt(bb, ab, strain) / normalization\n return strain_bp","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2021-08-31T12:09:50.801812Z","iopub.execute_input":"2021-08-31T12:09:50.802147Z","iopub.status.idle":"2021-08-31T12:09:50.809067Z","shell.execute_reply.started":"2021-08-31T12:09:50.802113Z","shell.execute_reply":"2021-08-31T12:09:50.807979Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def whiten(strain, interp_psd, dt, phase_shift=0, time_shift=0):\n \"\"\"Whitens strain data given the psd and sample rate, also applying a phase\n shift and time shift.\n Args:\n strain (ndarray): strain data\n interp_psd (interpolating function): function to take in freqs and output \n the average power at that freq \n dt (float): sample time interval of data\n phase_shift (float, optional): phase shift to apply to whitened data\n time_shift (float, optional): time shift to apply to whitened data (s)\n \n Returns:\n ndarray: array of whitened strain data\n \"\"\"\n Nt = len(strain)\n # take the fourier transform of the data\n freqs = np.fft.rfftfreq(Nt, dt)\n\n # whitening: transform to freq domain, divide by square root of psd, then\n # transform back, taking care to get normalization right.\n hf = np.fft.rfft(strain)\n \n # apply time and phase shift\n hf = hf * np.exp(-1.j * 2 * np.pi * time_shift * freqs - 1.j * phase_shift)\n norm = 1./np.sqrt(1./(dt*2))\n white_hf = hf / np.sqrt(interp_psd(freqs)) * norm\n white_ht = np.fft.irfft(white_hf, n=Nt)\n return white_ht","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2021-08-31T12:09:50.810616Z","iopub.execute_input":"2021-08-31T12:09:50.810955Z","iopub.status.idle":"2021-08-31T12:09:50.828139Z","shell.execute_reply.started":"2021-08-31T12:09:50.810919Z","shell.execute_reply":"2021-08-31T12:09:50.826719Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"time_span = 2 # 2 seconds\nsample_rate = 2048 # in Hz\nnum_samples = sample_rate*time_span\nhp_window = 1\nhp_tukey_alpha = 0.125\n#NFFT = 1*strain_len # why 16?\nNFFT = 4 * num_samples # Use 4 seconds of data for each fourier transform\nNOVL = 1 * NFFT / 2 # The number of points of overlap between segments used in Welch averaging\nfband = [15.0, 350.0]\ndt = 0.000244140625 # ?\nfig, ax = plt.subplots(5,2,figsize=(16,16))\n\nfor m in range(5):\n for k in range(2):\n melspecs = []\n for channel in [0,0,0]:#range(3):\n paths = train[train.target==k].reset_index(drop=True)\n path = paths.loc[m]\n example_strain = np.load(path.file_path)\n\n strain = example_strain[channel,:] / 2\n\n tukey_window = signal.tukey(num_samples*hp_window, hp_tukey_alpha)\n windowed_strain = strain*tukey_window\n\n psd_window = signal.tukey(NFFT, alpha=1./4)\n Pxx_strain, freqs = mlab.psd(windowed_strain, Fs = num_samples, NFFT = NFFT,\n window=psd_window, noverlap=NOVL)\n PSD = interp1d(freqs, Pxx_strain)\n \n strain_whitened = whiten(windowed_strain, \n PSD, dt)\n bandpassed_strain = bandpass(strain_whitened, fband, num_samples)\n ax[m,k].plot(bandpassed_strain)\n ax[m,k].set_ylim([-10,10])\n \n ax[m,k].set_title(\"Target {}\".format(k))","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2021-08-31T12:09:50.829600Z","iopub.execute_input":"2021-08-31T12:09:50.829923Z","iopub.status.idle":"2021-08-31T12:09:54.661380Z","shell.execute_reply.started":"2021-08-31T12:09:50.829891Z","shell.execute_reply":"2021-08-31T12:09:54.660072Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
QT","metadata":{}},{"cell_type":"markdown","source":"credit:https://www.kaggle.com/atamazian/nnaudio-constant-q-transform-demonstration","metadata":{}},{"cell_type":"code","source":"class G2NetDataset(Dataset):\n def __init__(self, paths, targets=None): \n self.paths = paths\n self.targets = targets\n\n def __len__(self):\n return len(self.paths)\n \n def __getitem__(self, index): \n signal = np.load(self.paths[index])\n signal = np.concatenate(signal, axis=0) # we concatenate data from 3 sensors to one signal\n signal = signal / np.max(signal)\n if self.targets is not None:\n targets = self.targets[index]\n return {\n \"signal\": torch.tensor(signal, dtype=torch.float),\n \"target\": torch.tensor(targets, dtype=torch.long),\n }\n else:\n return {\n 'signal': torch.tensor(signal, dtype=torch.float)\n }","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2021-08-31T12:09:54.663245Z","iopub.execute_input":"2021-08-31T12:09:54.663695Z","iopub.status.idle":"2021-08-31T12:09:54.675634Z","shell.execute_reply.started":"2021-08-31T12:09:54.663635Z","shell.execute_reply":"2021-08-31T12:09:54.674040Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ROOT_DIR = '../input/g2net-gravitational-wave-detection'\ndf = pd.read_csv(os.path.join(ROOT_DIR, 'training_labels.csv'))\ndf['path'] = df['id'].apply(lambda x: f'{ROOT_DIR}/train/{x[0]}/{x[1]}/{x[2]}/{x}.npy')","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2021-08-31T12:09:54.677370Z","iopub.execute_input":"2021-08-31T12:09:54.677860Z","iopub.status.idle":"2021-08-31T12:09:55.498737Z","shell.execute_reply.started":"2021-08-31T12:09:54.677809Z","shell.execute_reply":"2021-08-31T12:09:55.497599Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"transform = CQT(sr=2048, # sample rate\n fmin=20, # min freq\n fmax=1024, # max freq\n hop_length=64) # hop length\n\nds = G2NetDataset(df['path'], df['target'])\ncqts = []\nfor i in range(4):\n cqts.append(transform(ds.__getitem__(i)['signal']).squeeze())","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2021-08-31T12:09:55.500325Z","iopub.execute_input":"2021-08-31T12:09:55.500956Z","iopub.status.idle":"2021-08-31T12:09:55.603217Z","shell.execute_reply.started":"2021-08-31T12:09:55.500905Z","shell.execute_reply":"2021-08-31T12:09:55.601934Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"fig, axs = plt.subplots(4)\nfig.set_figheight(15)\nfig.set_figwidth(15)\ni=0\nnid = df['id'][i]\nntarget = df['target'][i]\naxs[i].title.set_text(f'{nid}.npy, target: {ntarget}')\naxs[i].pcolormesh(cqts[i])\n\ni=1\nnid = df['id'][i]\nntarget = df['target'][i]\naxs[i].title.set_text(f'{nid}.npy, target: {ntarget}')\naxs[i].pcolormesh(cqts[i])\n\ni=2\nnid = df['id'][i]\nntarget = df['target'][i]\naxs[i].title.set_text(f'{nid}.npy, target: {ntarget}')\naxs[i].pcolormesh(cqts[i])\n\ni=4\nnid = df['id'][i]\nntarget = df['target'][i]\naxs[3].title.set_text(f'{nid}.npy, target: {ntarget}')\naxs[3].pcolormesh(cqts[3])","metadata":{"_kg_hide-input":true,"execution":{"iopub.status.busy":"2021-08-31T12:09:55.604519Z","iopub.execute_input":"2021-08-31T12:09:55.604843Z","iopub.status.idle":"2021-08-31T12:09:56.213770Z","shell.execute_reply.started":"2021-08-31T12:09:55.604810Z","shell.execute_reply":"2021-08-31T12:09:56.212391Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
クーリエ変換 \nSTIMのコンペでは、コサイン波を除去することがポイントだったよう。今回も同じとはとても思えませんが、見てみる価値はあると思った。
\n見てみたがやはりわからない。","metadata":{}},{"cell_type":"markdown","source":"# target=1","metadata":{}},{"cell_type":"code","source":"import scipy.fftpack as sfft\ndata = np.load('../input/g2net-gravitational-wave-detection/train/0/0/0/00000e74ad.npy')\nh,w=data.shape \nfftsize=max(h,w) \nprint(fftsize) \nz=sfft.fftshift(sfft.fft2(data,(fftsize,fftsize)))#\nplt.plot(np.log(np.abs(z)))\nplt.show()\nprint(z.shape)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:10:55.303027Z","iopub.execute_input":"2021-08-31T12:10:55.303774Z","iopub.status.idle":"2021-08-31T12:11:05.605355Z","shell.execute_reply.started":"2021-08-31T12:10:55.303705Z","shell.execute_reply":"2021-08-31T12:11:05.604169Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"img =data.reshape(48,256)\nplt.imshow(img)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:11:05.606938Z","iopub.execute_input":"2021-08-31T12:11:05.607285Z","iopub.status.idle":"2021-08-31T12:11:05.749854Z","shell.execute_reply.started":"2021-08-31T12:11:05.607248Z","shell.execute_reply":"2021-08-31T12:11:05.748642Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"h,w=data.shape #h、wにGのサイズを代入\nfftsize=max(h,w) #フーリエ変換を行う際の行列のサイズを決定\nprint(fftsize)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:11:05.751348Z","iopub.execute_input":"2021-08-31T12:11:05.751690Z","iopub.status.idle":"2021-08-31T12:11:05.757762Z","shell.execute_reply.started":"2021-08-31T12:11:05.751649Z","shell.execute_reply":"2021-08-31T12:11:05.756587Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"z=sfft.fftshift(sfft.fft2(data,(fftsize,fftsize)))\nR=np.ones((4096,4096)) \nR[:1000,:]=0\nR[3000:,:]=0\n\nG2=np.uint8(np.abs(sfft.ifft2(sfft.fftshift(z*R))))\n#G2=G2.reshape(48,256)\nplt.imshow(G2[:48,:256])\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:12:27.176862Z","iopub.execute_input":"2021-08-31T12:12:27.177259Z","iopub.status.idle":"2021-08-31T12:12:29.505815Z","shell.execute_reply.started":"2021-08-31T12:12:27.177226Z","shell.execute_reply":"2021-08-31T12:12:29.504986Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data = np.load('../input/g2net-gravitational-wave-detection/train/0/0/0/0000a38978.npy')\nh,w=data.shape \nfftsize=max(h,w) \nprint(fftsize) \nz=sfft.fftshift(sfft.fft2(data,(fftsize,fftsize)))#\nplt.plot(np.log(np.abs(z)))\nplt.show()\nprint(z.shape)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:10:06.362522Z","iopub.execute_input":"2021-08-31T12:10:06.363011Z","iopub.status.idle":"2021-08-31T12:10:16.198895Z","shell.execute_reply.started":"2021-08-31T12:10:06.362970Z","shell.execute_reply":"2021-08-31T12:10:16.197454Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:10:38.709840Z","iopub.execute_input":"2021-08-31T12:10:38.710148Z","iopub.status.idle":"2021-08-31T12:10:40.972265Z","shell.execute_reply.started":"2021-08-31T12:10:38.710116Z","shell.execute_reply":"2021-08-31T12:10:40.971344Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# target=0","metadata":{}},{"cell_type":"code","source":"data = np.load('../input/g2net-gravitational-wave-detection/train/0/0/0/00001f4945.npy')\n\nh,w=data.shape \nfftsize=max(h,w) \nprint(fftsize) \nz=sfft.fftshift(sfft.fft2(data,(fftsize,fftsize)))#\nplt.plot(np.log(np.abs(z)))\nplt.show()\nprint(z.shape)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:10:16.201141Z","iopub.execute_input":"2021-08-31T12:10:16.202166Z","iopub.status.idle":"2021-08-31T12:10:26.568457Z","shell.execute_reply.started":"2021-08-31T12:10:16.202077Z","shell.execute_reply":"2021-08-31T12:10:26.567259Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data = np.load('../input/g2net-gravitational-wave-detection/train/0/0/0/0000661522.npy')\nh,w=data.shape \nfftsize=max(h,w) \nprint(fftsize) \nz=sfft.fftshift(sfft.fft2(data,(fftsize,fftsize)))#\nplt.plot(np.log(np.abs(z)))\nplt.show()\nprint(z.shape)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:10:26.570350Z","iopub.execute_input":"2021-08-31T12:10:26.570798Z","iopub.status.idle":"2021-08-31T12:10:36.335632Z","shell.execute_reply.started":"2021-08-31T12:10:26.570755Z","shell.execute_reply":"2021-08-31T12:10:36.334610Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"img =data.reshape(48,256)\nplt.imshow(img)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:10:36.337090Z","iopub.execute_input":"2021-08-31T12:10:36.337450Z","iopub.status.idle":"2021-08-31T12:10:36.483402Z","shell.execute_reply.started":"2021-08-31T12:10:36.337415Z","shell.execute_reply":"2021-08-31T12:10:36.480748Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"h,w=data.shape #h、wにGのサイズを代入\nfftsize=max(h,w) #フーリエ変換を行う際の行列のサイズを決定\nprint(fftsize)","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:10:36.485076Z","iopub.execute_input":"2021-08-31T12:10:36.485446Z","iopub.status.idle":"2021-08-31T12:10:36.496804Z","shell.execute_reply.started":"2021-08-31T12:10:36.485382Z","shell.execute_reply":"2021-08-31T12:10:36.495965Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"z=sfft.fftshift(sfft.fft2(data,(fftsize,fftsize)))\nR=np.ones((4096,4096)) \nR[:1000,:]=0\nR[3000:,:]=0\n\nG2=np.uint8(np.abs(sfft.ifft2(sfft.fftshift(z*R))))\n#G2=G2.reshape(48,256)\nplt.imshow(G2[:48,:256])\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2021-08-31T12:10:36.497988Z","iopub.execute_input":"2021-08-31T12:10:36.498465Z","iopub.status.idle":"2021-08-31T12:10:38.707716Z","shell.execute_reply.started":"2021-08-31T12:10:36.498417Z","shell.execute_reply":"2021-08-31T12:10:38.706567Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"㊙🔰🗑⬛🟥🟨🟩","metadata":{}}]}