validmind 2.7.12__py3-none-any.whl → 2.8.12__py3-none-any.whl

validmind/ai/test_result_description/image_processing.py

@@ -1,124 +0,0 @@
-# Copyright © 2023-2024 ValidMind Inc. All rights reserved.
-# See the LICENSE file in the root of this repository for details.
-# SPDX-License-Identifier: AGPL-3.0 AND ValidMind Commercial
-
-import base64
-import io
-
-import numpy as np
-from PIL import Image, ImageEnhance, ImageFilter
-
-DOWNSAMPLE_PERCENTAGE = 50
-
-
-def open_base64_image(base64_string):
-    if base64_string.startswith("data:image/png;base64,"):
-        base64_string = base64_string.split(",")[1]
-
-    image_data = base64.b64decode(base64_string)
-    image_buffer = io.BytesIO(image_data)
-    image = Image.open(image_buffer)
-
-    return image
-
-
-def downsample_image(base64_string):
-    image = open_base64_image(base64_string)
-
-    # Calculate the target dimensions based on the reduction percentage
-    target_width = int(image.width * (1 - DOWNSAMPLE_PERCENTAGE / 100))
-    target_height = int(image.height * (1 - DOWNSAMPLE_PERCENTAGE / 100))
-
-    # If the image is already smaller than the target size, return the original
-    if image.width <= target_width and image.height <= target_height:
-        return base64_string
-
-    # remove any margins from the image
-    # Find the bounding box of non-uniform pixels (margin detection)
-    width, height = image.size
-    background = image.getpixel((0, 0))  # Assume top-left pixel is background color
-
-    def is_different(pixel):
-        return pixel != background
-
-    left = next(
-        x
-        for x in range(width)
-        if any(is_different(image.getpixel((x, y))) for y in range(height))
-    )
-    right = next(
-        x
-        for x in range(width - 1, -1, -1)
-        if any(is_different(image.getpixel((x, y))) for y in range(height))
-    )
-    top = next(
-        y
-        for y in range(height)
-        if any(is_different(image.getpixel((x, y))) for x in range(width))
-    )
-    bottom = next(
-        y
-        for y in range(height - 1, -1, -1)
-        if any(is_different(image.getpixel((x, y))) for x in range(width))
-    )
-
-    # Crop the image to remove the uniform margin (with some padding)
-    bbox = (left - 5, top - 5, right + 6, bottom + 6)
-    image = image.crop(bbox)
-
-    # If the image has an alpha channel, remove any transparent margins
-    if image.mode in ("RGBA", "LA"):
-        alpha = image.getchannel("A")
-        bbox = alpha.getbbox()
-        if bbox:
-            image = image.crop(bbox)
-
-    # Apply unsharp mask to enhance edges
-    image = image.filter(ImageFilter.UnsharpMask(radius=2, percent=150, threshold=3))
-
-    # Calculate new dimensions
-    aspect_ratio = image.width / image.height
-    new_height = target_height
-    new_width = int(new_height * aspect_ratio)
-
-    # print(f"downsampling from {width}x{height} to {new_width}x{new_height}")
-
-    # Ensure we don't exceed the target width
-    if new_width > target_width:
-        new_width = target_width
-        new_height = int(new_width / aspect_ratio)
-
-    # print(f"downsampling from {image.width}x{image.height} to {new_width}x{new_height}")
-
-    # Convert to numpy array for custom downsampling
-    img_array = np.array(image)
-
-    # Optimized area interpolation
-    h_factor = img_array.shape[0] / new_height
-    w_factor = img_array.shape[1] / new_width
-
-    h_indices = (np.arange(new_height).reshape(-1, 1) * h_factor).astype(int)
-    w_indices = (np.arange(new_width).reshape(1, -1) * w_factor).astype(int)
-
-    h_indices = np.minimum(h_indices, img_array.shape[0] - 1)
-    w_indices = np.minimum(w_indices, img_array.shape[1] - 1)
-
-    # Convert back to PIL Image
-    image = Image.fromarray(img_array[h_indices, w_indices].astype(np.uint8))
-
-    # Enhance contrast slightly
-    enhancer = ImageEnhance.Contrast(image)
-    image = enhancer.enhance(1.2)
-
-    # Sharpen the image
-    image = image.filter(ImageFilter.SHARPEN)
-
-    # Convert the image to bytes in PNG format
-    buffered = io.BytesIO()
-    image.save(buffered, format="PNG")
-    img_bytes = buffered.getvalue()
-
-    # Encode the bytes to base64
-    b64_encoded = base64.b64encode(img_bytes).decode("utf-8")
-
-    return f"data:image/png;base64,{b64_encoded}"

validmind/ai/test_result_description/system.jinja

@@ -1,39 +0,0 @@
-You are an expert data scientist and MRM specialist.
-You are tasked with analyzing the results of a quantitative test run on some model or dataset.
-Your goal is to create a test description that will act as part of the model documentation.
-You will provide both the developer and other consumers of the documentation with a clear and concise "interpretation" of the results they will see.
-The overarching theme to maintain is MRM documentation.
-
-Examine the provided statistical test results and compose a description of the results.
-The results are either in the form of serialized tables or images of plots.
-Compose a description and interpretation of the result to accompany it in MRM documentation.
-It will be read by other data scientists and developers and by validators and stakeholders.
-
-Use valid Markdown syntax to format the response.
-Avoid long sentences and complex vocabulary.
-Avoid overly verbose explanations - the goal is to explain to a user what they are seeing in the results.
-Structure the response clearly and logically.
-Respond only with your analysis and insights, not the verbatim test results.
-Respond only with the markdown content, no explanation or context for your response is necessary.
-Use the Test ID that is provided to form the Test Name e.g. "ClassImbalance" -> "Class Imbalance".
-
-Explain the test, its purpose, its mechanism/formula etc and why it is useful.
-If relevant, provide a very brief description of the way this test is used in model/dataset evaluation and how it is interpreted.
-Highlight the key insights from the test results. The key insights should be concise and easily understood.
-An insight should only be included if it is something not entirely obvious from the test results.
-End the response with any closing remarks, summary or additional useful information.
-
-Use the following format for the response (feel free to stray from it if necessary - this is a suggested starting point):
-
-<ResponseFormat>
-**<Test Name>** calculates the xyz <continue to explain what it does in detail>...
-
-This test is useful for <explain why and for what this test is useful>...
-
-**Key Insights:**
-
-The following key insights can be identified in the test results:
-
-- **<key insight 1 - title>**: <concise explanation of key insight 1>
-- ...<continue with any other key insights using the same format>
-</ResponseFormat>

validmind/ai/test_result_description/user.jinja

@@ -1,30 +0,0 @@
-**Test ID**: `{{ test_name }}`
-
-**Test Description**:
-
-{{ test_description }}
-
----
-
-Generate a description of the following result of the test using the instructions given in your system prompt.
-
-{%- if context %}
-**Context**:
-{{ context }}
-{%- endif %}
-
-{%- if summary %}
-**Test Result Tables** *(Raw Data)*:
-{{ summary }}
-{%- endif %}
-
-{%- if figures %}
-The following images make up the results of the test.
-{%- for b64_image_url in figures %}
-[[IMAGE:{{ b64_image_url }}]]
-{%- endfor %}
-{%- endif %}
-
-Keep your response concise and to the point!
-Only include content in your response if its something truly insightful or interesting!
-DO NOT VERBOSELY EXPLAIN THE TEST OR THE RESULTS!!!