Version 1.2 #534
Version 1.2 #534
Conversation
* fix: directory for saving charts issue (#513) * (fix): update template (remove hardcode of directory) in `GeneratePythonCodePrompt` class * (fix): add passing "save_charts_path" from the config object to `get_prompt()` * tests: custom path for charts (#513) * (tests): add test method for a case passing custom path with `save_charts_path` parameter * test: add test for the custom chart prompt --------- Co-authored-by: Gabriele Venturi <[email protected]>
* feat[DFValidator]: Add return types * fix pre-commit errors * add return value * refactor: minor changes in validation --------- Co-authored-by: arslan-agory <[email protected]> Co-authored-by: Gabriele Venturi <[email protected]>
* fix: update type hint for config parameter (#526) * (fix): add `dict` as a possible type of object to pass in methods and functions where `config` parameter occurs * (feat): add logging of an exception to `load_config()` instead of silently suppressing an error * fix: use Union for config hints (#526) * (fix): replace "bitwise or" (`|`) usage with `Union` in type hints for `config` object
* feat(llm): deprecate falcon and starcoder * docs: mention Starcoder and Falcon being deprecated --------- Co-authored-by: Gabriele Venturi <[email protected]>
WalkthroughThis update introduces flexibility in configuration handling, deprecates certain classes, enhances dataframe validation with Pydantic schemas, and improves chart saving functionality. It also includes necessary test updates to validate these changes. Changes
TipsChat with CodeRabbit Bot (
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Review Status
Actionable comments generated: 3
Files ignored due to filter (2)
- mkdocs.yml
- pyproject.toml
Files selected for processing (13)
- docs/API/llms.md (1 hunks)
- docs/LLMs/llms.md (2 hunks)
- pandasai/init.py (4 hunks)
- pandasai/config.py (2 hunks)
- pandasai/helpers/code_manager.py (2 hunks)
- pandasai/helpers/df_validator.py (1 hunks)
- pandasai/llm/falcon.py (2 hunks)
- pandasai/llm/starcoder.py (2 hunks)
- pandasai/prompts/generate_python_code.py (1 hunks)
- pandasai/smart_dataframe/init.py (5 hunks)
- pandasai/smart_datalake/init.py (5 hunks)
- tests/prompts/test_generate_python_code_prompt.py (2 hunks)
- tests/test_smartdataframe.py (6 hunks)
Files skipped from review due to trivial changes (5)
- docs/API/llms.md
- docs/LLMs/llms.md
- pandasai/llm/falcon.py
- pandasai/llm/starcoder.py
- pandasai/prompts/generate_python_code.py
Additional comments (Suppressed): 23
pandasai/config.py (1)
- 12-15: The type hint for
override_confighas been updated to accept either aConfigobject or a dictionary. This is a good change as it provides more flexibility in how the configuration can be provided.pandasai/smart_dataframe/__init__.py (3)
56-62: The function signature has been updated to accept either a
Configobject or a dictionary for configuration. Ensure that all calls to this function throughout the codebase have been updated to match the new signature.65-71: The docstring has been updated to reflect the changes in the function signature. The description of the
configparameter now correctly indicates that it can be either aConfigobject or a dictionary.238-253: A new method
validatehas been added which uses theDfValidatorclass to validate the dataframe rows based on a Pydantic schema input. This is a good addition as it allows for robust data validation using Pydantic's powerful and flexible validation capabilities.pandasai/__init__.py (5)
103-117: The
__init__method of thePandasAIclass has been updated to accept either aConfigobject or a dictionary for configuration. This change enhances flexibility and allows users to pass in configurations in a more convenient way. However, ensure that all calls to this function throughout the codebase have been updated to match the new signature.146-148: Deprecation warning added for the
PandasAIclass, suggesting users to useSmartDataframeinstead. This is a good practice as it informs users about the upcoming changes and gives them time to adapt their code accordingly.150-150: The
_configattribute of thePandasAIclass is now initialized with aConfigobject. This is consistent with the change made in the__init__method where it can now accept either aConfigobject or a dictionary for configuration.163-172: No significant changes detected in the
runmethod of thePandasAIclass. The method signature and functionality remain the same.200-209: No significant changes detected in the
__call__method of thePandasAIclass. The method signature and functionality remain the same.pandasai/helpers/df_validator.py (2)
1-52: The
DfValidationResultclass is well implemented. It encapsulates the result of a validation operation, including whether it passed and any errors that occurred. The use of properties forpassedanderrorsensures that these attributes are read-only, which is good for immutability. Theadd_errormethod allows adding an error message to the results and automatically setspassedtoFalse, which is a logical behavior. The__bool__method is also correctly implemented to return the value ofpassed, allowing aDfValidationResultobject to be used in a boolean context.54-126: The
DfValidatorclass is responsible for validating a dataframe using a Pydantic schema. The_validate_batchmethod validates a batch of rows from the dataframe against the provided schema and returns any errors. The_df_to_list_of_dictmethod converts the dataframe to a list of dictionaries, which is the format required for validation. Thevalidatemethod orchestrates the validation process by determining the type of the dataframe, converting it to a list of dictionaries, validating it, and returning the results. This class is well designed and implemented.However, there's a potential issue with the
_df_to_list_of_dictmethod. If thedataframe_typeis neither "pandas" nor "polars", it returns an empty list. This could lead to silent failures if an unsupported dataframe type is used. It would be better to raise an exception in this case.tests/prompts/test_generate_python_code_prompt.py (1)
- 26-27: The new variable
save_charts_pathis being set but not used in the assertion. Please ensure that the test case checks for the correct usage of this variable.pandasai/helpers/code_manager.py (2)
21-27: The type hint for
_confighas been updated to accept either aConfigobject or a dictionary. This change enhances the flexibility of configuration, but it also introduces potential issues. If a dictionary is passed instead of aConfigobject, there might be missing keys that theConfigobject would have guaranteed. Ensure that all usages of_configin this class handle bothConfigobjects and dictionaries correctly.30-35: The
__init__method now accepts either aConfigobject or a dictionary for theconfigparameter. As mentioned above, ensure that all usages ofconfigin this class handle bothConfigobjects and dictionaries correctly.pandasai/smart_datalake/__init__.py (5)
44-50: The type of
_confighas been changed fromConfigtoUnion[Config, dict]. This change allows for more flexibility in the configuration of theSmartDatalakeclass. However, it's important to ensure that all parts of the code that interact with_configcan handle bothConfigobjects and dictionaries.60-72: The
__init__method now accepts either aConfigobject or a dictionary for theconfigparameter. This is a good change as it enhances the flexibility of theSmartDatalakeclass. However, make sure that the rest of the codebase is updated to reflect this change where necessary.138-154: The
_load_configmethod has been updated to accept either aConfigobject or a dictionary. Theload_configfunction is then used to load the configuration. This is a good change as it enhances the flexibility of theSmartDatalakeclass. However, make sure that theload_configfunction can handle bothConfigobjects and dictionaries.164-170: The check for whether an LLM is a PandasAI LLM has been simplified by checking if the LLM has the
_llm_typeattribute. This is a cleaner way of performing the check and should improve readability.262-268: The
default_valuesdictionary now includes thesave_charts_pathkey. This is a good addition as it allows the path for saving charts to be customized. However, make sure that the rest of the codebase is updated to reflect this change where necessary.tests/test_smartdataframe.py (4)
14-14: The import of
BaseModelandFieldfrompydanticis new. Ensure that thepydanticpackage is included in your project dependencies.125-127: The test case
test_init_without_llmhas been updated to pass a configuration dictionary withllmset toNone. This change aligns with the updates made to theSmartDataframeclass to accept a configuration dictionary. The test case should fail as expected if no LLM is provided.244-309: This is a new test case for saving charts in a non-default directory. It checks whether the chart is saved in the correct directory and whether the path to the saved chart is correctly returned. The test case seems well-written and covers all necessary assertions.
728-824: These are new test cases for validating dataframes using Pydantic schemas. They cover various scenarios, including validation passing, validation failing due to incorrect types, validation with Polars dataframes, validation failing for one record, and validation with complex Pydantic schemas. These tests are important for ensuring the correct functioning of the new dataframe validation feature.
| if config.get("callback") and not override_config.get("callback"): | ||
| config["callback"] = getattr(callbacks, config["callback"])() | ||
| except Exception: | ||
| pass | ||
| logger.error("Could not load configuration", exc_info=True) | ||
|
|
||
| if override_config: | ||
| config.update(override_config) | ||
|
|
||
| config = Config(**config) | ||
|
|
||
| return config |
There was a problem hiding this comment.
The error handling has been improved by logging the exception details instead of silently passing it. However, the code no longer converts the final config dictionary into a Config object before returning it. This could potentially lead to issues if the rest of the codebase expects a Config object rather than a dictionary. Please verify that this change is compatible with the rest of the codebase.
- return config
+ config = Config(**config)
+ return config| """ | ||
| Create list of dict of dataframe rows on basis of dataframe type | ||
| Supports only polars and pandas dataframe | ||
|
|
||
| Args: | ||
| df: dataframe to be converted | ||
| dataframe_type: type of dataframe | ||
|
|
||
| Returns: | ||
| list of dict of dataframe rows | ||
| """ | ||
| if dataframe_type == "pandas": | ||
| return df.to_dict(orient="records") | ||
| elif dataframe_type == "polars": | ||
| return df.to_dicts() | ||
| else: | ||
| return [] |
There was a problem hiding this comment.
- else:
- return []
+ else:
+ raise ValueError(f"Unsupported DataFrame type: {dataframe_type}")| def test_str_with_custom_save_charts_path(self): | ||
| """Test that the __str__ method is implemented""" | ||
|
|
||
| llm = FakeLLM("plt.show()") | ||
| dfs = [ | ||
| SmartDataframe( | ||
| pd.DataFrame({"a": [1], "b": [4]}), | ||
| config={"llm": llm}, | ||
| ) | ||
| ] | ||
|
|
||
| prompt = GeneratePythonCodePrompt() | ||
| prompt.set_var("dfs", dfs) | ||
| prompt.set_var("conversation", "Question") | ||
| prompt.set_var("save_charts_path", "custom_path") | ||
|
|
||
| assert ( | ||
| prompt.to_string() | ||
| == """ | ||
| You are provided with the following pandas DataFrames with the following metadata: | ||
|
|
||
| Dataframe dfs[0], with 1 rows and 2 columns. | ||
| This is the metadata of the dataframe dfs[0]: | ||
| a,b | ||
| 1,4 | ||
|
|
||
|
|
||
| This is the initial python code to be updated: | ||
| ```python | ||
| # TODO import all the dependencies required | ||
| import pandas as pd | ||
|
|
||
| # Analyze the data | ||
| # 1. Prepare: Preprocessing and cleaning data if necessary | ||
| # 2. Process: Manipulating data for analysis (grouping, filtering, aggregating, etc.) | ||
| # 3. Analyze: Conducting the actual analysis (if the user asks to create a chart save it to an image in custom_path/temp_chart.png and do not show the chart.) | ||
| # 4. Output: return a dictionary of: | ||
| # - type (possible values "text", "number", "dataframe", "plot") | ||
| # - value (can be a string, a dataframe or the path of the plot, NOT a dictionary) | ||
| # Example output: { "type": "text", "value": "The average loan amount is $15,000." } | ||
| def analyze_data(dfs: list[pd.DataFrame]) -> dict: | ||
| # Code goes here (do not add comments) | ||
|
|
||
|
|
||
| # Declare a result variable | ||
| result = analyze_data(dfs) | ||
| ``` |
There was a problem hiding this comment.
A new test case test_str_with_custom_save_charts_path has been added to verify the functionality of setting a custom path for saving charts. The test case seems well-structured and covers the necessary assertions. However, it would be beneficial to add an assertion to check if the chart is actually saved at the specified location.
+ assert os.path.exists("custom_path/temp_chart.png")
|
@gventuri Are we planning to also discontinue the |
|
@mspronesti I suppose we should basically change the implementation so that we go for the "official" way (not limited to 10 tokens with the "hacky" thing of running multiple calls to get the whole code generated |
Summary by CodeRabbit
Release Notes
DfValidatorfor validating dataframes using Pydantic schemas.Configobject or a dictionary in various classes and methods.sample_headparameter to theSmartDataframeclass.StarcoderandFalconclasses as deprecated with warnings suggesting alternatives.