start working on a quick encoder test, but it's far from finished

aimodel/src/encoderonly_test_rainfall.py

@@ -0,0 +1,59 @@
+from loguru import logger
+
+import os
+
+import tensorflow as tf
+
+from lib.ai.components.convnext import make_convnext
+
+
+# ███████ ███    ██ ██    ██
+# ██      ████   ██ ██    ██
+# █████   ██ ██  ██ ██    ██
+# ██      ██  ██ ██  ██  ██
+# ███████ ██   ████   ████
+
+# TODO: env vars & settings here
+
+
+
+
+# ██████   █████  ████████  █████  ███████ ███████ ████████
+# ██   ██ ██   ██    ██    ██   ██ ██      ██         ██
+# ██   ██ ███████    ██    ███████ ███████ █████      ██
+# ██   ██ ██   ██    ██    ██   ██      ██ ██         ██
+# ██████  ██   ██    ██    ██   ██ ███████ ███████    ██
+
+
+
+
+# ███    ███  ██████  ██████  ███████ ██
+# ████  ████ ██    ██ ██   ██ ██      ██
+# ██ ████ ██ ██    ██ ██   ██ █████   ██
+# ██  ██  ██ ██    ██ ██   ██ ██      ██
+# ██      ██  ██████  ██████  ███████ ███████
+
+
+# TODO: Have an 
+
+
+# ████████ ██████   █████  ██ ███    ██ ██ ███    ██  ██████  
+#    ██    ██   ██ ██   ██ ██ ████   ██ ██ ████   ██ ██       
+#    ██    ██████  ███████ ██ ██ ██  ██ ██ ██ ██  ██ ██   ███ 
+#    ██    ██   ██ ██   ██ ██ ██  ██ ██ ██ ██  ██ ██ ██    ██ 
+#    ██    ██   ██ ██   ██ ██ ██   ████ ██ ██   ████  ██████  
+
+
+
+# ██████  ██████  ███████ ██████  ██  ██████ ████████ ██  ██████  ███    ██
+# ██   ██ ██   ██ ██      ██   ██ ██ ██         ██    ██ ██    ██ ████   ██
+# ██████  ██████  █████   ██   ██ ██ ██         ██    ██ ██    ██ ██ ██  ██
+# ██      ██   ██ ██      ██   ██ ██ ██         ██    ██ ██    ██ ██  ██ ██
+# ██      ██   ██ ███████ ██████  ██  ██████    ██    ██  ██████  ██   ████
+
+
+
+
+
+
+# █████ █████ █████ █████ █████ █████ █████ █████ █████

aimodel/src/lib/dataset/dataset_encoderonly.py

@@ -0,0 +1,196 @@
+import os
+import math
+import json
+
+from loguru import logger
+
+import tensorflow as tf
+
+from lib.dataset.read_metadata import read_metadata
+
+from ..io.readfile import readfile
+from .shuffle import shuffle
+from .parse_heightmap import parse_heightmap
+
+
+
+# TO PARSE:
+def parse_item(metadata, output_size=100, input_size="same", water_threshold=0.1, water_bins=2, heightmap=None, rainfall_scale_up=1):
+	if input_size == "same":
+		input_size = output_size # This is almost always the case with e.g. the DeepLabV3+ model
+	
+	water_height_source, water_width_source = metadata["waterdepth"]
+	water_offset_x = math.ceil((water_width_source - output_size) / 2)
+	water_offset_y = math.ceil((water_height_source - output_size) / 2)
+	
+	rainfall_channels, rainfall_height_source, rainfall_width_source = metadata["rainfallradar"]
+	rainfall_height_source *= rainfall_scale_up
+	rainfall_width_source *= rainfall_scale_up
+	rainfall_offset_x = math.ceil((rainfall_width_source - input_size) / 2)
+	rainfall_offset_y = math.ceil((rainfall_height_source - input_size) / 2)
+	
+	print("DEBUG DATASET:rainfall shape", metadata["rainfallradar"], "/", f"w {rainfall_width_source} h {rainfall_height_source}")
+	print("DEBUG DATASET:water shape", metadata["waterdepth"])
+	print("DEBUG DATASET:water_threshold", water_threshold)
+	print("DEBUG DATASET:water_bins", water_bins)
+	print("DEBUG DATASET:output_size", output_size)
+	print("DEBUG DATASET:input_size", input_size)
+	print("DEBUG DATASET:water_offset x", water_offset_x, "y", water_offset_y)
+	print("DEBUG DATASET:rainfall_offset x", rainfall_offset_x, "y", rainfall_offset_y)
+	
+	if heightmap is not None:
+		heightmap = tf.expand_dims(heightmap, axis=-1)
+		# NORMALLY, this wouldn't work 'cause you'd need to know the max of ALL frames, but here we only have a single frame.
+		heightmap_max = tf.math.reduce_max(heightmap)
+		heightmap_min = tf.math.reduce_min(tf.where(tf.math.less(heightmap, -500), heightmap, tf.fill(heightmap.shape, 0.0)))
+		heightmap = (heightmap - heightmap_min) / heightmap_max
+		heightmap = tf.transpose(heightmap, [1, 0, 2]) # [width, height] → [height, width]
+		rainfall_channels += 1
+	
+	def parse_item_inner(item):
+		parsed = tf.io.parse_single_example(item, features={
+			"rainfallradar": tf.io.FixedLenFeature([], tf.string),
+			"waterdepth": tf.io.FixedLenFeature([], tf.string)
+		})
+		rainfall = tf.io.parse_tensor(parsed["rainfallradar"], out_type=tf.float32)
+		water = tf.io.parse_tensor(parsed["waterdepth"], out_type=tf.float32)
+		
+		
+		rainfall = tf.reshape(rainfall, tf.constant(metadata["rainfallradar"], dtype=tf.int32))
+		water = tf.reshape(water, tf.constant(metadata["waterdepth"], dtype=tf.int32))
+		
+		# Apparently the water depth data is also in HW instead of WH.... sighs
+		# * YES IT IS, BUT TENSORFLOW *wants* NHWC NOT NWHC....!
+		# water = tf.transpose(water, [1, 0])
+		
+		# [channels, height, weight] → [height, width, channels] - ref ConvNeXt does not support data_format=channels_first
+		# BUG: For some reasons we have data that's not transposed correctly still!! O.o
+		# I can't believe in this entire project I have yet to get the rotation of the rainfall radar data correct....!
+		# %TRANSPOSE%
+		rainfall = tf.transpose(rainfall, [1, 2, 0])
+		if heightmap is not None:
+			rainfall = tf.concat([rainfall, heightmap], axis=-1)
+		if rainfall_scale_up > 1:
+			rainfall = tf.repeat(tf.repeat(rainfall, rainfall_scale_up, axis=0), rainfall_scale_up, axis=1)
+		if input_size is not None:
+			rainfall = tf.image.crop_to_bounding_box(rainfall,
+				offset_width=rainfall_offset_x,
+				offset_height=rainfall_offset_y,
+				target_width=input_size,
+				target_height=input_size,
+			)
+		
+		# rainfall = tf.image.resize(rainfall, tf.cast(tf.constant(metadata["rainfallradar"]) / 2, dtype=tf.int32))
+		water = tf.expand_dims(water, axis=-1) # [height, width] → [height, width, channels=1]
+		water = tf.image.crop_to_bounding_box(water,
+			offset_width=water_offset_x,
+			offset_height=water_offset_y,
+			target_width=output_size,
+			target_height=output_size
+		)
+		
+		print("DEBUG:dataset BEFORE_SQUEEZE water", water.shape)
+		water = tf.squeeze(water)
+		print("DEBUG:dataset AFTER_SQUEEZE water", water.shape)
+		# LOSS cross entropy
+		# water = tf.cast(tf.math.greater_equal(water, water_threshold), dtype=tf.int32)
+		# water = tf.one_hot(water, water_bins, axis=-1, dtype=tf.int32)
+		# LOSS dice
+		water = tf.cast(tf.math.greater_equal(water, water_threshold), dtype=tf.float32)
+		
+		rainfall = tf.image.extract_patches(tf.expand_dims(rainfall, axis=0),
+			sizes=[1,windowsize,windowsize],
+			strides=[1,1,1,1],
+			rates=[1,1,1,1],
+			padding="VALID"
+		)
+		rainfall = tf.reshape(rainfall, [-1, windowsize, windowsize, rainfall_channels])
+		
+		# TODO: extract single water values here to match the above rainfall patches
+		
+		print("DEBUG DATASET_OUT:rainfall shape", rainfall.shape)
+		print("DEBUG DATASET_OUT:water shape", water.shape)
+		return rainfall, water
+	
+	return tf.function(parse_item_inner)
+
+def make_dataset(filepaths, compression_type="GZIP", parallel_reads_multiplier=1.5, shuffle_buffer_size=128, batch_size=64, prefetch=True, shuffle=True, filepath_heightmap=None, **kwargs):
+	if "NO_PREFETCH" in os.environ:
+		logger.info("disabling data prefetching.")
+	
+	heightmap = None
+	if filepath_heightmap is not None:
+		logger.info(f"Using heightmap from '{filepath_heightmap}'.")
+		heightmap = parse_heightmap(filepath_heightmap)
+	
+	dataset = tf.data.TFRecordDataset(filepaths,
+		compression_type=compression_type,
+		num_parallel_reads=math.ceil(os.cpu_count() * parallel_reads_multiplier) if parallel_reads_multiplier > 0 else None
+	)
+	if shuffle:
+		dataset = dataset.shuffle(shuffle_buffer_size)
+	dataset = dataset.map(parse_item(heightmap=heightmap, **kwargs), num_parallel_calls=tf.data.AUTOTUNE) \
+		.unbatch()
+	
+	if batch_size != None:
+		dataset = dataset.batch(batch_size, drop_remainder=True)
+	if prefetch:
+		dataset = dataset.prefetch(0 if "NO_PREFETCH" in os.environ else tf.data.AUTOTUNE)
+	
+	return dataset
+
+
+def get_filepaths(dirpath_input, do_shuffle=True):
+	result = list(filter(
+		lambda filepath: str(filepath).endswith(".tfrecord.gz"),
+		[ file.path for file in os.scandir(dirpath_input) ] # .path on a DirEntry object yields the absolute filepath
+	))
+	if do_shuffle:
+		result = shuffle(result)
+	else:
+		result = sorted(result, key=lambda filepath: int(os.path.basename(filepath).split(".", 1)[0]))
+	
+	return result
+
+def dataset_mono(dirpath_input, train_percentage=0.8, **kwargs):
+	filepaths = get_filepaths(dirpath_input)
+	filepaths_count = len(filepaths)
+	dataset_splitpoint = math.floor(filepaths_count * train_percentage)
+	
+	filepaths_train = filepaths[:dataset_splitpoint]
+	filepaths_validate = filepaths[dataset_splitpoint:]
+	
+	metadata = read_metadata(dirpath_input)
+	
+	dataset_train = make_dataset(filepaths_train, metadata=metadata, **kwargs)
+	dataset_validate = make_dataset(filepaths_validate, metadata=metadata, **kwargs)
+	
+	return dataset_train, dataset_validate #, filepaths
+
+def dataset_mono_predict(dirpath_input, **kwargs):
+	"""Creates a tf.data.Dataset() for prediction using the contrastive learning model.
+	Note that this WILL MANGLE THE ORDERING if you set parallel_reads_multiplier to anything other than 0!!
+	
+	Args:
+		dirpath_input (string): The path to the directory containing the input (.tfrecord.gz) files
+		parallel_reads_multiplier (float, optional): The number of files to read in parallel. Defaults to 1.5.
+		prefetch (bool, optional): Whether to prefetch data into memory or not. Defaults to True.
+
+	Returns:
+		tf.data.Dataset: A tensorflow Dataset for the given input files.
+	"""
+	filepaths = get_filepaths(dirpath_input, do_shuffle=False) if os.path.isdir(dirpath_input) else [ dirpath_input ]
+	
+	return make_dataset(
+		filepaths=filepaths,
+		metadata=read_metadata(dirpath_input),
+		batch_size=None,
+		shuffle=False, #even with shuffle=False we're not gonna get them all in the same order since we're reading in parallel
+		**kwargs
+	)
+
+if __name__ == "__main__":
+	ds_train, ds_validate = dataset_mono("/mnt/research-data/main/rainfallwater_records-viperfinal/")
+	for thing in ds_validate():
+		as_str = str(thing)
+		print(thing[:200])