diff --git a/attention.py b/attention.py
new file mode 100644
index 0000000000000000000000000000000000000000..69a94dc5f9e7b31209dc252e9912404ea178c1a0
--- /dev/null
+++ b/attention.py
@@ -0,0 +1,80 @@
+# Provided by under the Apache License, Version 2.0 (https://www.apache.org/licenses/LICENSE-2.0)
+# Source: https://github.com/philipperemy/keras-attention-mechanism
+import os
+
+from tensorflow.keras import backend as K
+from tensorflow.keras.layers import Dense, Lambda, Dot, Activation, Concatenate, Layer
+
+# KERAS_ATTENTION_DEBUG: If set to 1. Will switch to debug mode.
+# In debug mode, the class Attention is no longer a Keras layer.
+# What it means in practice is that we can have access to the internal values
+# of each tensor. If we don't use debug, Keras treats the object
+# as a layer and we can only get the final output.
+debug_flag = int(os.environ.get('KERAS_ATTENTION_DEBUG', 0))
+
+
+class Attention(object if debug_flag else Layer):
+
+ def __init__(self, units=128, **kwargs):
+ super(Attention, self).__init__(**kwargs)
+ self.units = units
+
+ # noinspection PyAttributeOutsideInit
+ def build(self, input_shape):
+ input_dim = int(input_shape[-1])
+ with K.name_scope(self.name if not debug_flag else 'attention'):
+ self.attention_score_vec = Dense(input_dim, use_bias=False, name='attention_score_vec')
+ self.h_t = Lambda(lambda x: x[:, -1, :], output_shape=(input_dim,), name='last_hidden_state')
+ self.attention_score = Dot(axes=[1, 2], name='attention_score')
+ self.attention_weight = Activation('softmax', name='attention_weight')
+ self.context_vector = Dot(axes=[1, 1], name='context_vector')
+ self.attention_output = Concatenate(name='attention_output')
+ self.attention_vector = Dense(self.units, use_bias=False, activation='tanh', name='attention_vector')
+ if not debug_flag:
+ # debug: the call to build() is done in call().
+ super(Attention, self).build(input_shape)
+
+ def compute_output_shape(self, input_shape):
+ return input_shape[0], self.units
+
+ def __call__(self, inputs, training=None, **kwargs):
+ if debug_flag:
+ return self.call(inputs, training, **kwargs)
+ else:
+ return super(Attention, self).__call__(inputs, training, **kwargs)
+
+ # noinspection PyUnusedLocal
+ def call(self, inputs, training=None, **kwargs):
+ """
+ Many-to-one attention mechanism for Keras.
+ @param inputs: 3D tensor with shape (batch_size, time_steps, input_dim).
+ @param training: not used in this layer.
+ @return: 2D tensor with shape (batch_size, units)
+ @author: felixhao28, philipperemy.
+ """
+ if debug_flag:
+ self.build(inputs.shape)
+ # Inside dense layer
+ # hidden_states dot W => score_first_part
+ # (batch_size, time_steps, hidden_size) dot (hidden_size, hidden_size) => (batch_size, time_steps, hidden_size)
+ # W is the trainable weight matrix of attention Luong's multiplicative style score
+ score_first_part = self.attention_score_vec(inputs)
+ # score_first_part dot last_hidden_state => attention_weights
+ # (batch_size, time_steps, hidden_size) dot (batch_size, hidden_size) => (batch_size, time_steps)
+ h_t = self.h_t(inputs)
+ score = self.attention_score([h_t, score_first_part])
+ attention_weights = self.attention_weight(score)
+ # (batch_size, time_steps, hidden_size) dot (batch_size, time_steps) => (batch_size, hidden_size)
+ context_vector = self.context_vector([inputs, attention_weights])
+ pre_activation = self.attention_output([context_vector, h_t])
+ attention_vector = self.attention_vector(pre_activation)
+ return attention_vector
+
+ def get_config(self):
+ """
+ Returns the config of a the layer. This is used for saving and loading from a model
+ :return: python dictionary with specs to rebuild layer
+ """
+ config = super(Attention, self).get_config()
+ config.update({'units': self.units})
+ return config
\ No newline at end of file
diff --git a/tf_model.ipynb b/tf_model.ipynb
index 259961689cdafba423c4f4cf1f2bd237855cab12..efcdfa79fba233262de74809a897831726664598 100644
--- a/tf_model.ipynb
+++ b/tf_model.ipynb
@@ -14,9 +14,11 @@
"outputs": [],
"source": [
"import tensorflow as tf\n",
+ "import numpy as np\n",
"from tensorflow import keras\n",
"from tensorflow.keras import layers\n",
- "import pandas as pd"
+ "import pandas as pd\n",
+ "from attention import Attention"
]
},
{
@@ -38,7 +40,7 @@
},
{
"cell_type": "code",
- "execution_count": 14,
+ "execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
@@ -47,20 +49,14 @@
"CLEANED_SLEEP_DATA_PATH = \".data/clean_bed_sleep-state.csv\""
]
},
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "## Parameters and Hyper-parameters"
- ]
- },
{
"cell_type": "code",
- "execution_count": 4,
+ "execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
- "SLEEP_STAGES = 4\n"
+ "## Parameters and Hyper-parameters\n",
+ "SLEEP_STAGES = 4"
]
},
{
@@ -140,7 +136,7 @@
},
{
"cell_type": "code",
- "execution_count": 156,
+ "execution_count": 201,
"metadata": {},
"outputs": [],
"source": [
@@ -149,12 +145,14 @@
"previous_duration = 60\n",
"with open(RAW_SLEEP_DATA_PATH, mode ='r') as raw_file:\n",
" csvFile = csv.reader(raw_file)\n",
- " max_count = 1\n",
- " stuff = set()\n",
+ " # max_count = 1\n",
+ " # stuff = set()\n",
" in_bed_time = None\n",
+ " current_sleep_id = -1\n",
" for index, lines in enumerate(csvFile):\n",
" if index == 0:\n",
" cleaned_info.append([\n",
+ " \"sleep_id\",\n",
" \"sleep_begin\",\n",
" \"stage_start\",\n",
" \"time_since_begin_sec\",\n",
@@ -172,6 +170,7 @@
" continue\n",
" date_seen.add(start_time)\n",
" if not in_bed_time or in_bed_time > start_time:\n",
+ " current_sleep_id += 1\n",
" in_bed_time = start_time\n",
" # for duration, stage in enumerate(\n",
" # for offset, (duration, stage) in enumerate(\n",
@@ -187,6 +186,7 @@
" # print(f\"{(index, duration) = } {stage = }\")\n",
" current_time = start_time + datetime.timedelta(seconds=offset*previous_duration)\n",
" cleaned_info.append([\n",
+ " current_sleep_id,\n",
" in_bed_time,\n",
" current_time, \n",
" (current_time - in_bed_time).seconds,\n",
@@ -207,13 +207,22 @@
},
{
"cell_type": "code",
- "execution_count": 157,
+ "execution_count": 202,
"metadata": {},
- "outputs": [],
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Finished Writing Cleaned Data\n"
+ ]
+ }
+ ],
"source": [
"with open(CLEANED_SLEEP_DATA_PATH, 'w') as clean_file:\n",
" write = csv.writer(clean_file)\n",
- " write.writerows(cleaned_info)"
+ " write.writerows(cleaned_info)\n",
+ "print(\"Finished Writing Cleaned Data\")"
]
},
{
@@ -225,7 +234,7 @@
},
{
"cell_type": "code",
- "execution_count": 158,
+ "execution_count": 6,
"metadata": {},
"outputs": [],
"source": [
@@ -235,7 +244,7 @@
},
{
"cell_type": "code",
- "execution_count": 160,
+ "execution_count": 7,
"metadata": {},
"outputs": [],
"source": [
@@ -244,12 +253,36 @@
"sleep_df_raw[\"sleep_begin\"] = pd.to_datetime(sleep_df_raw[\"sleep_begin\"], utc=True)\n",
"sleep_df_raw[\"stage_start\"] = pd.to_datetime(sleep_df_raw[\"stage_start\"], utc=True)\n",
"sleep_df_raw[\"stage_end\"] = pd.to_datetime(sleep_df_raw[\"stage_end\"], utc=True)\n",
- "# MAYBE 2. smaller units: int16 or int8 "
+ "# 2. Separate time, hour and minute\n",
+ "# MAYBE 3. smaller units: int16 or int8 "
]
},
{
"cell_type": "code",
- "execution_count": 161,
+ "execution_count": 8,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "def get_minute(row, index):\n",
+ " return row[index].time().minute\n",
+ "\n",
+ "def get_hour(row, index):\n",
+ " return row[index].time().hour"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 9,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "sleep_df_raw[\"stage_start_hour\"] = sleep_df_raw.apply (lambda row: get_hour(row, \"stage_start\"), axis=1)\n",
+ "sleep_df_raw[\"stage_start_minute\"] = sleep_df_raw.apply (lambda row: get_minute(row, \"stage_start\"), axis=1)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 10,
"metadata": {},
"outputs": [
{
@@ -258,21 +291,24 @@
"text": [
"<class 'pandas.core.frame.DataFrame'>\n",
"RangeIndex: 551042 entries, 0 to 551041\n",
- "Data columns (total 10 columns):\n",
+ "Data columns (total 13 columns):\n",
" # Column Non-Null Count Dtype \n",
"--- ------ -------------- ----- \n",
- " 0 sleep_begin 551042 non-null datetime64[ns, UTC]\n",
- " 1 stage_start 551042 non-null datetime64[ns, UTC]\n",
- " 2 time_since_begin_sec 551042 non-null int64 \n",
- " 3 stage_duration_sec 551042 non-null int64 \n",
- " 4 stage_end 551042 non-null datetime64[ns, UTC]\n",
- " 5 stage_value 551042 non-null int64 \n",
- " 6 awake_probability 551042 non-null float64 \n",
- " 7 light_probability 551042 non-null float64 \n",
- " 8 deep_probability 551042 non-null float64 \n",
- " 9 rem_probability 551042 non-null float64 \n",
- "dtypes: datetime64[ns, UTC](3), float64(4), int64(3)\n",
- "memory usage: 42.0 MB\n"
+ " 0 sleep_id 551042 non-null int64 \n",
+ " 1 sleep_begin 551042 non-null datetime64[ns, UTC]\n",
+ " 2 stage_start 551042 non-null datetime64[ns, UTC]\n",
+ " 3 time_since_begin_sec 551042 non-null int64 \n",
+ " 4 stage_duration_sec 551042 non-null int64 \n",
+ " 5 stage_end 551042 non-null datetime64[ns, UTC]\n",
+ " 6 stage_value 551042 non-null int64 \n",
+ " 7 awake_probability 551042 non-null float64 \n",
+ " 8 light_probability 551042 non-null float64 \n",
+ " 9 deep_probability 551042 non-null float64 \n",
+ " 10 rem_probability 551042 non-null float64 \n",
+ " 11 stage_start_hour 551042 non-null int64 \n",
+ " 12 stage_start_minute 551042 non-null int64 \n",
+ "dtypes: datetime64[ns, UTC](3), float64(4), int64(6)\n",
+ "memory usage: 54.7 MB\n"
]
}
],
@@ -282,7 +318,7 @@
},
{
"cell_type": "code",
- "execution_count": 162,
+ "execution_count": 11,
"metadata": {},
"outputs": [
{
@@ -306,6 +342,7 @@
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
+ " <th>sleep_id</th>\n",
" <th>sleep_begin</th>\n",
" <th>stage_start</th>\n",
" <th>time_since_begin_sec</th>\n",
@@ -316,11 +353,14 @@
" <th>light_probability</th>\n",
" <th>deep_probability</th>\n",
" <th>rem_probability</th>\n",
+ " <th>stage_start_hour</th>\n",
+ " <th>stage_start_minute</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
+ " <td>0</td>\n",
" <td>2022-04-21 08:18:00+00:00</td>\n",
" <td>2022-04-21 08:18:00+00:00</td>\n",
" <td>0</td>\n",
@@ -331,9 +371,12 @@
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
+ " <td>8</td>\n",
+ " <td>18</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
+ " <td>0</td>\n",
" <td>2022-04-21 08:18:00+00:00</td>\n",
" <td>2022-04-21 08:19:00+00:00</td>\n",
" <td>60</td>\n",
@@ -344,9 +387,12 @@
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
+ " <td>8</td>\n",
+ " <td>19</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
+ " <td>0</td>\n",
" <td>2022-04-21 08:18:00+00:00</td>\n",
" <td>2022-04-21 08:20:00+00:00</td>\n",
" <td>120</td>\n",
@@ -357,9 +403,12 @@
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
+ " <td>8</td>\n",
+ " <td>20</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
+ " <td>0</td>\n",
" <td>2022-04-21 08:18:00+00:00</td>\n",
" <td>2022-04-21 08:21:00+00:00</td>\n",
" <td>180</td>\n",
@@ -370,9 +419,12 @@
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
+ " <td>8</td>\n",
+ " <td>21</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
+ " <td>0</td>\n",
" <td>2022-04-21 08:18:00+00:00</td>\n",
" <td>2022-04-21 08:22:00+00:00</td>\n",
" <td>240</td>\n",
@@ -383,6 +435,8 @@
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
+ " <td>8</td>\n",
+ " <td>22</td>\n",
" </tr>\n",
" <tr>\n",
" <th>...</th>\n",
@@ -396,9 +450,13 @@
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
+ " <td>...</td>\n",
+ " <td>...</td>\n",
+ " <td>...</td>\n",
" </tr>\n",
" <tr>\n",
" <th>551037</th>\n",
+ " <td>1132</td>\n",
" <td>2019-02-11 06:11:00+00:00</td>\n",
" <td>2019-02-11 13:17:00+00:00</td>\n",
" <td>25560</td>\n",
@@ -409,9 +467,12 @@
" <td>1.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
+ " <td>13</td>\n",
+ " <td>17</td>\n",
" </tr>\n",
" <tr>\n",
" <th>551038</th>\n",
+ " <td>1132</td>\n",
" <td>2019-02-11 06:11:00+00:00</td>\n",
" <td>2019-02-11 13:18:00+00:00</td>\n",
" <td>25620</td>\n",
@@ -422,9 +483,12 @@
" <td>1.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
+ " <td>13</td>\n",
+ " <td>18</td>\n",
" </tr>\n",
" <tr>\n",
" <th>551039</th>\n",
+ " <td>1132</td>\n",
" <td>2019-02-11 06:11:00+00:00</td>\n",
" <td>2019-02-11 13:19:00+00:00</td>\n",
" <td>25680</td>\n",
@@ -435,9 +499,12 @@
" <td>1.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
+ " <td>13</td>\n",
+ " <td>19</td>\n",
" </tr>\n",
" <tr>\n",
" <th>551040</th>\n",
+ " <td>1132</td>\n",
" <td>2019-02-11 06:11:00+00:00</td>\n",
" <td>2019-02-11 13:20:00+00:00</td>\n",
" <td>25740</td>\n",
@@ -448,9 +515,12 @@
" <td>1.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
+ " <td>13</td>\n",
+ " <td>20</td>\n",
" </tr>\n",
" <tr>\n",
" <th>551041</th>\n",
+ " <td>1132</td>\n",
" <td>2019-02-11 06:11:00+00:00</td>\n",
" <td>2019-02-11 13:21:00+00:00</td>\n",
" <td>25800</td>\n",
@@ -461,25 +531,27 @@
" <td>1.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
+ " <td>13</td>\n",
+ " <td>21</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
- "<p>551042 rows × 10 columns</p>\n",
+ "<p>551042 rows × 13 columns</p>\n",
"</div>"
],
"text/plain": [
- " sleep_begin stage_start \\\n",
- "0 2022-04-21 08:18:00+00:00 2022-04-21 08:18:00+00:00 \n",
- "1 2022-04-21 08:18:00+00:00 2022-04-21 08:19:00+00:00 \n",
- "2 2022-04-21 08:18:00+00:00 2022-04-21 08:20:00+00:00 \n",
- "3 2022-04-21 08:18:00+00:00 2022-04-21 08:21:00+00:00 \n",
- "4 2022-04-21 08:18:00+00:00 2022-04-21 08:22:00+00:00 \n",
- "... ... ... \n",
- "551037 2019-02-11 06:11:00+00:00 2019-02-11 13:17:00+00:00 \n",
- "551038 2019-02-11 06:11:00+00:00 2019-02-11 13:18:00+00:00 \n",
- "551039 2019-02-11 06:11:00+00:00 2019-02-11 13:19:00+00:00 \n",
- "551040 2019-02-11 06:11:00+00:00 2019-02-11 13:20:00+00:00 \n",
- "551041 2019-02-11 06:11:00+00:00 2019-02-11 13:21:00+00:00 \n",
+ " sleep_id sleep_begin stage_start \\\n",
+ "0 0 2022-04-21 08:18:00+00:00 2022-04-21 08:18:00+00:00 \n",
+ "1 0 2022-04-21 08:18:00+00:00 2022-04-21 08:19:00+00:00 \n",
+ "2 0 2022-04-21 08:18:00+00:00 2022-04-21 08:20:00+00:00 \n",
+ "3 0 2022-04-21 08:18:00+00:00 2022-04-21 08:21:00+00:00 \n",
+ "4 0 2022-04-21 08:18:00+00:00 2022-04-21 08:22:00+00:00 \n",
+ "... ... ... ... \n",
+ "551037 1132 2019-02-11 06:11:00+00:00 2019-02-11 13:17:00+00:00 \n",
+ "551038 1132 2019-02-11 06:11:00+00:00 2019-02-11 13:18:00+00:00 \n",
+ "551039 1132 2019-02-11 06:11:00+00:00 2019-02-11 13:19:00+00:00 \n",
+ "551040 1132 2019-02-11 06:11:00+00:00 2019-02-11 13:20:00+00:00 \n",
+ "551041 1132 2019-02-11 06:11:00+00:00 2019-02-11 13:21:00+00:00 \n",
"\n",
" time_since_begin_sec stage_duration_sec stage_end \\\n",
"0 0 60 2022-04-21 08:19:00+00:00 \n",
@@ -507,23 +579,23 @@
"551040 1 0.0 1.0 0.0 \n",
"551041 1 0.0 1.0 0.0 \n",
"\n",
- " rem_probability \n",
- "0 0.0 \n",
- "1 0.0 \n",
- "2 0.0 \n",
- "3 0.0 \n",
- "4 0.0 \n",
- "... ... \n",
- "551037 0.0 \n",
- "551038 0.0 \n",
- "551039 0.0 \n",
- "551040 0.0 \n",
- "551041 0.0 \n",
+ " rem_probability stage_start_hour stage_start_minute \n",
+ "0 0.0 8 18 \n",
+ "1 0.0 8 19 \n",
+ "2 0.0 8 20 \n",
+ "3 0.0 8 21 \n",
+ "4 0.0 8 22 \n",
+ "... ... ... ... \n",
+ "551037 0.0 13 17 \n",
+ "551038 0.0 13 18 \n",
+ "551039 0.0 13 19 \n",
+ "551040 0.0 13 20 \n",
+ "551041 0.0 13 21 \n",
"\n",
- "[551042 rows x 10 columns]"
+ "[551042 rows x 13 columns]"
]
},
- "execution_count": 162,
+ "execution_count": 11,
"metadata": {},
"output_type": "execute_result"
}
@@ -534,35 +606,451 @@
},
{
"cell_type": "code",
- "execution_count": 143,
+ "execution_count": 12,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "sleep_data = sleep_df_raw[[\"sleep_id\", \"stage_start_hour\", \"stage_start_minute\", \"awake_probability\", \"rem_probability\",\"light_probability\", \"deep_probability\"]]\n",
+ "sleep_data.insert(loc=1, column=\"minutes_since_begin\" , value= sleep_df_raw[\"time_since_begin_sec\"]//60)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 13,
"metadata": {},
"outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ " sleep_id minutes_since_begin stage_start_hour stage_start_minute \\\n",
+ "0 0 0 8 18 \n",
+ "1 0 1 8 19 \n",
+ "2 0 2 8 20 \n",
+ "3 0 3 8 21 \n",
+ "4 0 4 8 22 \n",
+ "\n",
+ " awake_probability rem_probability light_probability deep_probability \n",
+ "0 1.0 0.0 0.0 0.0 \n",
+ "1 1.0 0.0 0.0 0.0 \n",
+ "2 1.0 0.0 0.0 0.0 \n",
+ "3 1.0 0.0 0.0 0.0 \n",
+ "4 1.0 0.0 0.0 0.0 \n",
+ "<class 'pandas.core.frame.DataFrame'>\n",
+ "RangeIndex: 551042 entries, 0 to 551041\n",
+ "Data columns (total 8 columns):\n",
+ " # Column Non-Null Count Dtype \n",
+ "--- ------ -------------- ----- \n",
+ " 0 sleep_id 551042 non-null int64 \n",
+ " 1 minutes_since_begin 551042 non-null int64 \n",
+ " 2 stage_start_hour 551042 non-null int64 \n",
+ " 3 stage_start_minute 551042 non-null int64 \n",
+ " 4 awake_probability 551042 non-null float64\n",
+ " 5 rem_probability 551042 non-null float64\n",
+ " 6 light_probability 551042 non-null float64\n",
+ " 7 deep_probability 551042 non-null float64\n",
+ "dtypes: float64(4), int64(4)\n",
+ "memory usage: 33.6 MB\n",
+ "None\n"
+ ]
+ }
+ ],
+ "source": [
+ "print(sleep_data.head())\n",
+ "print(sleep_data.info())"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Model Development"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Helper functions and class"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 14,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "TEST_SIZE = 122\n",
+ "VALIDATION_SIZE = 183\n",
+ "TIME_STEP_INPUT = 10 # in minutes\n",
+ "BATCH_SIZE = 32"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 15,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "def training_test_split_by_unique_index(data, index: str, test_size: int = 10):\n",
+ " test_ids = np.random.choice(data[index].unique(), size = test_size, replace=False)\n",
+ " return data[~data[index].isin(test_ids)], data[data[index].isin(test_ids)]"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 16,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# Adapted from https://www.tensorflow.org/tutorials/structured_data/time_series\n",
+ "class WindowGenerator():\n",
+ " def __init__(self, data, index: str = \"sleep_id\", input_width: int = TIME_STEP_INPUT, validation_size: int = VALIDATION_SIZE, test_size: int = TEST_SIZE, generate_data_now: bool = True):\n",
+ " # Partition data\n",
+ " self.training, self.testing = training_test_split_by_unique_index(sleep_data, index, test_size)\n",
+ " self.training, self.validation = training_test_split_by_unique_index(self.training, index, validation_size)\n",
+ "\n",
+ " # Window paramters\n",
+ " self.input_width = input_width\n",
+ " self.label_width = 1\n",
+ " self.shift = 1\n",
+ "\n",
+ " self.total_window_size = self.input_width + self.shift\n",
+ "\n",
+ " self.input_slice = slice(0, input_width)\n",
+ " self.input_indices = np.arange(self.total_window_size)[self.input_slice]\n",
+ "\n",
+ " self.label_start = self.total_window_size - self.label_width\n",
+ " self.labels_slice = slice(self.label_start, None)\n",
+ " self.label_indices = np.arange(self.total_window_size)[self.labels_slice]\n",
+ " \n",
+ " if generate_data_now:\n",
+ " self.training_ds = self.make_dataset(self.training)\n",
+ " self.validation_ds = self.make_dataset(self.validation)\n",
+ " self.testing_ds = self.make_dataset(self.testing)\n",
+ "\n",
+ "\n",
+ " def __repr__(self):\n",
+ " return \"WindowGenerator:\\n\\t\" +'\\n\\t'.join([\n",
+ " f'Total window size: {self.total_window_size}',\n",
+ " f'Input indices: {self.input_indices}',\n",
+ " f'Label indices: {self.label_indices}',\n",
+ " ])\n",
+ "\n",
+ " def split_window(self, features):\n",
+ " inputs = features[:, self.input_slice, :]\n",
+ " labels = features[:, self.labels_slice, :]\n",
+ " inputs.set_shape([None, self.input_width, None])\n",
+ " labels.set_shape([None, self.label_width, None])\n",
+ "\n",
+ " return inputs, labels\n",
+ "\n",
+ " def make_dataset(self, data, index_group: str = \"sleep_id\", sort_by: str = \"minutes_since_begin\"):\n",
+ " ds_all = None\n",
+ " for i_group in data[index_group].unique():\n",
+ " subset_data = np.array(data[data[index_group] == i_group].sort_values(by=[sort_by]), dtype=np.float32)\n",
+ " ds = tf.keras.utils.timeseries_dataset_from_array(\n",
+ " data=subset_data,\n",
+ " targets=None,\n",
+ " sequence_length=self.total_window_size,\n",
+ " sequence_stride=1,\n",
+ " shuffle=False,\n",
+ " batch_size=BATCH_SIZE,)\n",
+ " ds_all = ds if ds_all is None else ds_all.concatenate(ds)\n",
+ " ds_all = ds_all.map(self.split_window)\n",
+ "\n",
+ " return ds_all\n",
+ "\n",
+ " # def generate_all_datasets(self):\n",
+ " # self._training_ds = self.make_dataset(self.training)\n",
+ " # self._validation_ds = self.make_dataset(self.validation)\n",
+ " # self._testing_ds = self.make_dataset(self.testing)\n",
+ "\n",
+ " # def training_dataset(self):\n",
+ " # if self._training_ds is None:\n",
+ " # self._training_ds = self.make_dataset(self.training)\n",
+ " # return self._training_ds\n",
+ "\n",
+ " # def validation_dataset(self):\n",
+ " # if self._validation_ds is None:\n",
+ " # self._validation_ds = self.make_dataset(self.validation)\n",
+ " # return self._validation_ds\n",
+ "\n",
+ " # def test_dataset(self):\n",
+ " # if self._testing_ds is None:\n",
+ " # self._testing_ds = self.make_dataset(self.testing)\n",
+ " # return self._testing_ds"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 17,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "2022-05-11 00:38:01.423873: I tensorflow/core/platform/cpu_feature_guard.cc:151] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2 FMA\n",
+ "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n"
+ ]
+ },
{
"data": {
"text/plain": [
- "start 2019-02-11 06:11:00+00:00\n",
- "duration 60\n",
- "end 2019-02-11 06:12:00+00:00\n",
- "stage 0\n",
- "awake_probability 0.0\n",
- "light_probability 0.0\n",
- "deep_probability 0.0\n",
- "rem_probability 0.0\n",
- "dtype: object"
+ "WindowGenerator:\n",
+ "\tTotal window size: 11\n",
+ "\tInput indices: [0 1 2 3 4 5 6 7 8 9]\n",
+ "\tLabel indices: [10]"
]
},
- "execution_count": 143,
+ "execution_count": 17,
"metadata": {},
"output_type": "execute_result"
}
],
+ "source": [
+ "wg = WindowGenerator(sleep_data)\n",
+ "wg"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 20,
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/plain": [
+ "1836"
+ ]
+ },
+ "execution_count": 20,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "len(wg.testing_ds)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Data Prep\n",
+ "\n",
+ "All inputs follow: (batch_size, timesteps, input_dim)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "metadata": {},
+ "outputs": [
+ {
+ "ename": "NameError",
+ "evalue": "name 'training_test_split_by_unique_index' is not defined",
+ "output_type": "error",
+ "traceback": [
+ "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+ "\u001b[0;31mNameError\u001b[0m Traceback (most recent call last)",
+ "\u001b[1;32m/Users/nowadmin/Documents/School Folder/CS 437/Lab/Final Project/tf_model.ipynb Cell 32'\u001b[0m in \u001b[0;36m<cell line: 1>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> <a href='vscode-notebook-cell:/Users/nowadmin/Documents/School%20Folder/CS%20437/Lab/Final%20Project/tf_model.ipynb#ch0000068?line=0'>1</a>\u001b[0m training_sleep_data, test_sleep_data \u001b[39m=\u001b[39m training_test_split_by_unique_index(sleep_data, \u001b[39m\"\u001b[39m\u001b[39msleep_id\u001b[39m\u001b[39m\"\u001b[39m, \u001b[39m5\u001b[39m)\n",
+ "\u001b[0;31mNameError\u001b[0m: name 'training_test_split_by_unique_index' is not defined"
+ ]
+ }
+ ],
+ "source": [
+ "training_sleep_data, test_sleep_data = training_test_split_by_unique_index(sleep_data, \"sleep_id\", 5)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 288,
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/plain": [
+ "set()"
+ ]
+ },
+ "execution_count": 288,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "set(training.sleep_id.unique()).intersection(set(test.sleep_id.unique()))"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 239,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "sleep_data_tensor = tf.convert_to_tensor(sleep_data)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 240,
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/plain": [
+ "TensorShape([551042, 7])"
+ ]
+ },
+ "execution_count": 240,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "sleep_data_tensor.shape"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 241,
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/plain": [
+ "<tf.Tensor: shape=(7,), dtype=float64, numpy=array([ 1., 8., 19., 1., 0., 0., 0.])>"
+ ]
+ },
+ "execution_count": 241,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "sleep_data_tensor[1]"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Model 1: LSTM"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 215,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# Hyper-parameters\n",
+ "LSTM_UNITS = 16\n",
+ "LSTM_LEARNING_RATE = 0.0001"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 278,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Model: \"sequential_12\"\n",
+ "_________________________________________________________________\n",
+ " Layer (type) Output Shape Param # \n",
+ "=================================================================\n",
+ " lstm_14 (LSTM) (None, 10, 16) 1600 \n",
+ " \n",
+ " dense_9 (Dense) (None, 10, 4) 68 \n",
+ " \n",
+ "=================================================================\n",
+ "Total params: 1,668\n",
+ "Trainable params: 1,668\n",
+ "Non-trainable params: 0\n",
+ "_________________________________________________________________\n",
+ "None\n"
+ ]
+ }
+ ],
+ "source": [
+ "# Model Definition\n",
+ "lstm_model = keras.Sequential()\n",
+ "lstm_model.add(layers.Input(shape=(TIME_STEP_INPUT, 8)))\n",
+ "lstm_model.add(layers.LSTM(LSTM_UNITS, stateful=False, return_sequences=True))\n",
+ "# lstm_model.add(layers.LSTM(LSTM_UNITS, stateful=False, return_sequences=True))\n",
+ "lstm_model.add(layers.Dense(SLEEP_STAGES))\n",
+ "lstm_model.build()\n",
+ "print(lstm_model.summary())"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# Model Training\n",
+ "lstm_loss = keras.losses.SparseCategoricalCrossentropy(from_logits=True)\n",
+ "lstm_optm = keras.optimizers.Adam(learning_rate=LSTM_LEARNING_RATE)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "### Model 2: GRU"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# Hyper-paramters\n",
+ "GRU_UNITS = 16"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "gru_model = keras.Sequential([\n",
+ " layers.GRU(GRU_UNITS),\n",
+ " layers.Dense(SLEEP_STAGES)\n",
+ "])\n",
+ "gru_model.add(layers.Embedding(input_dim=1000, output_dim=64))\n",
+ "print(gru_model.summary())"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
"source": []
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
- "## Model 1: GRU"
+ "### Model 3: Attention Mechanism"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "ATTENTION_UNITS = 16"
]
},
{
@@ -571,15 +1059,32 @@
"metadata": {},
"outputs": [],
"source": [
- "model = keras.Sequential()\n",
- "model.add(layers.Embedding(input_dim=1000, output_dim=64))\n",
- "model.add(layers.GRU(128))\n",
- "model.add(layers.Dense(10))"
+ "am_model = keras.Sequential([\n",
+ " Attention(ATTENTION_UNITS)\n",
+ " layers.Dense(SLEEP_STAGES)\n",
+ "])\n",
+ "print(am_model.summary())"
]
},
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": []
+ },
{
"cell_type": "markdown",
"metadata": {},
+ "source": [
+ "### Model Head-to-Head testing"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
"source": []
},
{