Skip to content

feat: add quantiles to forecast and cross validation methods #242

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
Mar 7, 2024
Merged

feat: add quantiles to forecast and cross validation methods #242

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
a076bff
feat: add quantiles argument to forecast and cv methods
AzulGarza Mar 7, 2024
313a464
feat: add distributed quantile
AzulGarza Mar 7, 2024
a6755dc
feat: add quantile forecasts tutorial
AzulGarza Mar 7, 2024
e424b52
fix: add correct test pred intervals test
AzulGarza Mar 7, 2024
f722bc1
feat: add cross validation to quantile forecasts
AzulGarza Mar 7, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
108 changes: 103 additions & 5 deletions nbs/distributed.timegpt.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -160,6 +160,7 @@
" target_col: str = 'y',\n",
" X_df: Optional[fugue.AnyDataFrame] = None,\n",
" level: Optional[List[Union[int, float]]] = None,\n",
" quantiles: Optional[List[float]] = None,\n",
" finetune_steps: int = 0,\n",
" finetune_loss: str = 'default',\n",
" clean_ex_first: bool = True,\n",
Expand All @@ -177,6 +178,7 @@
" time_col=time_col,\n",
" target_col=target_col,\n",
" level=level,\n",
" quantiles=quantiles,\n",
" finetune_steps=finetune_steps,\n",
" finetune_loss=finetune_loss,\n",
" clean_ex_first=clean_ex_first,\n",
Expand All @@ -186,7 +188,12 @@
" date_features_to_one_hot=date_features_to_one_hot,\n",
" model=model,\n",
" )\n",
" schema = self._get_forecast_schema(id_col=id_col, time_col=time_col, level=level)\n",
" schema = self._get_forecast_schema(\n",
" id_col=id_col, \n",
" time_col=time_col, \n",
" level=level,\n",
" quantiles=quantiles,\n",
" )\n",
" fcst_df = self._distribute_method(\n",
" method=self._forecast if X_df is None else self._forecast_x,\n",
" df=df,\n",
Expand Down Expand Up @@ -246,6 +253,7 @@
" time_col: str = 'ds',\n",
" target_col: str = 'y',\n",
" level: Optional[List[Union[int, float]]] = None,\n",
" quantiles: Optional[List[float]] = None,\n",
" finetune_steps: int = 0,\n",
" finetune_loss: str = 'default',\n",
" clean_ex_first: bool = True,\n",
Expand All @@ -264,6 +272,7 @@
" time_col=time_col,\n",
" target_col=target_col,\n",
" level=level,\n",
" quantiles=quantiles,\n",
" finetune_steps=finetune_steps,\n",
" finetune_loss=finetune_loss,\n",
" clean_ex_first=clean_ex_first,\n",
Expand All @@ -274,7 +283,13 @@
" n_windows=n_windows,\n",
" step_size=step_size,\n",
" )\n",
" schema = self._get_forecast_schema(id_col=id_col, time_col=time_col, level=level, cv=True)\n",
" schema = self._get_forecast_schema(\n",
" id_col=id_col, \n",
" time_col=time_col, \n",
" level=level,\n",
" quantiles=quantiles,\n",
" cv=True,\n",
" )\n",
" fcst_df = self._distribute_method(\n",
" method=self._cross_validation,\n",
" df=df,\n",
Expand Down Expand Up @@ -330,15 +345,23 @@
" return timegpt._cross_validation(df=df, **kwargs)\n",
" \n",
" @staticmethod\n",
" def _get_forecast_schema(id_col, time_col, level, cv=False):\n",
" def _get_forecast_schema(id_col, time_col, level, quantiles, cv=False):\n",
" schema = f'{id_col}:string,{time_col}:datetime'\n",
" if cv:\n",
" schema = f'{schema},cutoff:datetime'\n",
" schema = f'{schema},TimeGPT:double'\n",
" if (level is not None) and (quantiles is not None):\n",
" raise Exception(\n",
" \"you should include `level` or `quantiles` but not both.\"\n",
" )\n",
" if level is not None:\n",
" level = sorted(level)\n",
" schema = f'{schema},{\",\".join([f\"TimeGPT-lo-{lv}:double\" for lv in reversed(level)])}'\n",
" schema = f'{schema},{\",\".join([f\"TimeGPT-hi-{lv}:double\" for lv in level])}'\n",
" if quantiles is not None:\n",
" quantiles = sorted(quantiles)\n",
" q_cols = [f'TimeGPT-q-{int(q * 100)}:double' for q in quantiles]\n",
" schema = f'{schema},{\",\".join(q_cols)}'\n",
" return Schema(schema)\n",
" \n",
" @staticmethod\n",
Expand Down Expand Up @@ -798,6 +821,34 @@
" #test_anomalies_same_results_num_partitions(df, id_col=id_col, time_col=time_col, date_features=True, clean_ex_first=False)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"#| hide\n",
"def test_quantiles(df: fugue.AnyDataFrame, id_col: str = 'id_col', time_col: str = 'time_col'):\n",
" test_qls = list(np.arange(0.1, 1, 0.1))\n",
" exp_q_cols = [f\"TimeGPT-q-{int(q * 100)}\" for q in test_qls]\n",
" def test_method_qls(method, **kwargs):\n",
" df_qls = method(\n",
" df=df, \n",
" h=12, \n",
" id_col=id_col,\n",
" time_col=time_col, \n",
" quantiles=test_qls,\n",
" **kwargs\n",
" )\n",
" df_qls = fa.as_pandas(df_qls)\n",
" assert all(col in df_qls.columns for col in exp_q_cols)\n",
" # test monotonicity of quantiles\n",
" df_qls.apply(lambda x: x.is_monotonic_increasing, axis=1).sum() == len(exp_q_cols)\n",
" test_method_qls(distributed_timegpt.forecast)\n",
" test_method_qls(distributed_timegpt.forecast, add_history=True)\n",
" test_method_qls(distributed_timegpt.cross_validation)"
]
},
{
"cell_type": "code",
"execution_count": null,
Expand Down Expand Up @@ -867,6 +918,16 @@
"spark_diff_cols_df = spark.createDataFrame(series_diff_cols)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"#| hide\n",
"test_quantiles(spark_df, id_col=\"unique_id\", time_col=\"ds\")"
]
},
{
"cell_type": "code",
"execution_count": null,
Expand Down Expand Up @@ -941,7 +1002,26 @@
"source": [
"#| hide\n",
"dask_df = dd.from_pandas(series, npartitions=2)\n",
"dask_diff_cols_df = dd.from_pandas(series_diff_cols, npartitions=2)\n",
"dask_diff_cols_df = dd.from_pandas(series_diff_cols, npartitions=2)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"#| hide\n",
"test_quantiles(dask_df, id_col=\"unique_id\", time_col=\"ds\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"#| hide\n",
"test_forecast_dataframe(dask_df)\n",
"test_forecast_dataframe_diff_cols(dask_diff_cols_df)\n",
"test_anomalies_dataframe(dask_df)\n",
Expand Down Expand Up @@ -1007,7 +1087,25 @@
"# add mock node to simulate a cluster\n",
"mock_node = ray_cluster.add_node(num_cpus=2)\n",
"ray_df = ray.data.from_pandas(series)\n",
"ray_diff_cols_df = ray.data.from_pandas(series_diff_cols)\n",
"ray_diff_cols_df = ray.data.from_pandas(series_diff_cols)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"test_quantiles(ray_df, id_col=\"unique_id\", time_col=\"ds\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"#| hide\n",
"test_forecast_dataframe(ray_df)\n",
"test_forecast_dataframe_diff_cols(ray_diff_cols_df)\n",
"test_anomalies_dataframe(ray_df)\n",
Expand Down
Loading