Updated Notebook comments
This commit is contained in:
Radu C. Martin
parent
12c879f016
commit
633c4d12d3
5 changed files with 216 additions and 2413 deletions
|
|
@ -4,7 +4,7 @@
|
|||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"# Bayesian Optimisation of starting Gaussian Process hyperparameters"
|
||||
"# Gaussian Process Model Training and Performance Evaluation"
|
||||
]
|
||||
},
|
||||
{
|
||||
|
|
@ -213,7 +213,7 @@
|
|||
"id": "0aba0df5-b0e3-4738-bb61-1dad869d1ea3"
|
||||
},
|
||||
"source": [
|
||||
"## Load previously exported data"
|
||||
"## Load previously exported CARNOT 'experimental' data"
|
||||
]
|
||||
},
|
||||
{
|
||||
|
|
@ -226,6 +226,13 @@
|
|||
"dfs_test = []"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Separate into training and testing data sets:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 16,
|
||||
|
|
@ -249,6 +256,13 @@
|
|||
" dfs_test.append(pd.read_csv(f\"../Data/Good_CARNOT/{exp}_table.csv\").rename(columns = {'Power': 'SimulatedHeat'}))"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Separate columns into exogenous inputs, controlled inputs and outputs:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 18,
|
||||
|
|
@ -262,6 +276,13 @@
|
|||
"y_cols = ['SimulatedTemp']"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Impose the autoregressive lags for each input group:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 19,
|
||||
|
|
@ -511,6 +532,13 @@
|
|||
" return df_gpr"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Merge all the training dataframes:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 28,
|
||||
|
|
@ -667,13 +695,19 @@
|
|||
"df_gpr_train.head()"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Select all points in the training dataset:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 29,
|
||||
"execution_count": 1,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"train_dataset_size = 15 * 96\n",
|
||||
"train_dataset_size = -1"
|
||||
]
|
||||
},
|
||||
|
|
@ -1401,6 +1435,13 @@
|
|||
"y_range = np.arange(1,6)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Iterate over all combination of lags and compute for each the RMSE, SMSE, LPD and MSLL errors:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
|
|
@ -1545,6 +1586,13 @@
|
|||
"## Multistep prediction"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Select the dataset which will be used for multistep prediction:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 54,
|
||||
|
|
@ -1556,6 +1604,13 @@
|
|||
"df_output = dfs_gpr_test[test_dataset_idx][dict_cols['y'][1]]"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Select the starting index in the test dataset and the number of consecutive points to simulate:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 55,
|
||||
|
|
@ -1617,26 +1672,6 @@
|
|||
"plt.title(f\"Multi step prediction over {N_pred} steps for Test dataset {test_dataset_idx}\")\n",
|
||||
"plt.savefig(f\"../Thesis/Plots/GP_{w_lags}{u_lags}{y_lags}_{train_dataset_size}pts_test_prediction_{N_pred}_steps.pdf\")"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 141,
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"data": {
|
||||
"text/plain": [
|
||||
"TensorShape([2612, 7])"
|
||||
]
|
||||
},
|
||||
"execution_count": 141,
|
||||
"metadata": {},
|
||||
"output_type": "execute_result"
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"m.data[0].shape"
|
||||
]
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
|
|
|
|||
|
|
@ -4,7 +4,7 @@
|
|||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"# Bayesian Optimisation of starting Gaussian Process hyperparameters"
|
||||
"# Sparse and Variational Gaussian Process Model Training and Performance Evaluation"
|
||||
]
|
||||
},
|
||||
{
|
||||
|
|
@ -78,9 +78,6 @@
|
|||
"cell_type": "markdown",
|
||||
"metadata": {
|
||||
"id": "90fdac33-eed4-4ab4-b2b1-de0f1f27727b",
|
||||
"jupyter": {
|
||||
"source_hidden": true
|
||||
},
|
||||
"tags": []
|
||||
},
|
||||
"source": [
|
||||
|
|
@ -199,7 +196,7 @@
|
|||
"id": "0aba0df5-b0e3-4738-bb61-1dad869d1ea3"
|
||||
},
|
||||
"source": [
|
||||
"## Load previously exported data"
|
||||
"## Load previously exported CARNOT 'experimental' data"
|
||||
]
|
||||
},
|
||||
{
|
||||
|
|
@ -212,6 +209,13 @@
|
|||
"dfs_test = []"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Separate into training and testing data sets:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 13,
|
||||
|
|
@ -235,6 +239,13 @@
|
|||
" dfs_test.append(pd.read_csv(f\"../Data/Good_CARNOT/{exp}_table.csv\").rename(columns = {'Power': 'SimulatedHeat'}))"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Separate columns into exogenous inputs, controlled inputs and outputs:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 15,
|
||||
|
|
@ -248,6 +259,13 @@
|
|||
"y_cols = ['SimulatedTemp']"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Impose the autoregressive lags for each input group:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 16,
|
||||
|
|
@ -497,6 +515,13 @@
|
|||
" return df_gpr"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Merge all the training dataframes:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 25,
|
||||
|
|
@ -647,15 +672,6 @@
|
|||
"df_gpr_train.head()"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 26,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"#df_gpr_train = df_gpr_train.sample(n = 500)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 27,
|
||||
|
|
@ -1245,6 +1261,13 @@
|
|||
"y_lags = 5"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Iterate over all combination of lags and compute for each the RMSE, SMSE, LPD and MSLL errors:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 54,
|
||||
|
|
@ -1470,6 +1493,13 @@
|
|||
"## Multistep prediction"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Select the dataset which will be used for multistep prediction:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 47,
|
||||
|
|
@ -1481,6 +1511,13 @@
|
|||
"df_output = dfs_gpr_test[test_dataset_idx][dict_cols['y'][1]]"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Select the starting index in the test dataset and the number of consecutive points to simulate:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 48,
|
||||
|
|
|
|||
|
|
@ -75,6 +75,13 @@
|
|||
"## GP model"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Define the underlying function that will be modeled:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 8,
|
||||
|
|
@ -108,6 +115,13 @@
|
|||
"})"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Sample `n` points and add measurement noise:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 11,
|
||||
|
|
@ -142,6 +156,13 @@
|
|||
"Y_sampled = Y_sampled + noise"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Plot the function and the sampled points:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 14,
|
||||
|
|
@ -167,6 +188,13 @@
|
|||
"plt.show()"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"### Train the GP"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 15,
|
||||
|
|
@ -362,6 +390,13 @@
|
|||
"## CasADi part"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Define the CasADi callback:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 23,
|
||||
|
|
@ -1074,6 +1109,13 @@
|
|||
"grads"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Define another casadi callback, which directly uses the GPflow gradients:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 29,
|
||||
|
|
|
|||
File diff suppressed because one or more lines are too long
|
|
@ -169,6 +169,14 @@
|
|||
"### Compute the time index"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "0a39bb25-f673-488b-99a4-57ee0d789f54",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"The time index is computed by adding the elapsed time (`sample nr.` * `Tsample`) to the dataset start time. Since the CARNOT weather set represents the year 2010, the starting time is taken as 2010-01-01 at midnight."
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 14,
|
||||
|
|
@ -278,6 +286,14 @@
|
|||
"### Get reference temperature"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "8b53a7c0-966f-4c75-8857-1a96ec3135b0",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"The reference temperature is computed according to the SIA norm, as a function of the last 48h of outside temperature (at a `Tsample` of 15min this comes out to 2 * 96 sample points)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 16,
|
||||
|
|
@ -292,6 +308,14 @@
|
|||
"df_tref = df_tref.shift(1) # The reference at time t is computed using info up to t-1"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "5d2e03c9-044a-454b-8a6b-c1a2a6adf6cc",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Compute mean value and standard deviation of tracking error:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 17,
|
||||
|
|
@ -358,6 +382,14 @@
|
|||
"}"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "b0fc6fbc-ed75-4724-a5cb-8896a89f5039",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Scale the dataset and compute the autoregressive inputs that are passed to the GP:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 20,
|
||||
|
|
@ -548,6 +580,14 @@
|
|||
"df_output = df_gpr[dict_cols['y'][1]]"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "ddbfd08f-7d5d-4f2e-b43e-a282a0503931",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Load the only trained model in the GP case, and the first trained model in the SVGP case:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 23,
|
||||
|
|
@ -571,6 +611,14 @@
|
|||
" m = model"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "c18135f0-06d8-44e3-b86c-4a84b2508f79",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Plot the multistep prediction performance for 25 consecutive points:"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 25,
|
||||
|
|
@ -1091,6 +1139,14 @@
|
|||
"output_notebook()"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "3ec059bc-9ce6-4271-b5cd-5003d1330a0b",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"### Plot evolution of reference/measured temperature"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 22,
|
||||
|
|
@ -1393,6 +1449,14 @@
|
|||
"plt.savefig(f\"../Thesis/Plots/{sim_id}_abserr.pdf\", bbox_inches='tight')"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "4b0595d2-59d7-4536-b8a3-8c8aced19dc4",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"### Plot evolution of hyperparameters"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 29,
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue