"<a href=\"https://colab.research.google.com/github/microsoft/autogen/blob/main/notebook/agentchat_groupchat_RAG.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"# Auto Generated Agent Chat: Group Chat with Retrieval Augmented Generation\n",
"\n",
"AutoGen supports conversable agents powered by LLMs, tools or humans, performing tasks collectively via automated chat. This framework allows tool use and human participation through multi-agent conversation.\n",
"Please find documentation about this feature [here](https://microsoft.github.io/autogen/docs/Use-Cases/agent_chat).\n",
"\n",
"## Requirements\n",
"\n",
"AutoGen requires `Python>=3.8`. To run this notebook example, please install:\n",
"```bash\n",
"pip install pyautogen\n",
"```"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [],
"source": [
"%%capture --no-stderr\n",
"# %pip install pyautogen[retrievechat]~=0.1.11"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"## Set your API Endpoint\n",
"\n",
"The [`config_list_from_json`](https://microsoft.github.io/autogen/docs/reference/oai/openai_utils#config_list_from_json) function loads a list of configurations from an environment variable or a json file."
"print(\"LLM models: \", [config_list[i][\"model\"] for i in range(len(config_list))])"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"It first looks for environment variable \"OAI_CONFIG_LIST\" which needs to be a valid json string. If that variable is not found, it then looks for a json file named \"OAI_CONFIG_LIST\". It filters the configs by models (you can filter by other keys as well).\n",
"\n",
"The config list looks like the following:\n",
"```python\n",
"config_list = [\n",
" {\n",
" \"model\": \"gpt-4\",\n",
" \"api_key\": \"<your OpenAI API key>\",\n",
" }, # OpenAI API endpoint for gpt-4\n",
" {\n",
" \"model\": \"gpt-35-turbo-0631\", # 0631 or newer is needed to use functions\n",
" boss_aid.human_input_mode = \"NEVER\" # Disable human input for boss_aid since it only retrieves content.\n",
" \n",
" llm_config = {\n",
" \"functions\": [\n",
" {\n",
" \"name\": \"retrieve_content\",\n",
" \"description\": \"retrieve content for code generation and question answering.\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"message\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"Refined message which keeps the original meaning and can be used to retrieve content for code generation and question answering.\",\n",
" # Start chatting with boss as this is the user proxy agent.\n",
" boss.initiate_chat(\n",
" manager,\n",
" message=PROBLEM,\n",
" )"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"## Start Chat\n",
"\n",
"### UserProxyAgent doesn't get the correct code\n",
"[FLAML](https://github.com/microsoft/FLAML) was open sourced in 2020, so ChatGPT is familiar with it. However, Spark-related APIs were added in 2022, so they were not in ChatGPT's training data. As a result, we end up with invalid code."
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"\u001b[33mBoss\u001b[0m (to chat_manager):\n",
"\n",
"How to use spark for parallel training in FLAML? Give me sample code.\n",
"In this code, we first load the credit dataset. Then, we define the search space for the hyperparameters. We create an `AutoML` instance with `SparkTrials` as the `trials` parameter. We set the `parallelism` parameter to 2 to use 2 Spark workers for parallel training. Finally, we fit the model and evaluate it.\n",
"### RetrieveUserProxyAgent get the correct code\n",
"Since RetrieveUserProxyAgent can perform retrieval-augmented generation based on the given documentation file, ChatGPT can generate the correct code for us!"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Trying to create collection.\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"INFO:autogen.retrieve_utils:Found 2 chunks.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"doc_ids: [['doc_0', 'doc_1', 'doc_4']]\n",
"\u001b[32mAdding doc_id doc_0 to context.\u001b[0m\n",
"\u001b[32mAdding doc_id doc_1 to context.\u001b[0m\n",
"\u001b[32mAdding doc_id doc_4 to context.\u001b[0m\n",
"You're a retrieve augmented coding assistant. You answer user's questions based on your own knowledge and the\n",
"context provided by the user.\n",
"If you can't answer the question with or without the current context, you should reply exactly `UPDATE CONTEXT`.\n",
"For code generation, you must obey the following rules:\n",
"Rule 1. You MUST NOT install any packages because all the packages needed are already installed.\n",
"Rule 2. You must follow the formats below to write your code:\n",
"```language\n",
"# your code\n",
"```\n",
"\n",
"User's question is: How to use spark for parallel training in FLAML? Give me sample code.\n",
"\n",
"Context is: # Integrate - Spark\n",
"\n",
"FLAML has integrated Spark for distributed training. There are two main aspects of integration with Spark:\n",
"- Use Spark ML estimators for AutoML.\n",
"- Use Spark to run training in parallel spark jobs.\n",
"\n",
"## Spark ML Estimators\n",
"\n",
"FLAML integrates estimators based on Spark ML models. These models are trained in parallel using Spark, so we called them Spark estimators. To use these models, you first need to organize your data in the required format.\n",
"\n",
"### Data\n",
"\n",
"For Spark estimators, AutoML only consumes Spark data. FLAML provides a convenient function `to_pandas_on_spark` in the `flaml.automl.spark.utils` module to convert your data into a pandas-on-spark (`pyspark.pandas`) dataframe/series, which Spark estimators require.\n",
"\n",
"This utility function takes data in the form of a `pandas.Dataframe` or `pyspark.sql.Dataframe` and converts it into a pandas-on-spark dataframe. It also takes `pandas.Series` or `pyspark.sql.Dataframe` and converts it into a [pandas-on-spark](https://spark.apache.org/docs/latest/api/python/user_guide/pandas_on_spark/index.html) series. If you pass in a `pyspark.pandas.Dataframe`, it will not make any changes.\n",
"\n",
"This function also accepts optional arguments `index_col` and `default_index_type`.\n",
"- `index_col` is the column name to use as the index, default is None.\n",
"- `default_index_type` is the default index type, default is \"distributed-sequence\". More info about default index type could be found on Spark official [documentation](https://spark.apache.org/docs/latest/api/python/user_guide/pandas_on_spark/options.html#default-index-type)\n",
"\n",
"Here is an example code snippet for Spark Data:\n",
"To use Spark ML models you need to format your data appropriately. Specifically, use [`VectorAssembler`](https://spark.apache.org/docs/latest/api/python/reference/api/pyspark.ml.feature.VectorAssembler.html) to merge all feature columns into a single vector column.\n",
"Later in conducting the experiment, use your pandas-on-spark data like non-spark data and pass them using `X_train, y_train` or `dataframe, label`.\n",
"\n",
"### Estimators\n",
"#### Model List\n",
"- `lgbm_spark`: The class for fine-tuning Spark version LightGBM models, using [SynapseML](https://microsoft.github.io/SynapseML/docs/features/lightgbm/about/) API.\n",
"\n",
"#### Usage\n",
"First, prepare your data in the required format as described in the previous section.\n",
"\n",
"By including the models you intend to try in the `estimators_list` argument to `flaml.automl`, FLAML will start trying configurations for these models. If your input is Spark data, FLAML will also use estimators with the `_spark` postfix by default, even if you haven't specified them.\n",
"\n",
"Here is an example code snippet using SparkML models in AutoML:\n",
"\n",
"```python\n",
"import flaml\n",
"# prepare your data in pandas-on-spark format as we previously mentioned\n",
"\n",
"automl = flaml.AutoML()\n",
"settings = {\n",
" \"time_budget\": 30,\n",
" \"metric\": \"r2\",\n",
" \"estimator_list\": [\"lgbm_spark\"], # this setting is optional\n",
" \"task\": \"regression\",\n",
"}\n",
"\n",
"automl.fit(\n",
" dataframe=psdf,\n",
" label=label,\n",
" **settings,\n",
")\n",
"```\n",
"\n",
"\n",
"[Link to notebook](https://github.com/microsoft/FLAML/blob/main/notebook/automl_bankrupt_synapseml.ipynb) | [Open in colab](https://colab.research.google.com/github/microsoft/FLAML/blob/main/notebook/automl_bankrupt_synapseml.ipynb)\n",
"\n",
"## Parallel Spark Jobs\n",
"You can activate Spark as the parallel backend during parallel tuning in both [AutoML](/docs/Use-Cases/Task-Oriented-AutoML#parallel-tuning) and [Hyperparameter Tuning](/docs/Use-Cases/Tune-User-Defined-Function#parallel-tuning), by setting the `use_spark` to `true`. FLAML will dispatch your job to the distributed Spark backend using [`joblib-spark`](https://github.com/joblib/joblib-spark).\n",
"\n",
"Please note that you should not set `use_spark` to `true` when applying AutoML and Tuning for Spark Data. This is because only SparkML models will be used for Spark Data in AutoML and Tuning. As SparkML models run in parallel, there is no need to distribute them with `use_spark` again.\n",
"\n",
"All the Spark-related arguments are stated below. These arguments are available in both Hyperparameter Tuning and AutoML:\n",
"\n",
"\n",
"- `use_spark`: boolean, default=False | Whether to use spark to run the training in parallel spark jobs. This can be used to accelerate training on large models and large datasets, but will incur more overhead in time and thus slow down training in some cases. GPU training is not supported yet when use_spark is True. For Spark clusters, by default, we will launch one trial per executor. However, sometimes we want to launch more trials than the number of executors (e.g., local mode). In this case, we can set the environment variable `FLAML_MAX_CONCURRENT` to override the detected `num_executors`. The final number of concurrent trials will be the minimum of `n_concurrent_trials` and `num_executors`.\n",
"- `n_concurrent_trials`: int, default=1 | The number of concurrent trials. When n_concurrent_trials > 1, FLAML performes parallel tuning.\n",
"- `force_cancel`: boolean, default=False | Whether to forcely cancel Spark jobs if the search time exceeded the time budget. Spark jobs include parallel tuning jobs and Spark-based model training jobs.\n",
"\n",
"An example code snippet for using parallel Spark jobs:\n",
"```python\n",
"import flaml\n",
"automl_experiment = flaml.AutoML()\n",
"automl_settings = {\n",
" \"time_budget\": 30,\n",
" \"metric\": \"r2\",\n",
" \"task\": \"regression\",\n",
" \"n_concurrent_trials\": 2,\n",
" \"use_spark\": True,\n",
" \"force_cancel\": True, # Activating the force_cancel option can immediately halt Spark jobs once they exceed the allocated time_budget.\n",
"}\n",
"\n",
"automl.fit(\n",
" dataframe=dataframe,\n",
" label=label,\n",
" **automl_settings,\n",
")\n",
"```\n",
"\n",
"\n",
"[Link to notebook](https://github.com/microsoft/FLAML/blob/main/notebook/integrate_spark.ipynb) | [Open in colab](https://colab.research.google.com/github/microsoft/FLAML/blob/main/notebook/integrate_spark.ipynb)\n",
"Artifacts can differ among various machine learning libraries. More detailed information can be found [here](https://mlflow.org/docs/latest/tracking.html#automatic-logging).\n",
"\n",
"\n",
"\n",
"\n",
"## Plot Experiment Result\n",
"The `flaml.visualization` module provides utility functions for plotting the optimization process using [plotly](https://plotly.com/python/). Leveraging `plotly`, users can interactively explore experiment results. To use these plotting functions, simply provide your optimized `flaml.AutoML` or `flaml.tune.tune.ExperimentAnalysis` object as input. Optional parameters can be added using keyword arguments.\n",
"\n",
"Avaliable plotting functions:\n",
"- `plot_optimization_history`: Plot optimization history of all trials in the experiment.\n",
"- `plot_feature_importance`: Plot importance for each feature in the dataset.\n",
"- `plot_parallel_coordinate`: Plot the high-dimensional parameter relationships in the experiment.\n",
"- `plot_contour`: Plot the parameter relationship as contour plot in the experiment.\n",
"- `plot_edf`: Plot the objective value EDF (empirical distribution function) of the experiment.\n",
"- `plot_timeline`: Plot the timeline of the experiment.\n",
"- `plot_slice`: Plot the parameter relationship as slice plot in a study.\n",
"\n",
"### Figure Examples\n",
"\n",
"\n",
"Check out our example [notebook](../../notebook/trident/automl_plot.ipynb) for a preview of all interactive plots.\n",
"You're a retrieve augmented coding assistant. You answer user's questions based on your own knowledge and the\n",
"context provided by the user.\n",
"If you can't answer the question with or without the current context, you should reply exactly `UPDATE CONTEXT`.\n",
"For code generation, you must obey the following rules:\n",
"Rule 1. You MUST NOT install any packages because all the packages needed are already installed.\n",
"Rule 2. You must follow the formats below to write your code:\n",
"```language\n",
"# your code\n",
"```\n",
"\n",
"User's question is: How to use spark for parallel training in FLAML? Give me sample code.\n",
"\n",
"Context is: # Integrate - Spark\n",
"\n",
"FLAML has integrated Spark for distributed training. There are two main aspects of integration with Spark:\n",
"- Use Spark ML estimators for AutoML.\n",
"- Use Spark to run training in parallel spark jobs.\n",
"\n",
"## Spark ML Estimators\n",
"\n",
"FLAML integrates estimators based on Spark ML models. These models are trained in parallel using Spark, so we called them Spark estimators. To use these models, you first need to organize your data in the required format.\n",
"\n",
"### Data\n",
"\n",
"For Spark estimators, AutoML only consumes Spark data. FLAML provides a convenient function `to_pandas_on_spark` in the `flaml.automl.spark.utils` module to convert your data into a pandas-on-spark (`pyspark.pandas`) dataframe/series, which Spark estimators require.\n",
"\n",
"This utility function takes data in the form of a `pandas.Dataframe` or `pyspark.sql.Dataframe` and converts it into a pandas-on-spark dataframe. It also takes `pandas.Series` or `pyspark.sql.Dataframe` and converts it into a [pandas-on-spark](https://spark.apache.org/docs/latest/api/python/user_guide/pandas_on_spark/index.html) series. If you pass in a `pyspark.pandas.Dataframe`, it will not make any changes.\n",
"\n",
"This function also accepts optional arguments `index_col` and `default_index_type`.\n",
"- `index_col` is the column name to use as the index, default is None.\n",
"- `default_index_type` is the default index type, default is \"distributed-sequence\". More info about default index type could be found on Spark official [documentation](https://spark.apache.org/docs/latest/api/python/user_guide/pandas_on_spark/options.html#default-index-type)\n",
"\n",
"Here is an example code snippet for Spark Data:\n",
"To use Spark ML models you need to format your data appropriately. Specifically, use [`VectorAssembler`](https://spark.apache.org/docs/latest/api/python/reference/api/pyspark.ml.feature.VectorAssembler.html) to merge all feature columns into a single vector column.\n",
"Later in conducting the experiment, use your pandas-on-spark data like non-spark data and pass them using `X_train, y_train` or `dataframe, label`.\n",
"\n",
"### Estimators\n",
"#### Model List\n",
"- `lgbm_spark`: The class for fine-tuning Spark version LightGBM models, using [SynapseML](https://microsoft.github.io/SynapseML/docs/features/lightgbm/about/) API.\n",
"\n",
"#### Usage\n",
"First, prepare your data in the required format as described in the previous section.\n",
"\n",
"By including the models you intend to try in the `estimators_list` argument to `flaml.automl`, FLAML will start trying configurations for these models. If your input is Spark data, FLAML will also use estimators with the `_spark` postfix by default, even if you haven't specified them.\n",
"\n",
"Here is an example code snippet using SparkML models in AutoML:\n",
"\n",
"```python\n",
"import flaml\n",
"# prepare your data in pandas-on-spark format as we previously mentioned\n",
"\n",
"automl = flaml.AutoML()\n",
"settings = {\n",
" \"time_budget\": 30,\n",
" \"metric\": \"r2\",\n",
" \"estimator_list\": [\"lgbm_spark\"], # this setting is optional\n",
" \"task\": \"regression\",\n",
"}\n",
"\n",
"automl.fit(\n",
" dataframe=psdf,\n",
" label=label,\n",
" **settings,\n",
")\n",
"```\n",
"\n",
"\n",
"[Link to notebook](https://github.com/microsoft/FLAML/blob/main/notebook/automl_bankrupt_synapseml.ipynb) | [Open in colab](https://colab.research.google.com/github/microsoft/FLAML/blob/main/notebook/automl_bankrupt_synapseml.ipynb)\n",
"\n",
"## Parallel Spark Jobs\n",
"You can activate Spark as the parallel backend during parallel tuning in both [AutoML](/docs/Use-Cases/Task-Oriented-AutoML#parallel-tuning) and [Hyperparameter Tuning](/docs/Use-Cases/Tune-User-Defined-Function#parallel-tuning), by setting the `use_spark` to `true`. FLAML will dispatch your job to the distributed Spark backend using [`joblib-spark`](https://github.com/joblib/joblib-spark).\n",
"\n",
"Please note that you should not set `use_spark` to `true` when applying AutoML and Tuning for Spark Data. This is because only SparkML models will be used for Spark Data in AutoML and Tuning. As SparkML models run in parallel, there is no need to distribute them with `use_spark` again.\n",
"\n",
"All the Spark-related arguments are stated below. These arguments are available in both Hyperparameter Tuning and AutoML:\n",
"\n",
"\n",
"- `use_spark`: boolean, default=False | Whether to use spark to run the training in parallel spark jobs. This can be used to accelerate training on large models and large datasets, but will incur more overhead in time and thus slow down training in some cases. GPU training is not supported yet when use_spark is True. For Spark clusters, by default, we will launch one trial per executor. However, sometimes we want to launch more trials than the number of executors (e.g., local mode). In this case, we can set the environment variable `FLAML_MAX_CONCURRENT` to override the detected `num_executors`. The final number of concurrent trials will be the minimum of `n_concurrent_trials` and `num_executors`.\n",
"- `n_concurrent_trials`: int, default=1 | The number of concurrent trials. When n_concurrent_trials > 1, FLAML performes parallel tuning.\n",
"- `force_cancel`: boolean, default=False | Whether to forcely cancel Spark jobs if the search time exceeded the time budget. Spark jobs include parallel tuning jobs and Spark-based model training jobs.\n",
"\n",
"An example code snippet for using parallel Spark jobs:\n",
"```python\n",
"import flaml\n",
"automl_experiment = flaml.AutoML()\n",
"automl_settings = {\n",
" \"time_budget\": 30,\n",
" \"metric\": \"r2\",\n",
" \"task\": \"regression\",\n",
" \"n_concurrent_trials\": 2,\n",
" \"use_spark\": True,\n",
" \"force_cancel\": True, # Activating the force_cancel option can immediately halt Spark jobs once they exceed the allocated time_budget.\n",
"}\n",
"\n",
"automl.fit(\n",
" dataframe=dataframe,\n",
" label=label,\n",
" **automl_settings,\n",
")\n",
"```\n",
"\n",
"\n",
"[Link to notebook](https://github.com/microsoft/FLAML/blob/main/notebook/integrate_spark.ipynb) | [Open in colab](https://colab.research.google.com/github/microsoft/FLAML/blob/main/notebook/integrate_spark.ipynb)\n",
"Artifacts can differ among various machine learning libraries. More detailed information can be found [here](https://mlflow.org/docs/latest/tracking.html#automatic-logging).\n",
"\n",
"\n",
"\n",
"\n",
"## Plot Experiment Result\n",
"The `flaml.visualization` module provides utility functions for plotting the optimization process using [plotly](https://plotly.com/python/). Leveraging `plotly`, users can interactively explore experiment results. To use these plotting functions, simply provide your optimized `flaml.AutoML` or `flaml.tune.tune.ExperimentAnalysis` object as input. Optional parameters can be added using keyword arguments.\n",
"\n",
"Avaliable plotting functions:\n",
"- `plot_optimization_history`: Plot optimization history of all trials in the experiment.\n",
"- `plot_feature_importance`: Plot importance for each feature in the dataset.\n",
"- `plot_parallel_coordinate`: Plot the high-dimensional parameter relationships in the experiment.\n",
"- `plot_contour`: Plot the parameter relationship as contour plot in the experiment.\n",
"- `plot_edf`: Plot the objective value EDF (empirical distribution function) of the experiment.\n",
"- `plot_timeline`: Plot the timeline of the experiment.\n",
"- `plot_slice`: Plot the parameter relationship as slice plot in a study.\n",
"\n",
"### Figure Examples\n",
"\n",
"\n",
"Check out our example [notebook](../../notebook/trident/automl_plot.ipynb) for a preview of all interactive plots.\n",
"To use Spark for parallel training in FLAML, you can activate Spark as the parallel backend during parallel tuning in both AutoML and Hyperparameter Tuning, by setting the `use_spark` to `true`. FLAML will dispatch your job to the distributed Spark backend using `joblib-spark`. \n",
"\n",
"Here is an example code snippet for using parallel Spark jobs:\n",
"Note that you should not set `use_spark` to `true` when applying AutoML and Tuning for Spark Data. This is because only SparkML models will be used for Spark Data in AutoML and Tuning. As SparkML models run in parallel, there is no need to distribute them with `use_spark` again.\n",
"\n",
"You can also use Spark ML estimators for AutoML. FLAML integrates estimators based on Spark ML models. These models are trained in parallel using Spark, so we called them Spark estimators. To use these models, you first need to organize your data in the required format.\n",
"\n",
"Here is an example code snippet for Spark Data:\n",
"To use Spark ML models you need to format your data appropriately. Specifically, use `VectorAssembler` to merge all feature columns into a single vector column.\n",
"Later in conducting the experiment, use your pandas-on-spark data like non-spark data and pass them using `X_train, y_train` or `dataframe, label`.\n",
"\n",
"You can also plot the optimization process using `plotly` by providing your optimized `flaml.AutoML` or `flaml.tune.tune.ExperimentAnalysis` object as input. Optional parameters can be added using keyword arguments. Available plotting functions include `plot_optimization_history`, `plot_feature_importance`, `plot_parallel_coordinate`, `plot_contour`, `plot_edf`, `plot_timeline`, and `plot_slice`.\n",
"### Call RetrieveUserProxyAgent while init chat with another user proxy agent\n",
"Sometimes, there might be a need to use RetrieveUserProxyAgent in group chat without initializing the chat with it. In such scenarios, it becomes essential to create a function that wraps the RAG agents and allows them to be called from other agents."
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"\u001b[33mBoss\u001b[0m (to chat_manager):\n",
"\n",
"How to use spark for parallel training in FLAML? Give me sample code.\n",
"\u001b[32m***** Response from calling function \"retrieve_content\" *****\u001b[0m\n",
"You're a retrieve augmented coding assistant. You answer user's questions based on your own knowledge and the\n",
"context provided by the user.\n",
"If you can't answer the question with or without the current context, you should reply exactly `UPDATE CONTEXT`.\n",
"For code generation, you must obey the following rules:\n",
"Rule 1. You MUST NOT install any packages because all the packages needed are already installed.\n",
"Rule 2. You must follow the formats below to write your code:\n",
"```language\n",
"# your code\n",
"```\n",
"\n",
"User's question is: How to use spark for parallel training in FLAML?\n",
"\n",
"Context is: # Integrate - Spark\n",
"\n",
"FLAML has integrated Spark for distributed training. There are two main aspects of integration with Spark:\n",
"- Use Spark ML estimators for AutoML.\n",
"- Use Spark to run training in parallel spark jobs.\n",
"\n",
"## Spark ML Estimators\n",
"\n",
"FLAML integrates estimators based on Spark ML models. These models are trained in parallel using Spark, so we called them Spark estimators. To use these models, you first need to organize your data in the required format.\n",
"\n",
"### Data\n",
"\n",
"For Spark estimators, AutoML only consumes Spark data. FLAML provides a convenient function `to_pandas_on_spark` in the `flaml.automl.spark.utils` module to convert your data into a pandas-on-spark (`pyspark.pandas`) dataframe/series, which Spark estimators require.\n",
"\n",
"This utility function takes data in the form of a `pandas.Dataframe` or `pyspark.sql.Dataframe` and converts it into a pandas-on-spark dataframe. It also takes `pandas.Series` or `pyspark.sql.Dataframe` and converts it into a [pandas-on-spark](https://spark.apache.org/docs/latest/api/python/user_guide/pandas_on_spark/index.html) series. If you pass in a `pyspark.pandas.Dataframe`, it will not make any changes.\n",
"\n",
"This function also accepts optional arguments `index_col` and `default_index_type`.\n",
"- `index_col` is the column name to use as the index, default is None.\n",
"- `default_index_type` is the default index type, default is \"distributed-sequence\". More info about default index type could be found on Spark official [documentation](https://spark.apache.org/docs/latest/api/python/user_guide/pandas_on_spark/options.html#default-index-type)\n",
"\n",
"Here is an example code snippet for Spark Data:\n",
"To use Spark ML models you need to format your data appropriately. Specifically, use [`VectorAssembler`](https://spark.apache.org/docs/latest/api/python/reference/api/pyspark.ml.feature.VectorAssembler.html) to merge all feature columns into a single vector column.\n",
"Later in conducting the experiment, use your pandas-on-spark data like non-spark data and pass them using `X_train, y_train` or `dataframe, label`.\n",
"\n",
"### Estimators\n",
"#### Model List\n",
"- `lgbm_spark`: The class for fine-tuning Spark version LightGBM models, using [SynapseML](https://microsoft.github.io/SynapseML/docs/features/lightgbm/about/) API.\n",
"\n",
"#### Usage\n",
"First, prepare your data in the required format as described in the previous section.\n",
"\n",
"By including the models you intend to try in the `estimators_list` argument to `flaml.automl`, FLAML will start trying configurations for these models. If your input is Spark data, FLAML will also use estimators with the `_spark` postfix by default, even if you haven't specified them.\n",
"\n",
"Here is an example code snippet using SparkML models in AutoML:\n",
"\n",
"```python\n",
"import flaml\n",
"# prepare your data in pandas-on-spark format as we previously mentioned\n",
"\n",
"automl = flaml.AutoML()\n",
"settings = {\n",
" \"time_budget\": 30,\n",
" \"metric\": \"r2\",\n",
" \"estimator_list\": [\"lgbm_spark\"], # this setting is optional\n",
" \"task\": \"regression\",\n",
"}\n",
"\n",
"automl.fit(\n",
" dataframe=psdf,\n",
" label=label,\n",
" **settings,\n",
")\n",
"```\n",
"\n",
"\n",
"[Link to notebook](https://github.com/microsoft/FLAML/blob/main/notebook/automl_bankrupt_synapseml.ipynb) | [Open in colab](https://colab.research.google.com/github/microsoft/FLAML/blob/main/notebook/automl_bankrupt_synapseml.ipynb)\n",
"\n",
"## Parallel Spark Jobs\n",
"You can activate Spark as the parallel backend during parallel tuning in both [AutoML](/docs/Use-Cases/Task-Oriented-AutoML#parallel-tuning) and [Hyperparameter Tuning](/docs/Use-Cases/Tune-User-Defined-Function#parallel-tuning), by setting the `use_spark` to `true`. FLAML will dispatch your job to the distributed Spark backend using [`joblib-spark`](https://github.com/joblib/joblib-spark).\n",
"\n",
"Please note that you should not set `use_spark` to `true` when applying AutoML and Tuning for Spark Data. This is because only SparkML models will be used for Spark Data in AutoML and Tuning. As SparkML models run in parallel, there is no need to distribute them with `use_spark` again.\n",
"\n",
"All the Spark-related arguments are stated below. These arguments are available in both Hyperparameter Tuning and AutoML:\n",
"\n",
"\n",
"- `use_spark`: boolean, default=False | Whether to use spark to run the training in parallel spark jobs. This can be used to accelerate training on large models and large datasets, but will incur more overhead in time and thus slow down training in some cases. GPU training is not supported yet when use_spark is True. For Spark clusters, by default, we will launch one trial per executor. However, sometimes we want to launch more trials than the number of executors (e.g., local mode). In this case, we can set the environment variable `FLAML_MAX_CONCURRENT` to override the detected `num_executors`. The final number of concurrent trials will be the minimum of `n_concurrent_trials` and `num_executors`.\n",
"- `n_concurrent_trials`: int, default=1 | The number of concurrent trials. When n_concurrent_trials > 1, FLAML performes parallel tuning.\n",
"- `force_cancel`: boolean, default=False | Whether to forcely cancel Spark jobs if the search time exceeded the time budget. Spark jobs include parallel tuning jobs and Spark-based model training jobs.\n",
"\n",
"An example code snippet for using parallel Spark jobs:\n",
"```python\n",
"import flaml\n",
"automl_experiment = flaml.AutoML()\n",
"automl_settings = {\n",
" \"time_budget\": 30,\n",
" \"metric\": \"r2\",\n",
" \"task\": \"regression\",\n",
" \"n_concurrent_trials\": 2,\n",
" \"use_spark\": True,\n",
" \"force_cancel\": True, # Activating the force_cancel option can immediately halt Spark jobs once they exceed the allocated time_budget.\n",
"}\n",
"\n",
"automl.fit(\n",
" dataframe=dataframe,\n",
" label=label,\n",
" **automl_settings,\n",
")\n",
"```\n",
"\n",
"\n",
"[Link to notebook](https://github.com/microsoft/FLAML/blob/main/notebook/integrate_spark.ipynb) | [Open in colab](https://colab.research.google.com/github/microsoft/FLAML/blob/main/notebook/integrate_spark.ipynb)\n",
"Artifacts can differ among various machine learning libraries. More detailed information can be found [here](https://mlflow.org/docs/latest/tracking.html#automatic-logging).\n",
"\n",
"\n",
"\n",
"\n",
"## Plot Experiment Result\n",
"The `flaml.visualization` module provides utility functions for plotting the optimization process using [plotly](https://plotly.com/python/). Leveraging `plotly`, users can interactively explore experiment results. To use these plotting functions, simply provide your optimized `flaml.AutoML` or `flaml.tune.tune.ExperimentAnalysis` object as input. Optional parameters can be added using keyword arguments.\n",
"\n",
"Avaliable plotting functions:\n",
"- `plot_optimization_history`: Plot optimization history of all trials in the experiment.\n",
"- `plot_feature_importance`: Plot importance for each feature in the dataset.\n",
"- `plot_parallel_coordinate`: Plot the high-dimensional parameter relationships in the experiment.\n",
"- `plot_contour`: Plot the parameter relationship as contour plot in the experiment.\n",
"- `plot_edf`: Plot the objective value EDF (empirical distribution function) of the experiment.\n",
"- `plot_timeline`: Plot the timeline of the experiment.\n",
"- `plot_slice`: Plot the parameter relationship as slice plot in a study.\n",
"\n",
"### Figure Examples\n",
"\n",
"\n",
"Check out our example [notebook](../../notebook/trident/automl_plot.ipynb) for a preview of all interactive plots.\n",
"To use Spark for parallel training in FLAML, you can follow these steps:\n",
"\n",
"1. Prepare your data in the required format. FLAML only consumes Spark data for Spark estimators. You can use the `to_pandas_on_spark` function from the `flaml.automl.spark.utils` module to convert your data into a pandas-on-spark dataframe. Here's an example:\n",
"2. Use the Spark ML estimators in FLAML. FLAML integrates estimators based on Spark ML models. You can include the models you want to try in the `estimator_list` argument when creating an instance of `flaml.AutoML`. Here's an example:\n",
"\n",
"```python\n",
"import flaml\n",
"\n",
"automl = flaml.AutoML()\n",
"settings = {\n",
" \"time_budget\": 30,\n",
" \"metric\": \"r2\",\n",
" \"estimator_list\": [\"lgbm_spark\"], # Optional: specify the Spark ML estimator\n",
" \"task\": \"regression\"\n",
"}\n",
"\n",
"automl.fit(\n",
" dataframe=psdf,\n",
" label=label,\n",
" **settings\n",
")\n",
"```\n",
"\n",
"3. Activate Spark as the parallel backend. You can set the `use_spark` parameter to `True` to activate Spark as the parallel backend during parallel tuning. FLAML will dispatch your job to the distributed Spark backend using `joblib-spark`. Here's an example:\n",
"\n",
"```python\n",
"import flaml\n",
"\n",
"automl_experiment = flaml.AutoML()\n",
"automl_settings = {\n",
" \"time_budget\": 30,\n",
" \"metric\": \"r2\",\n",
" \"task\": \"regression\",\n",
" \"n_concurrent_trials\": 2,\n",
" \"use_spark\": True,\n",
" \"force_cancel\": True # Optional: force cancel Spark jobs if time budget is exceeded\n",
"}\n",
"\n",
"automl_experiment.fit(\n",
" dataframe=dataframe,\n",
" label=label,\n",
" **automl_settings\n",
")\n",
"```\n",
"\n",
"These are the steps to use Spark for parallel training in FLAML. Let me know if you need any further assistance!\n",
