LSTM2_simple.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "import datetime\n",
    "import time\n",
    "import multiprocessing  as mp\n",
    "from itertools import product\n",
    "\n",
    "from keras.models import Sequential,Model\n",
    "from keras.layers import *#Dense, Input, Concatenate, Lambda,LSTM,Dropout,Layer,Flatten\n",
    "from keras.layers import ConvLSTM2D,BatchNormalization,TimeDistributed,MaxPooling3D,RepeatVector\n",
    "from keras.optimizers import RMSprop\n",
    "from keras.utils.vis_utils import plot_model\n",
    "import keras.losses as kl\n",
    "import pydot\n",
    "\n",
    "from sys import getsizeof\n",
    "import sys\n",
    "import warnings\n",
    "\n",
    "from functions import *\n",
    "\n",
    "idx = pd.IndexSlice"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# sys.modules[__name__].__dict__.clear()\n",
    "# Data.columns\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data import"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "startTime = 7*3600\n",
    "endTime = 10*3600\n",
    "dt = 30\n",
    "ncoeur = 8\n",
    "\n",
    "Data = get_time_data('data/data2/Times/fahrzeiten_soll_ist_20200105_20200111.csv')\n",
    "Data = Data.loc[Data['datum_nach'].dt.dayofweek < 5,:]\n",
    "#delete depot run\n",
    "Data = Data[~Data['fw_lang'].str.contains('DEP|Einfahrt|Ausfahrt')]\n",
    "\n",
    "\n",
    "Studied_line = 32\n",
    "# Data = Data[Data['linie']<100]\n",
    "Data = Data.sort_values(by ='halt_diva_nach')\n",
    "Data['ID'] = 1000000000*Data['richtung']+ 100000*Data['linie']+Data['halt_diva_nach']\n",
    "Data = Data.sort_values(by ='ID')\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Selection\n",
    "### Stops\n",
    "Select the 3 lateness (travel time, stops and total ) for every stops every 30s, if no new bus came, the last one is kept. It was first thought to take as time step the updating of the bus lateness, but as we now have multiple bus it does not work."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Time spent:  173.9890992641449\n"
     ]
    }
   ],
   "source": [
    "\n",
    "t1 = time.time()\n",
    "def features_selector(Data_d):\n",
    "    Data_d = pd.DataFrame({\n",
    "        'Delta_stops'  :(Data_d['ist_ab_nach']-Data_d['ist_an_nach1']-Data_d['soll_ab_nach']+Data_d['soll_an_nach'])/(\n",
    "                        Data_d['soll_ab_nach']-Data_d['soll_an_nach']),\n",
    "        'Delta_trip'   :(Data_d['ist_an_nach1']-Data_d['ist_ab_von']-Data_d['soll_an_nach']+Data_d['soll_ab_von'])/(\n",
    "                          Data_d['soll_an_nach']-Data_d['soll_an_von'])  ,\n",
    "        'Tot_lat'      :Data_d['ist_ab_nach']-Data_d['soll_ab_nach'],\n",
    "        'ID'           :Data_d['ID'],\n",
    "        'time'         :Data_d['ist_ab_nach']})\n",
    "\n",
    "    colname = [np.repeat(Data['ID'].unique(),3),np.resize(['Delta_stops','Delta_trip','Tot_lat'\n",
    "                                                             ],3*Data['ID'].unique().shape[0])]\n",
    "\n",
    "    out = pd.DataFrame(index = range(startTime,endTime,dt) ,columns= colname)\n",
    "    t_last = startTime\n",
    "\n",
    "    Data_t = Data_d.loc[(Data_d['time'] < t_last) ,['Delta_stops','Delta_trip','Tot_lat','ID']]\n",
    "\n",
    "    out.loc[t_last,idx[Data_t['ID'].values,:]] = Data_t.groupby('ID').last().values.ravel()\n",
    "\n",
    "    for t in range(startTime,endTime,dt):\n",
    "        Data_t = Data_d.loc[(Data_d['time'] > t_last) &(Data_d['time'] <= t),['Delta_stops','Delta_trip','Tot_lat','ID']]\n",
    "\n",
    "        out.loc[t] = out.loc[t_last]\n",
    "        out.loc[t,idx[Data_t['ID'].values,:]] = Data_t.groupby('ID').mean().values.ravel()\n",
    "\n",
    "        t_last = t\n",
    "        \n",
    "    return np.array([out.loc[:,idx[:,'Delta_stops']].to_numpy(),out.loc[:,idx[:,'Delta_trip']].to_numpy(),out.loc[:,idx[:,'Tot_lat']].to_numpy()])   \n",
    "    del(Data_t,Data_d,out)\n",
    "\n",
    "pool = mp.Pool(ncoeur) \n",
    "i = 0\n",
    "for day in Data['datum_nach'].unique():\n",
    "    if i == 0:\n",
    "        splitData = list([Data.loc[Data['datum_nach'] == day,:]])\n",
    "    else:\n",
    "        splitData.append(Data.loc[Data['datum_nach'] == day,:])\n",
    "    i+=1\n",
    "    \n",
    "FrameT = np.array(pool.map(features_selector,splitData),np.float32)#np.array(,np.float32)\n",
    "pool.close()\n",
    "pool.join() # 28\n",
    "\n",
    "print('Time spent: ',time.time()-t1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Reshape into (#sample , time ,features, #stop, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(5, 360, 3, 2472, 1)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "FrameT = np.moveaxis(FrameT, [0, 1, 2, 3], [0, 2, 1, 3])\n",
    "FrameT = np.expand_dims(FrameT,axis = 4)\n",
    "FrameT.shape #(5, 360, 3, 2472, 1) 1539\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Bus\n",
    "The bus delay for all the bus in the line is computed every 30s.  ( don't need reshape already good)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/gneven/anaconda3/lib/python3.7/site-packages/ipykernel_launcher.py:5: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  \"\"\"\n",
      "/gneven/anaconda3/lib/python3.7/site-packages/ipykernel_launcher.py:5: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  \"\"\"\n",
      "/gneven/anaconda3/lib/python3.7/site-packages/ipykernel_launcher.py:5: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  \"\"\"\n",
      "/gneven/anaconda3/lib/python3.7/site-packages/ipykernel_launcher.py:5: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  \"\"\"\n",
      "/gneven/anaconda3/lib/python3.7/site-packages/ipykernel_launcher.py:5: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  \"\"\"\n"
     ]
    }
   ],
   "source": [
    "def get_out(Data_d):\n",
    "    Data_d = splitData[0]\n",
    "    Data_d = Data_d.loc[Data_d['linie'] == Studied_line,:]\n",
    "    \n",
    "    Data_d['Tot_latness'] = Data_d['ist_ab_nach']-Data_d['soll_ab_nach']\n",
    "    \n",
    "    idBus = Data_d.loc[(Data_d['soll_ab_nach']>=startTime ) &(\n",
    "                        (Data_d['soll_ab_nach']<=endTime )\n",
    "            ) & (Data_d['halt_diva_nach'] == 1143) ,'fahrzeug'].unique()\n",
    "    \n",
    "    \n",
    "    out= pd.DataFrame(index = range(startTime,endTime,dt), columns = idBus)\n",
    "    \n",
    "    t_last = startTime\n",
    "    Data_t = Data_d.loc[(Data_d['ist_ab_nach'] < t_last),['Tot_latness','fahrzeug']]\n",
    "    out.loc[t_last].update(Data_t.groupby('fahrzeug').last()['Tot_latness'])\n",
    "\n",
    "    for t in range(startTime+dt,endTime,dt):\n",
    "        Data_t = Data_d.loc[(Data_d['ist_ab_nach'] > t_last) &(Data_d['ist_ab_nach'] <= t),['Tot_latness','fahrzeug']]\n",
    "        if Data_t['fahrzeug'].size == 0:\n",
    "            out.loc[t,:] = out.loc[t_last,:]\n",
    "        else:\n",
    "            out.loc[t,:] = out.loc[t_last,:]\n",
    "            out.loc[t,Data_t['fahrzeug'].unique()] =  Data_t.groupby('fahrzeug').mean().values.ravel()\n",
    "\n",
    "        t_last = t\n",
    "        \n",
    "        out = out.dropna(axis='columns',how='all')\n",
    "        \n",
    "    return  out.to_numpy()\n",
    "    del(Data_t,Data_d,out)\n",
    "\n",
    "pool = mp.Pool(ncoeur) \n",
    "\n",
    "out = np.array(pool.map(get_out,splitData),np.float32)\n",
    "pool.close()\n",
    "pool.join() # 28  \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## LSTM model\n",
    "\n",
    "do loop as proposed with prediction from before as training "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_6\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv_lst_m2d_1 (ConvLSTM2D)  (None, 360, 3, 2472, 20)  16880     \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 360, 3, 2472, 20)  0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_1 (Batch (None, 360, 3, 2472, 20)  80        \n",
      "_________________________________________________________________\n",
      "conv_lst_m2d_2 (ConvLSTM2D)  (None, 360, 3, 2472, 10)  6040      \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 360, 3, 2472, 10)  0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_2 (Batch (None, 360, 3, 2472, 10)  40        \n",
      "_________________________________________________________________\n",
      "conv_lst_m2d_3 (ConvLSTM2D)  (None, 360, 3, 2472, 10)  8040      \n",
      "_________________________________________________________________\n",
      "dropout_3 (Dropout)          (None, 360, 3, 2472, 10)  0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_3 (Batch (None, 360, 3, 2472, 10)  40        \n",
      "_________________________________________________________________\n",
      "conv_lstm_4 (ConvLSTM2D)     (None, 360, 3, 2472, 20)  12080     \n",
      "_________________________________________________________________\n",
      "time_distributed_2 (TimeDist (None, 360, 148320)       0         \n",
      "_________________________________________________________________\n",
      "time_distributed_3 (TimeDist (None, 360, 512)          75940352  \n",
      "_________________________________________________________________\n",
      "time_distributed_4 (TimeDist (None, 360, 32)           16416     \n",
      "_________________________________________________________________\n",
      "test (TimeDistributed)       (None, 360, 13)           429       \n",
      "=================================================================\n",
      "Total params: 76,000,397\n",
      "Trainable params: 76,000,317\n",
      "Non-trainable params: 80\n",
      "_________________________________________________________________\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\n",
    "model = Sequential([\n",
    "    ConvLSTM2D(filters=20, kernel_size=(10, 1)\n",
    "                       , padding='same', return_sequences=True\n",
    "                       , input_shape = (FrameT.shape[1],FrameT.shape[2],FrameT.shape[3],FrameT.shape[4])),\n",
    "    Dropout(0.2),\n",
    "    \n",
    "    BatchNormalization(),\n",
    "        \n",
    "    ConvLSTM2D(filters=10, kernel_size=(5, 1)\n",
    "                         ,padding='same', return_sequences=True),\n",
    "    \n",
    "    Dropout(0.1),\n",
    "    \n",
    "    BatchNormalization(),\n",
    "        \n",
    "    ConvLSTM2D(filters=10,  kernel_size=(10, 1)\n",
    "                       , padding='same', return_sequences=True),\n",
    "    Dropout(0.1),\n",
    "    BatchNormalization(),\n",
    "    ConvLSTM2D(name ='conv_lstm_4',\n",
    "                         filters = 20, kernel_size = (5, 1), \n",
    "                         padding='same',\n",
    "                         return_sequences = True),\n",
    "        \n",
    "    TimeDistributed(Flatten()),\n",
    "    \n",
    "    TimeDistributed(Dense(512,)),\n",
    "    \n",
    "    TimeDistributed(Dense(32,)),\n",
    "    \n",
    "    TimeDistributed(Dense(out.shape[2]), name='test'),\n",
    "\n",
    "\n",
    "    \n",
    "    \n",
    "])\n",
    "\n",
    "model.compile(loss='mean_absolute_error', optimizer = RMSprop())\n",
    "model.summary()\n",
    "\n",
    "plot_model(model)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 1\n",
    "epochs = 5\n",
    "t1 = time.time()\n",
    "history = model.fit(FrameT[:4,:,:,:,:],out[:4,:,:], batch_size=\n",
    "                            batch_size,epochs=epochs,verbose=1,validation_data=(FrameT[4:5,:,:,:,:], out[4:5,:,:]))\n",
    "\n",
    "print('time spend:',t1-time.time())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "t1 = time.time()\n",
    "latness_predict = model.predict(FrameT[4:5,:,:,:,:])\n",
    "print('time spend:',t1-time.time())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "loss = history.history['loss']\n",
    "loss_val = history.history['val_loss']\n",
    "epochs_plot = range(epochs)\n",
    "plt.figure(2)\n",
    "plt.plot(epochs_plot, loss, 'bo', label='Training loss') #val_loss\n",
    "plt.plot(epochs_plot, loss_val, 'bs', label='Validation loss') #val_loss\n",
    "\n",
    "plt.title('Loss')\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('Loss')\n",
    "plt.legend()\n",
    "plt.show()\n",
    "plt.savefig('test.png')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# plt.figure(3,figsize = (12,5))\n",
    "\n",
    "# which = 3\n",
    "\n",
    "# plt.plot(latness_predict[0,:,which], label='Predict') #val_loss\n",
    "# plt.plot(out[0,:,which], label='Real') #val_loss\n",
    "\n",
    "# plt.title('Loss')\n",
    "# plt.xlabel('Time')\n",
    "# plt.ylabel('Latness')\n",
    "# plt.legend()\n",
    "# plt.show()\n",
    "# out.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_5\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "batch_norm_0 (BatchNormaliza (None, 32, 32, 1, 1)      4         \n",
      "_________________________________________________________________\n",
      "conv_lstm_1 (ConvLSTM2D)     (None, 32, 32, 1, 64)     166656    \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 32, 32, 1, 64)     0         \n",
      "_________________________________________________________________\n",
      "batch_norm_1 (BatchNormaliza (None, 32, 32, 1, 64)     256       \n",
      "_________________________________________________________________\n",
      "conv_lstm_2 (ConvLSTM2D)     (None, 32, 1, 64)         164096    \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 32, 1, 64)         0         \n",
      "_________________________________________________________________\n",
      "batch_norm_2 (BatchNormaliza (None, 32, 1, 64)         256       \n",
      "_________________________________________________________________\n",
      "flatten_5 (Flatten)          (None, 2048)              0         \n",
      "_________________________________________________________________\n",
      "repeat_vector_3 (RepeatVecto (None, 3, 2048)           0         \n",
      "_________________________________________________________________\n",
      "reshape_1 (Reshape)          (None, 3, 32, 1, 64)      0         \n",
      "_________________________________________________________________\n",
      "conv_lstm_3 (ConvLSTM2D)     (None, 3, 32, 1, 64)      327936    \n",
      "_________________________________________________________________\n",
      "dropout_3 (Dropout)          (None, 3, 32, 1, 64)      0         \n",
      "_________________________________________________________________\n",
      "batch_norm_3 (BatchNormaliza (None, 3, 32, 1, 64)      256       \n",
      "_________________________________________________________________\n",
      "conv_lstm_4 (ConvLSTM2D)     (None, 3, 32, 1, 64)      164096    \n",
      "_________________________________________________________________\n",
      "time_distributed_1 (TimeDist (None, 3, 32, 1, 1)       65        \n",
      "=================================================================\n",
      "Total params: 823,621\n",
      "Trainable params: 823,235\n",
      "Non-trainable params: 386\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = Sequential()\n",
    "model.add(BatchNormalization(name = 'batch_norm_0', input_shape = (32, 32, 1, 1)))\n",
    "model.add(ConvLSTM2D(name ='conv_lstm_1',\n",
    "                         filters = 64, kernel_size = (10, 1),                       \n",
    "                         padding = 'same', \n",
    "                         return_sequences = True))\n",
    "\n",
    "model.add(Dropout(0.2, name = 'dropout_1'))\n",
    "model.add(BatchNormalization(name = 'batch_norm_1'))\n",
    "\n",
    "model.add(ConvLSTM2D(name ='conv_lstm_2',\n",
    "                         filters = 64, kernel_size = (5, 1), \n",
    "                         padding='same',\n",
    "                         return_sequences = False))\n",
    "\n",
    "model.add(Dropout(0.1, name = 'dropout_2'))\n",
    "model.add(BatchNormalization(name = 'batch_norm_2'))\n",
    "\n",
    "model.add(Flatten())\n",
    "model.add(RepeatVector(3))\n",
    "model.add(Reshape((3, 32, 1, 64)))\n",
    "\n",
    "model.add(ConvLSTM2D(name ='conv_lstm_3',\n",
    "                         filters = 64, kernel_size = (10, 1), \n",
    "                         padding='same',\n",
    "                         return_sequences = True))\n",
    "\n",
    "model.add(Dropout(0.1, name = 'dropout_3'))\n",
    "model.add(BatchNormalization(name = 'batch_norm_3'))\n",
    "model.add(ConvLSTM2D(name ='conv_lstm_4',\n",
    "                         filters = 64, kernel_size = (5, 1), \n",
    "                         padding='same',\n",
    "                         return_sequences = True))\n",
    "\n",
    "model.add(TimeDistributed(Dense(units=1, name = 'dense_1', activation = 'relu')))\n",
    "    #model.add(Dense(units=1, name = 'dense_2'))\n",
    "    \n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}