{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import os\nimport pytorch_lightning as pl\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torch.optim as optim\nfrom torch.utils.data import Dataset, DataLoader\nimport pandas as pd\nimport numpy as np\nimport polars\nfrom sklearn.model_selection import train_test_split","metadata":{"papermill":{"duration":9.057793,"end_time":"2023-04-18T11:42:58.161541","exception":false,"start_time":"2023-04-18T11:42:49.103748","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-04-20T00:16:01.244962Z","iopub.execute_input":"2023-04-20T00:16:01.246123Z","iopub.status.idle":"2023-04-20T00:16:10.402087Z","shell.execute_reply.started":"2023-04-20T00:16:01.246062Z","shell.execute_reply":"2023-04-20T00:16:10.400969Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def normalize_targets(y):\n    y_min = np.array([0.0, 0.0], dtype=np.float32)  # azimuth: [0, 2*pi], zenith: [0, pi]\n    y_max = np.array([2 * np.pi, np.pi], dtype=np.float32)\n    return (y - y_min) / (y_max - y_min)\n\ndef denormalize_targets(y_norm):\n    y_min = np.array([0.0, 0.0], dtype=np.float32)\n    y_max = np.array([2 * np.pi, np.pi], dtype=np.float32)\n    return y_norm * (y_max - y_min) + y_min\n\ndef normalize_features(df, features):\n    df = df.copy()\n    cols_min = df[features].min()\n    cols_max = df[features].max()\n    df[features] = (df[features] - cols_min) / (cols_max - cols_min)\n    return df","metadata":{"papermill":{"duration":0.019445,"end_time":"2023-04-18T11:42:58.187123","exception":false,"start_time":"2023-04-18T11:42:58.167678","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-04-20T00:16:10.404374Z","iopub.execute_input":"2023-04-20T00:16:10.404759Z","iopub.status.idle":"2023-04-20T00:16:10.413882Z","shell.execute_reply.started":"2023-04-20T00:16:10.404717Z","shell.execute_reply":"2023-04-20T00:16:10.412986Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def angular_dist_score(az_true, zen_true, az_pred, zen_pred):\n    '''\n    calculate the MAE of the angular distance between two directions.\n    The two vectors are first converted to cartesian unit vectors,\n    and then their scalar product is computed, which is equal to\n    the cosine of the angle between the two vectors. The inverse \n    cosine (arccos) thereof is then the angle between the two input vectors\n    \n    Parameters:\n    -----------\n    \n    az_true : float (or array thereof)\n        true azimuth value(s) in radian\n    zen_true : float (or array thereof)\n        true zenith value(s) in radian\n    az_pred : float (or array thereof)\n        predicted azimuth value(s) in radian\n    zen_pred : float (or array thereof)\n        predicted zenith value(s) in radian\n    \n    Returns:\n    --------\n    \n    dist : float\n        mean over the angular distance(s) in radian\n    '''\n    \n    if not (np.all(np.isfinite(az_true)) and\n            np.all(np.isfinite(zen_true)) and\n            np.all(np.isfinite(az_pred)) and\n            np.all(np.isfinite(zen_pred))):\n        raise ValueError(\"All arguments must be finite\")\n    \n    # pre-compute all sine and cosine values\n    sa1 = np.sin(az_true)\n    ca1 = np.cos(az_true)\n    sz1 = np.sin(zen_true)\n    cz1 = np.cos(zen_true)\n    \n    sa2 = np.sin(az_pred)\n    ca2 = np.cos(az_pred)\n    sz2 = np.sin(zen_pred)\n    cz2 = np.cos(zen_pred)\n    \n    # scalar product of the two cartesian vectors (x = sz*ca, y = sz*sa, z = cz)\n    scalar_prod = sz1*sz2*(ca1*ca2 + sa1*sa2) + (cz1*cz2)\n    \n    # scalar product of two unit vectors is always between -1 and 1, this is against nummerical instability\n    # that might otherwise occure from the finite precision of the sine and cosine functions\n    scalar_prod =  np.clip(scalar_prod, -1, 1)\n    \n    # convert back to an angle (in radian)\n    return np.average(np.abs(np.arccos(scalar_prod)))","metadata":{"papermill":{"duration":0.023044,"end_time":"2023-04-18T11:42:58.216084","exception":false,"start_time":"2023-04-18T11:42:58.193040","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-04-20T00:16:10.415439Z","iopub.execute_input":"2023-04-20T00:16:10.415975Z","iopub.status.idle":"2023-04-20T00:16:10.429655Z","shell.execute_reply.started":"2023-04-20T00:16:10.415940Z","shell.execute_reply":"2023-04-20T00:16:10.428435Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def generating_some_features(batch_file, metadata, sensor_geometry):\n    \n    \n    def join_tables_all(df_meta, df_batch, df_sensor):\n        return df_meta.join(df_batch, on='event_id').join(df_sensor, on='sensor_id').with_columns([\n            (polars.col('time') - polars.col('time').min()).over('event_id')\n        ])\n\n    def generate_features_grouped(dataf):\n        return dataf.groupby('event_id').agg([\n        polars.col('x').mean().alias('x_mean'),\n        polars.col('x').median().alias('x_median'),\n        polars.col('y').mean().alias('y_mean'),\n        polars.col('y').median().alias('y_median'),\n        polars.col('z').mean().alias('z_mean'),\n        polars.col('z').median().alias('z_median'),    \n        polars.col('time').mean().alias('event_mean_time'),\n        polars.col('time').max().alias('event_max_time'),\n        polars.col('charge').min().alias('event_min_charge'),\n        polars.col('charge').mean().alias('event_mean_charge'),\n        polars.col('charge').max().alias('event_max_charge'),\n        polars.col('charge').count().alias('overall_count'),\n        polars.col('auxiliary').sum().alias('overall_aux_sum'),\n        polars.col('charge').sum().alias('sum_charge'),\n        (polars.col('auxiliary').sum() / polars.col('auxiliary').count()).alias('aux_ratio'),\n        polars.col('sensor_id').n_unique().alias('sensor_count'),\n    ])\n    \n    def add_ranks(dataf):\n        return dataf.with_columns(\n[\n    polars.col('time').rank('ordinal').over('event_id').alias('time_rank_asc'),\n    polars.col('time').rank('ordinal', descending=True).over('event_id').alias('time_rank_des'),\n    polars.col('charge').rank('ordinal').over('event_id').alias('charge_rank_asc'),\n    polars.col('charge').rank('ordinal').over('event_id').alias('charge_rank_des')\n])\n\n    def make_geometrical_features(dataf):\n        geometrical_features = dataf.select('event_id').unique()\n        for direction in ['time_rank_asc','time_rank_des', 'charge_rank_asc', 'charge_rank_des']:\n            for direction_axis in ['x', 'y', 'z']:    \n                temp_col_1 = dataf.filter(polars.col(direction) == 1).select([\n                    polars.col('event_id'),\n                    polars.col(direction_axis).over('event_id')\n                ]).with_columns([\n                    polars.col(direction_axis).alias(direction_axis+'_'+direction+'_1')]\n                ).select(polars.col('event_id'), polars.col(direction_axis+'_'+direction+'_1'))\n\n                temp_col_2 = dataf.filter(polars.col(\"time_rank_asc\") == 2).select([\n                    polars.col('event_id'),\n                    polars.col(direction_axis).over('event_id')\n                ]).with_columns([\n                    polars.col(direction_axis).alias(direction_axis+'_'+direction+'_2')]\n                ).select(polars.col('event_id'), polars.col(direction_axis+'_'+direction+'_2'))\n\n                temp_col_3 = dataf.filter(polars.col(\"time_rank_asc\") == 3).select([\n                    polars.col('event_id'),\n                    polars.col(direction_axis).over('event_id')\n                ]).with_columns([\n                    polars.col(direction_axis).alias(direction_axis+'_'+direction+'_3')]\n                ).select(polars.col('event_id'), polars.col(direction_axis+'_'+direction+'_3'))\n\n                geometrical_features = geometrical_features.join(temp_col_1, on='event_id', how='left'\n                               ).join(temp_col_2, on='event_id', how='left'\n                               ).join(temp_col_3, on='event_id', how='left'\n                               )\n        return geometrical_features.fill_null(1000)\n\n    train_batch = polars.scan_parquet(batch_file).lazy()\n    df_train_meta = polars.DataFrame(metadata).lazy()\n    df_sensor_geometry = polars.DataFrame(sensor_geometry).with_columns(polars.col('sensor_id').cast(polars.Int16)).lazy()\n    \n        #Not accounting for aux\n    features_grouped_metrics = df_train_meta.pipe(join_tables_all, train_batch, df_sensor_geometry\n                      ).pipe(generate_features_grouped).collect()\n\n    geometrical_features = df_train_meta.pipe(join_tables_all, train_batch, df_sensor_geometry\n                      ).pipe(add_ranks\n                      ).collect().pipe(make_geometrical_features)\n\n    temp_1 = features_grouped_metrics.join(geometrical_features, on='event_id', how='left')\n\n\n    #AUX = FALSE\n\n    features_grouped_metrics = df_train_meta.pipe(join_tables_all, train_batch, df_sensor_geometry\n                      ).filter(polars.col('auxiliary') == False).pipe(generate_features_grouped).collect()\n\n    geometrical_features = df_train_meta.pipe(join_tables_all, train_batch, df_sensor_geometry\n                      ).filter(polars.col('auxiliary') == False).pipe(add_ranks\n                      ).collect().pipe(make_geometrical_features)\n\n    temp_2 = features_grouped_metrics.join(geometrical_features, on='event_id', how='left')\n    \n    temp_3 = temp_1.join(temp_2, on = 'event_id', how='left').fill_null(0)\n    del temp_1, temp_2, features_grouped_metrics, geometrical_features\n    \n    temp_3 = temp_3.to_pandas().set_index('event_id')\n    \n    temp_3 = (temp_3-temp_3.mean())/temp_3.std()\n    \n    return temp_3","metadata":{"papermill":{"duration":0.045874,"end_time":"2023-04-18T11:42:58.268841","exception":false,"start_time":"2023-04-18T11:42:58.222967","status":"completed"},"tags":[],"jupyter":{"source_hidden":true},"execution":{"iopub.status.busy":"2023-04-20T00:16:10.431394Z","iopub.execute_input":"2023-04-20T00:16:10.431756Z","iopub.status.idle":"2023-04-20T00:16:10.460110Z","shell.execute_reply.started":"2023-04-20T00:16:10.431723Z","shell.execute_reply":"2023-04-20T00:16:10.458871Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class NeutrinoDataset(Dataset):\n    def __init__(self, metadata, sensor_geometry, batch_file, mode=\"train\", max_nodes=100):\n        self.metadata = metadata\n        self.sensor_geometry = sensor_geometry\n        self.sensor_geometry = normalize_features(sensor_geometry, ['x', 'y', 'z']) \n        batch_df = pd.read_parquet(batch_file)\n        self.mode = mode\n        self.max_nodes = max_nodes\n        \n        self.batch_features = generating_some_features(batch_file, metadata, sensor_geometry)\n        \n        # Normalize time and charge features\n        batch_df['time'] = (batch_df['time'] - batch_df['time'].min()) / (batch_df['time'].max() - batch_df['time'].min())\n        batch_df['charge'] = (batch_df['charge'] - batch_df['charge'].min()) / (batch_df['charge'].max() - batch_df['charge'].min())\n        \n        self.batch_data = batch_df\n        \n    def get_event_features(self, event_data, idx):\n#         pca_features = process_event_pca(event_data)\n        batch_features = self.batch_features.loc[self.metadata.iloc[idx]['event_id']].fillna(0).values\n\n        combined_features = np.hstack([\n#             pca_features,\n            batch_features\n        ])\n        assert np.all(~np.isnan(combined_features))\n        return combined_features.astype(np.float32)\n        \n    def __len__(self):\n        return len(self.metadata)\n\n    def __getitem__(self, idx):\n        event = self.metadata.iloc[idx]\n        event_df = self.batch_data.loc[event.event_id]\n        if len(event_df) > 100:\n            event_df = event_df.sample(100)\n        \n        nodes_groupby = event_df.groupby('sensor_id').agg({\n            'sensor_id': 'size',\n            'time': ['min', 'max', 'mean'],\n            'charge': ['min', 'max', 'mean'],\n            'auxiliary': 'mean',\n        })\n\n        nodes = list(nodes_groupby.index)\n        nodes_features = np.concatenate((nodes_groupby.values, self.sensor_geometry.loc[nodes, ['x', 'y', 'z']].values), axis=1)\n        nodes_features = torch.tensor(nodes_features, dtype=torch.float)\n        # Pad node features\n        nodes_features_padded = np.zeros((self.max_nodes, nodes_features.shape[1]))\n        nodes_features_padded[:nodes_features.shape[0], :] = nodes_features\n        nodes_features_padded = torch.tensor(nodes_features_padded, dtype=torch.float)\n\n        \n        nodes_dict = {n: i for i, n in enumerate(nodes)}\n\n        seq = event_df.sort_values(by='time').sensor_id.values\n        edges = []\n        for i in range(1, len(seq)):\n            edges.append([nodes_dict[seq[i-1]], nodes_dict[seq[i]]])\n            \n            \n        adjacency_matrix = torch.zeros((self.max_nodes, self.max_nodes))\n        for i, j in edges:\n            adjacency_matrix[i, j] = 1\n            adjacency_matrix[j, i] = 1\n            \n        event_features = self.get_event_features(event_df, idx)\n\n        if self.mode == \"train\":\n            y = torch.tensor([event['azimuth'], event['zenith']], dtype=torch.float32)\n            y_norm = normalize_targets(y)\n            return nodes_features_padded, adjacency_matrix, event_features, y_norm\n        else:\n            return nodes_features_padded, adjacency_matrix, event_features, event.event_id\n","metadata":{"papermill":{"duration":0.033115,"end_time":"2023-04-18T11:42:58.310775","exception":false,"start_time":"2023-04-18T11:42:58.277660","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-04-20T00:16:10.463566Z","iopub.execute_input":"2023-04-20T00:16:10.464182Z","iopub.status.idle":"2023-04-20T00:16:10.483056Z","shell.execute_reply.started":"2023-04-20T00:16:10.464141Z","shell.execute_reply":"2023-04-20T00:16:10.481571Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class GraphConvolution(nn.Module):\n    def __init__(self, in_features, out_features):\n        super(GraphConvolution, self).__init__()\n        self.linear = nn.Linear(in_features, out_features)\n\n    def forward(self, adjacency_matrix, node_features):\n        # Normalize adjacency matrix\n        epsilon = 1e-8\n        degree_matrix_inv_sqrt = torch.diag_embed(torch.pow(adjacency_matrix.sum(dim=-1) + epsilon, -0.5))\n        normalized_adjacency_matrix = degree_matrix_inv_sqrt @ adjacency_matrix @ degree_matrix_inv_sqrt\n        \n        # Perform graph convolution\n        output = self.linear(torch.matmul(normalized_adjacency_matrix, node_features))\n        return output\n\nclass GCN(nn.Module):\n    def __init__(self, input_dim, hidden_dim, output_dim):\n        super(GCN, self).__init__()\n        self.graph_conv1 = GraphConvolution(input_dim, hidden_dim)\n        self.graph_conv2 = GraphConvolution(hidden_dim, hidden_dim)\n        self.graph_conv3 = GraphConvolution(hidden_dim, hidden_dim)\n        self.graph_conv4 = GraphConvolution(hidden_dim, hidden_dim)\n        self.graph_conv5 = GraphConvolution(hidden_dim, output_dim)\n\n    def forward(self, adjacency_matrix, node_features):\n        x = F.relu(self.graph_conv1(adjacency_matrix, node_features))\n        x = F.dropout(x, training=self.training)\n        x = F.relu(self.graph_conv2(adjacency_matrix, x))\n        x = F.dropout(x, training=self.training)\n        x = F.relu(self.graph_conv3(adjacency_matrix, x))\n        x = F.dropout(x, training=self.training)\n        x = F.relu(self.graph_conv4(adjacency_matrix, x))\n        x = F.dropout(x, training=self.training)\n        x = self.graph_conv5(adjacency_matrix, x)\n        return x\n\n","metadata":{"_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","papermill":{"duration":0.023821,"end_time":"2023-04-18T11:42:58.343340","exception":false,"start_time":"2023-04-18T11:42:58.319519","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-04-20T00:16:10.484468Z","iopub.execute_input":"2023-04-20T00:16:10.484796Z","iopub.status.idle":"2023-04-20T00:16:10.501054Z","shell.execute_reply.started":"2023-04-20T00:16:10.484763Z","shell.execute_reply":"2023-04-20T00:16:10.499851Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class NeutrinoDirectionModel(pl.LightningModule):\n    def __init__(self, input_features, hidden_features, output_gnn_features, num_fc_layers, input_fc_features, hidden_fc_units, output_features, learning_rate=1e-3):\n        super().__init__()\n        self.gcn = GCN(input_features, hidden_features, output_gnn_features)\n        \n        layers = []\n        layers.append(nn.BatchNorm1d(input_fc_features))\n\n        for i in range(num_fc_layers):\n            in_dim = input_fc_features if i == 0 else hidden_fc_units\n            out_dim = output_features if i == num_fc_layers - 1 else hidden_fc_units\n            layers.append(nn.Linear(in_dim, out_dim))\n\n            if i < num_fc_layers - 1:\n                layers.append(nn.ReLU())\n            else:\n                layers.append(nn.Sigmoid())\n\n        self.fc_layers = nn.Sequential(*layers)\n        \n        \n        self.lr = learning_rate\n        \n    def forward(self, node_features, adjacency_matrix, event_features):\n        x = self.gcn(adjacency_matrix, node_features)\n        x = torch.mean(x, dim=1)\n        x = torch.cat((x, event_features), 1)\n        x = self.fc_layers(x)\n        return x\n\n    def training_step(self, batch, batch_idx):\n        node_features, adjacency_matrix, event_features, y = batch\n        y_hat = self(node_features, adjacency_matrix, event_features)\n        loss = F.l1_loss(y_hat, y, reduction='mean')\n        self.log(\"train_loss\", loss, on_step=True, on_epoch=True, prog_bar=True, logger=True)\n        return loss\n\n    def validation_step(self, batch, batch_idx):\n        node_features, adjacency_matrix, event_features, y_norm = batch\n        y_hat_norm = self(node_features, adjacency_matrix, event_features)\n        loss = F.l1_loss(y_hat_norm, y_norm, reduction='mean')\n\n        # Denormalize target values and predictions\n        y = denormalize_targets(y_norm.cpu().numpy())\n        y_hat = denormalize_targets(y_hat_norm.detach().cpu().numpy())\n\n        # Calculate angular distance score\n        az_true, zen_true = y[:, 0], y[:, 1]\n        az_pred, zen_pred = y_hat[:, 0], y_hat[:, 1]\n        ang_dist = angular_dist_score(az_true, zen_true, az_pred, zen_pred)\n\n        self.log(\"val_loss\", loss, on_step=False, on_epoch=True, prog_bar=True, logger=True)\n        self.log(\"angular_dist_score\", ang_dist, on_step=False, on_epoch=True, prog_bar=True, logger=True)\n\n        return loss\n        \n    def configure_optimizers(self):\n        return torch.optim.Adam(self.parameters(), lr=self.lr)\n","metadata":{"papermill":{"duration":0.026631,"end_time":"2023-04-18T11:42:58.375936","exception":false,"start_time":"2023-04-18T11:42:58.349305","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-04-20T00:16:10.502389Z","iopub.execute_input":"2023-04-20T00:16:10.502820Z","iopub.status.idle":"2023-04-20T00:16:10.519452Z","shell.execute_reply.started":"2023-04-20T00:16:10.502789Z","shell.execute_reply":"2023-04-20T00:16:10.518248Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def train_one_epoch_on_batch(model, metadata, sensor_geometry, batch_file, val_split=0.2):\n    batch_id = int(batch_file.split('_')[-1].split('.')[0])\n    batch_metadata = metadata[metadata['batch_id'] == batch_id]\n    \n    train_metadata, val_metadata = train_test_split(batch_metadata, test_size=val_split, random_state=42)\n    \n    train_dataset = NeutrinoDataset(train_metadata, sensor_geometry, batch_file, mode=\"train\")\n    train_dataloader = DataLoader(train_dataset, batch_size=64, shuffle=True, num_workers=4)\n    \n    val_dataset = NeutrinoDataset(val_metadata, sensor_geometry, batch_file, mode=\"train\")\n    val_dataloader = DataLoader(val_dataset, batch_size=64, shuffle=False, num_workers=4)\n\n    trainer = pl.Trainer(max_epochs=1)\n    trainer.fit(model, train_dataloader, val_dataloader)\n    \n    del train_dataset, train_dataloader, val_dataset, val_dataloader\n    \n    return model","metadata":{"papermill":{"duration":0.018853,"end_time":"2023-04-18T11:42:58.503556","exception":false,"start_time":"2023-04-18T11:42:58.484703","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-04-20T00:16:10.521244Z","iopub.execute_input":"2023-04-20T00:16:10.521608Z","iopub.status.idle":"2023-04-20T00:16:10.534638Z","shell.execute_reply.started":"2023-04-20T00:16:10.521573Z","shell.execute_reply":"2023-04-20T00:16:10.533689Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data_dir = '/kaggle/input/icecube-neutrinos-in-deep-ice/'\nmetadata_file = f'{data_dir}/train_meta.parquet'\nbatch_files = [f'{data_dir}train/batch_{i}.parquet' for i in range(1, 2)] # You can set up to 660 files\n\nsensor_geometry = pd.read_csv(os.path.join(data_dir, 'sensor_geometry.csv'))\nmetadata = pd.read_parquet(metadata_file)","metadata":{"papermill":{"duration":45.15714,"end_time":"2023-04-18T11:43:43.666684","exception":false,"start_time":"2023-04-18T11:42:58.509544","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-04-20T00:16:10.536181Z","iopub.execute_input":"2023-04-20T00:16:10.536933Z","iopub.status.idle":"2023-04-20T00:16:50.452636Z","shell.execute_reply.started":"2023-04-20T00:16:10.536872Z","shell.execute_reply":"2023-04-20T00:16:50.451530Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"config = {\n    'input_features': 11,\n    'hidden_features': 64,\n    'output_gnn_features': 32,\n    'num_fc_layers': 4,\n    'input_fc_features': 104 + 32,\n    'hidden_fc_units': 512,\n    'output_features': 2,\n    'learning_rate': 0.001\n}\n\n\nmodel = NeutrinoDirectionModel(**config)","metadata":{"papermill":{"duration":0.061909,"end_time":"2023-04-18T11:43:43.735712","exception":false,"start_time":"2023-04-18T11:43:43.673803","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-04-20T00:16:50.454232Z","iopub.execute_input":"2023-04-20T00:16:50.455204Z","iopub.status.idle":"2023-04-20T00:16:50.504749Z","shell.execute_reply.started":"2023-04-20T00:16:50.455153Z","shell.execute_reply":"2023-04-20T00:16:50.503776Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"for epoch in range(1): # You can set any number of epochs but first use all of 660 files\n    for batch_file in batch_files:\n        model = train_one_epoch_on_batch(model, metadata, sensor_geometry, batch_file)","metadata":{"papermill":{"duration":12898.95943,"end_time":"2023-04-18T15:18:42.701006","exception":false,"start_time":"2023-04-18T11:43:43.741576","status":"completed"},"tags":[],"execution":{"iopub.status.busy":"2023-04-20T00:16:50.506090Z","iopub.execute_input":"2023-04-20T00:16:50.506470Z"},"trusted":true},"execution_count":null,"outputs":[]}]}