\u306f\u3058\u3081\u306b \u3053\u306e\u8a18\u4e8b\u3067\u306f\u3001\u81ea\u5206\u304c\u7d44\u3093\u3060LSTM\u306e\u4e88\u6e2c\u30e2\u30c7\u30eb\u3092\u4f7f\u3063\u3066\u5b9f\u969b\u306b\u682a\u5f0f\u6295\u8cc7\u3092\u3057\u3066\u307f\u308b\u3068\u3044\u3063\u305f\u8da3\u65e8\u306e\u5185\u5bb9\u3092\u7d39\u4ecb\u3057\u3066\u3044\u307e\u3059\u3002 \u524d\u56de\u306e\u8a18\u4e8b\u3092\u307e\u3060\u3054\u89a7\u306b\u306a\u3063\u3066\u3044\u306a\u3044\u65b9\u306f\u4ee5\u4e0b\u304b\u3089\u95b2\u89a7\u3067\u304d...<\/div><\/div><\/a><\/div>\n<\/div><\/figure>\n\n\n\n
\u3053\u306e\u8a18\u4e8b\u3067\u306f\u305d\u306e\u5185\u5bb9\u3092\u5171\u6709\u3057\u3088\u3046\u3068\u601d\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
<\/p>\n\n\n\n
\u6700\u521d\u306e\u4e00\u30f6\u6708\u9593\u306f\u304b\u306a\u308a\u52dd\u3064\u3053\u3068\u304c\u3067\u304d\u305f\u306e\u3067\u3059\u304c\u3001\u305d\u308c\u4ee5\u964d\u8ca0\u3051\u7d9a\u3051\u305f\u3053\u3068\u306b\u52a0\u3048\u3001\u53d6\u5f15\u3092\u5b8c\u5168\u81ea\u52d5\u5316\u3055\u305b\u308b\u3053\u3068\u304c\u96e3\u3057\u3044\u3068\u3044\u3046\u3053\u3068\u3067\u3001\u4e00\u65e6\u3053\u306e\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u81ea\u4f53\u3092\u30af\u30ed\u30fc\u30ba\u3059\u308b\u3053\u3068\u306b\u3057\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
\u6700\u7d42\u7684\u306b\u8ca0\u3051\u3066\u30af\u30ed\u30fc\u30ba\u3057\u305f\u30e2\u30c7\u30eb\u3067\u306f\u3042\u308a\u307e\u3059\u304c\u3001\u53c2\u8003\u306b\u3067\u304d\u308b\u90e8\u5206\u306f\u3042\u308b\u3068\u601d\u3046\u306e\u3067\u3053\u3046\u3057\u3066\u8a18\u4e8b\u306b\u6b8b\u305d\u3046\u3068\u601d\u3044\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
<\/p>\n\n\n\n
\u69cb\u6210<\/h2>\n\n\n\n
\u5148\u305a\u306f\u3001\u81ea\u5206\u304c\u69cb\u7bc9\u3057\u305f\u5168\u4f53\u50cf\u304b\u3089\u7d39\u4ecb\u3057\u3066\u3044\u3053\u3046\u3068\u601d\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
\u304b\u306a\u308a\u30b6\u30c3\u30af\u30ea\u3067\u3059\u304c\u5168\u4f53\u50cf\u3067\u3059\u3002<\/p>\n\n\n\n![\"\"]()
<\/figure>\n\n\n\n<\/p>\n\n\n\n
\u56f3\u3092\u898b\u308b\u3068\u308f\u304b\u308a\u307e\u3059\u304c\u3001\u7c21\u5358\u306a\u69cb\u6210\u306b\u306a\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
Heroku\u30b5\u30fc\u30d0\u3092\u5229\u7528\u3057\u3066\u3001\u65e5\u6bce\u306b\u7fcc\u65e5\u306e\u4e88\u6e2c\u7d50\u679c\u3092line\u901a\u77e5\u3068\u3057\u3066\u8fd4\u3059\u3088\u3046\u306a\u69cb\u6210\u306b\u306a\u3063\u3066\u3044\u307e\u3059\u3002\u4f7f\u7528\u8a00\u8a9e\u306fpython\u3092\u4f7f\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
\u30b3\u30fc\u30c9\u306e\u7ba1\u7406\u306fgit\u3067\u884c\u306a\u3063\u3066\u304a\u308a\u3001heroku\u30b5\u30fc\u30d0\u306b\u30d7\u30c3\u30b7\u30e5\u3059\u308b\u3053\u3068\u3067\u66f4\u65b0\u304c\u8d70\u308b\u3088\u3046\u306b\u306a\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
Line\u901a\u77e5\u306b\u98db\u3093\u3067\u304d\u305f\u60c5\u5831\u3092\u3082\u3068\u306b\u30ab\u30d6\u30b3\u30e0\u8a3c\u5238\u3067\u6307\u5b9a\u306e\u9298\u67c4\u3092\u8cfc\u5165\uff06\u58f2\u5374\u3057\u3066\u3044\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
\u3053\u306e\u8cfc\u5165\u3068\u58f2\u5374\u306e\u90e8\u5206\u306f\u624b\u52d5\u3067\u884c\u306a\u3063\u3066\u304a\u308a\u3001\u672c\u5f53\u306f\u3053\u3053\u3092\u81ea\u52d5\u5316\u3057\u305f\u304b\u3063\u305f\u306e\u3067\u3059\u304c\u3001\u30ab\u30d6\u30b3\u30e0\u8a3c\u5238\u306e\u500b\u4eba\u30a2\u30ab\u30a6\u30f3\u30c8\u3067\u306fAPI\u306e\u5236\u9650\u4e0a\u96e3\u3057\u305d\u3046\u3060\u3063\u305f\u306e\u3067\u65ad\u5ff5\u3057\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
<\/p>\n\n\n\n
<\/p>\n\n\n\n
\u5b9f\u88c5<\/h2>\n\n\n\n
\u5b9f\u88c5\u306f\u4ee5\u4e0b\u306e\uff13\u30d1\u30fc\u30c4\u306b\u5206\u304b\u308c\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
\u30fb\u524d\u65e5\u307e\u3067\u306e\u682a\u4fa1\u30c7\u30fc\u30bf\u53ce\u96c6<\/p>\n\n\n\n
\u30fbLSTM\u3067\u30e2\u30c7\u30eb\u3092\u69cb\u7bc9(pytorch)<\/p>\n\n\n\n
\u30fbline notify\u3067\u901a\u77e5\u3059\u308b<\/p>\n\n\n\n
\u682a\u4fa1\u306e\u53ce\u96c6\u306b\u306fyahoo\u306eAPI\u3092\u5229\u7528\u3057\u3066\u3044\u307e\u3059\u3002yahoo_finance_api2\u3067\u691c\u7d22\u3057\u3066\u3082\u3089\u3048\u308b\u3068\u4ed6\u306e\u65b9\u3005\u304c\u307e\u3068\u3081\u3066\u304f\u308c\u3066\u3044\u308b\u8a18\u4e8b\u304c\u305f\u304f\u3055\u3093\u30d2\u30c3\u30c8\u3059\u308b\u3068\u601d\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
\u3053\u3061\u3089\u3092\u5229\u7528\u3057\u3066\u3001\u682a\u5f0f\u60c5\u5831\u3092\u53ce\u96c6\u3057\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
<\/p>\n\n\n\n
API\u3092\u53e9\u3044\u3066\u3001\u682a\u5f0f\u306e\u30c7\u30fc\u30bf\u3092\u5168\u3066pandas\u306edataframe\u5f62\u5f0f\u306b\u8a70\u3081\u8fbc\u3093\u3067\u8fd4\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
from yahoo_finance_api2 import share\nfrom yahoo_finance_api2.exceptions import YahooFinanceError\n\ndef get_stock_info(code, N_day=180):\n my_share = share.Share(f'{code}.T')\n symbol_data = None\n try:\n symbol_data = my_share.get_historical(\n share.PERIOD_TYPE_DAY, N_day,\n share.FREQUENCY_TYPE_DAY, 1)\n except YahooFinanceError as e:\n sys.exit(1)\n \n df = pd.DataFrame(symbol_data)\n df["datetime"] = pd.to_datetime(df.timestamp, unit="ms")\n\n if df.isnull().values.sum() != 0:\n raise ValueError("Exception")\n\n if len(df) != row:\n raise ValueError("Exception")\n\n # \u53d6\u5f15\u91cf\u304c\u5c11\u306a\u3051\u308c\u3070\u524a\u9664(500\u4e07\u4ee5\u4e0b)\n if df['volume'].min()*df['low'].min() < 5000000:\n raise ValueError("Exception")\n \n #\u65e5\u672c\u6642\u9593\u3078\u5909\u63db\n df["datetime_JST"] = df["datetime"] + datetime.timedelta(hours=9)\n df = df.set_index('datetime_JST')\n df = df.drop(['datetime', 'timestamp'], axis=1)\n\n return df<\/code><\/pre><\/div>\n\n\n\n<\/p>\n\n\n\n
\u30e2\u30c7\u30eb\u69cb\u6210\u3068\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u69cb\u6210\u306f\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u69cb\u7bc9\u3057\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306b\u306fpytorch\u3092\u4f7f\u7528\u3057\u3066\u304a\u308a\u3001\u4e0a\u8a18\u306edataframe\u306b\u5bfe\u3057\u3066\u3001\u81ea\u4f5c\u306e\u30c7\u30fc\u30bf\u30d9\u30fc\u30b9\u30af\u30e9\u30b9\u3092\u4f5c\u6210\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
class LSTMPVClassifier(nn.Module):\n def __init__(self, embedding_dim, lstm_hidden_size, mlp_hidden_size, output_size, dropout):\n super(LSTMPVClassifier, self).__init__()\n self.lstm_hidden_size = lstm_hidden_size\n self.dropout = nn.Dropout(dropout)\n \n self.lstm = nn.LSTM(embedding_dim, self.lstm_hidden_size, batch_first=True, \n num_layers=1, bidirectional=False, dropout=0.0)\n \n self.fc1 = nn.Linear(self.lstm_hidden_size, mlp_hidden_size)\n self.fc2 = nn.Linear(mlp_hidden_size, output_size)\n \n def forward(self, x):\n b_size = x.size(0) # batch size\n seq_len = x.size(1) # sentence length\n\n # init hidden and cells of LSTM\n h0 = torch.zeros(1, b_size, self.lstm_hidden_size).to(device)\n c0 = torch.zeros(1, b_size, self.lstm_hidden_size).to(device)\n \n # execute LSTM\n lstm_output_seq, (h_n, c_n) = self.lstm(x, (h0, c0))\n \n out = self.dropout(self.fc1(lstm_output_seq))\n out = self.fc2(out)\n\n return out\n\n\n\nclass MyDataset(torch.utils.data.Dataset):\n def __init__(self, X, y, z):\n self.data = X\n self.teacher = y\n self.code = z\n\n def __len__(self):\n return len(self.teacher)\n\n def __getitem__(self, idx):\n out_data = self.data[idx]\n out_label = self.teacher[idx]\n out_code = self.code[idx]\n\n return out_data, out_label, out_code<\/code><\/pre><\/div>\n\n\n\ntrain_dataset = MyDataset(X_train, y_train, z_train)\ntrain_dataloader = DataLoader(train_dataset, batch_size=batch_size)\nval_dataset = MyDataset(X_val, y_val, z_val)\nval_dataloader = DataLoader(val_dataset, batch_size=batch_size)\n\ntest_dataset = MyDataset(X_test, y_val, z_val)\ntest_dataloader = DataLoader(test_dataset, batch_size=1)\n\nprint("train_size, val_size",len(X_train), len(X_val))<\/code><\/pre><\/div>\n\n\n\n\u5404\u7a2e\u30cf\u30a4\u30d1\u30fc\u30d1\u30e9\u30e1\u30fc\u30bf\u8a2d\u5b9a\u3068\u5b9f\u969b\u306e\u5b66\u7fd2\u904e\u7a0b\u304c\u3053\u3061\u3089\u3067\u3059\u3002<\/p>\n\n\n\n
# define network and hyperparameter for training\ndropout = 0.0\nlstm_hidden_size = 256\nemb_dim = X.shape[2]\nmlp_hidden = 64\noutput_size = 1\n\nnet = LSTMPVClassifier(emb_dim, lstm_hidden_size, mlp_hidden, \n output_size, dropout)\nnet = net.to(device)\n\ncriterion = nn.MSELoss()\noptimizer = torch.optim.Adam(filter(lambda p: p.requires_grad, net.parameters()))\n\n# training\nnum_epochs = 10\ntrain_loss_list = []\nval_loss_list = []\n\nfor epoch in range(num_epochs):\n train_loss = 0.0\n val_loss = 0.0\n train_acc = 0.0\n val_acc = 0.0\n train_y = []\n train_predicted = []\n\n val_y = []\n val_predicted = []\n\n #train\n net.train()\n for i, batch in enumerate(train_dataloader):\n input, labels, _ = batch\n\n input = input.to(device)\n labels = labels.to(device)\n f_labels = labels.float()\n\n optimizer.zero_grad()\n output = net(input)\n outputs = output[:, -1].squeeze()\n if f_labels.size()[0] == 1:\n continue\n\n loss = criterion(outputs, f_labels)\n train_loss += loss.item()\n loss.backward()\n optimizer.step()\n\n avg_train_loss = train_loss \/ len(train_dataloader.dataset)\n \n #val\n net.eval()\n with torch.no_grad():\n for batch in val_dataloader:\n input, labels, _ = batch\n input = input.to(device)\n labels = labels.to(device)\n\n f_labels = labels.float()\n outputs = net(input)\n outputs = outputs[:, -1].squeeze()\n\n loss = criterion(outputs, f_labels)\n val_loss += loss.item()\n \n avg_val_loss = val_loss \/ len(val_dataloader.dataset)\n print ('Epoch [{}\/{}], Loss: {loss:.4f}, val_loss: {val_loss:.4f}' \n .format(epoch+1, num_epochs, i+1, loss=avg_train_loss, val_loss=avg_val_loss,))<\/code><\/pre><\/div>\n\n\n\n<\/p>\n\n\n\n
\u6700\u5f8c\u306b\u7d50\u679c\u3092Line\u3067\u901a\u77e5\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
def send_line_notify(notification_message):\n line_notify_token = \u2018YOUR\u2019TOKEN\u2019\n line_notify_api = 'https:\/\/notify-api.line.me\/api\/notify'\n headers = {'Authorization': f'Bearer {line_notify_token}'}\n data = {'message': f'{notification_message}'}\n requests.post(line_notify_api, headers = headers, data = data)<\/code><\/pre><\/div>\n\n\n\n<\/p>\n\n\n\n
\u3088\u308a\u8a73\u7d30\u306a\u4f7f\u3044\u65b9\u306f\u4ee5\u4e0b\u306e\u8a18\u4e8b\u3067\u307e\u3068\u3081\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n\n
![\"[python]]()
<\/div>
Python\u3067\u81ea\u5206\u306eline\u306b\u30a8\u30e9\u30fc\u30e1\u30c3\u30bb\u30fc\u30b8\u3092\u9001\u4fe1\u3059\u308b\u3010\u5c0f\u30cd\u30bf\u3011<\/div>
2022.5.23<\/div>
\u682a\u4fa1\u4e88\u6e2c\u306a\u3069\u3001\u5b9a\u671f\u5b9f\u884c\u3057\u3066\u3044\u308b\u30b3\u30fc\u30c9\u304c\u3044\u304f\u3064\u304b\u3042\u308a\u308b\u304c\u3001\u30a8\u30e9\u30fc\u3092\u5410\u3044\u3066\u52d5\u4f5c\u3057\u306a\u304f\u306a\u3063\u3066\u3044\u305f\u308a\u3059\u308b\u5834\u5408\u304c\u3042\u308a\u307e\u3059\u3002 \u5e38\u306b\u30e2\u30cb\u30bf\u30ea\u30f3\u30b0\u3057\u3066\u3044\u308b\u308f\u3051\u3067\u306f\u306a\u3044\u306e\u3067\u3001\u30a8\u30e9\u30fc\u306b\u6c17\u3065\u304f\u307e\u3067\u306b\u306f\u30bf...<\/div><\/div><\/a><\/div>\n<\/div><\/figure>\n\n\n\n
<\/p>\n\n\n\n
<\/p>\n\n\n\n
\u304a\u308f\u308a\u306b<\/h2>\n\n\n\n
\u4e0a\u8a18\u306e\u3088\u3046\u306a\u30b3\u30fc\u30c9\u7528\u3044\u3067\u5b9f\u88c5\u3092\u884c\u3044\u307e\u3057\u305f\u3002\u30c8\u30fc\u30bf\u30eb\u8ca0\u3051\u3066\u3044\u308b\u69cb\u6210\u306a\u306e\u3067\u3001\u53c2\u8003\u7a0b\u5ea6\u306b\u3068\u3069\u3081\u3066\u304a\u3044\u3066\u9802\u3051\u305f\u3089\u3068\u601d\u3044\u307e\u3059!<\/p>\n","protected":false},"excerpt":{"rendered":"
\u306f\u3058\u3081\u306b \u6df1\u5c64\u5b66\u7fd2\u306a\u3069\u3092\u52c9\u5f37\u3057\u305f\u308a\u3001\u89e6\u3063\u3066\u3044\u305f\u308a\u3059\u308b\u3068\u3001\u7d50\u69cb\u306a\u4eba\u304c\u682a\u4fa1\u3084\u70ba\u66ff\u306a\u3069\u306b\u5229\u7528\u3067\u304d\u306a\u3044\u304b\u3068\u8003\u3048\u308b\u3082\u306e\u3067\u3059\u3088\u306d\u3002\u81ea\u5206\u3082\u305d\u308c\u306b\u6f0f\u308c\u305a\u3001\u81ea\u5206\u3067\u4f5c\u3063\u3066\u307f\u305f\u3044\u3068\u5e38\u3005\u601d\u3063\u3066\u3044\u307e\u3057\u305f\u3002 \u305d\u3057\u3066\u5b9f\u969b\u306b\u4f5c\u6210\u3057\u3001\u56db\u30f6\u6708\u9593\u306b\u308f\u305f\u308a\u6df1\u5c64\u2026<\/p>\n","protected":false},"author":1,"featured_media":495,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[10,25],"tags":[49,7,32],"_links":{"self":[{"href":"https:\/\/is-ai.jp\/index.php?rest_route=\/wp\/v2\/posts\/493"}],"collection":[{"href":"https:\/\/is-ai.jp\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/is-ai.jp\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/is-ai.jp\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/is-ai.jp\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=493"}],"version-history":[{"count":1,"href":"https:\/\/is-ai.jp\/index.php?rest_route=\/wp\/v2\/posts\/493\/revisions"}],"predecessor-version":[{"id":496,"href":"https:\/\/is-ai.jp\/index.php?rest_route=\/wp\/v2\/posts\/493\/revisions\/496"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/is-ai.jp\/index.php?rest_route=\/wp\/v2\/media\/495"}],"wp:attachment":[{"href":"https:\/\/is-ai.jp\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=493"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/is-ai.jp\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=493"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/is-ai.jp\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=493"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
![\"[python]](\"https:\/\/is-ai.jp\/wp-content\/uploads\/2022\/04\/22607921.jpg\")
<\/div>![\"\"](\"https:\/\/is-ai.jp\/wp-content\/uploads\/2022\/09\/full.drawio.png\")
<\/figure>\n\n\n\n![\"[python]](\"https:\/\/is-ai.jp\/wp-content\/uploads\/2022\/05\/44c23b6b15d70994d766716b66bcaf1c.png\")
<\/div>