{ "cells": [ { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# setup autoload\n", "%load_ext autoreload\n", "%autoreload 2" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import ast\n", "import os\n", "import shutil\n", "import sys\n", "from collections import defaultdict\n", "import json\n", "\n", "import numpy as np\n", "import pandas as pd\n", "from preference_data_preparation_diff import *\n", "from sklearn.model_selection import train_test_split\n", "from suno_utils.utils.s3 import download_s3_files\n", "from suno_utils.utils.text import read_json, read_jsonl, write_json, write_jsonl\n", "from tqdm import tqdm\n", "import matplotlib.pyplot as plt\n", "from suno_utils.audio import Audio\n", "from suno_analytics.preference_helper import get_preference_counts\n", "\n", "pd.set_option(\"display.max_rows\", 500)\n", "pd.set_option(\"display.max_columns\", 500)\n", "pd.set_option(\"display.width\", 1000)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "OUT_DATA_DIR = \"/app/suno/data/dpo/diff_v6_t3_comb/\"\n", "os.makedirs(OUT_DATA_DIR, exist_ok=True)\n", "NPZ_DIR = \"/app/suno/data/dpo/diff_v6\"" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "df = pd.read_pickle(\n", " \"/home/tony/Data/Preference/up_v6/interesting_clips_up_u_6_20250320_full.pkl\"\n", ") # , engine='python')\n", "print(\"Preference data shape\", df.shape)\n", "print(\"unique users\", df[\"user_id\"].nunique())" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "print(df[\"is_public\"].value_counts())\n", "# # remove public for now cause fucking users\n", "# df = df[~df[\"is_public\"]]\n", "# print(df[\"is_public\"].value_counts())" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "df[\"upsample_clip_id\"] = df[\"metadata\"].apply(lambda x: x.get(\"upsample_clip_id\", \"\"))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "all_converted_paths = os.listdir(NPZ_DIR)\n", "print(len(all_converted_paths))\n", "\n", "converted_paths = set(\n", " [f.replace(\".npz\", \"\") for f in all_converted_paths if \"vae\" not in f]\n", ")\n", "print(len(converted_paths))\n", "vae_converted_paths = set(\n", " [f.replace(\"_vae.npz\", \"\") for f in all_converted_paths if \"vae\" in f]\n", ")\n", "print(len(vae_converted_paths))\n", "\n", "print(\"pre-downloaded df\", df.shape)\n", "df[df[\"upsample_clip_id\"].isin(converted_paths)].shape\n", "df = df[df[\"upsample_clip_id\"].isin(converted_paths)].copy()\n", "print(\"downloaded df\", df.shape)\n", "df[df[\"s3_id\"].isin(vae_converted_paths)].shape\n", "df = df[df[\"s3_id\"].isin(vae_converted_paths)].copy()\n", "print(\"vae downloaded df\", df.shape)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "df[\"is_up\"] = df[\"model_name\"].str.contains(\"up\")\n", "df[\"is_up\"].value_counts()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# LET's do the data prep" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "## for 13b this is easy for now\n", "print(df.groupby([\"preference\"])[\"model_name\"].value_counts())\n", "print(df.shape)\n", "df = df[df[\"model_name\"].isin([\"chirp-v4-up-u-6\"])]\n", "print(df.shape)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "print(df.shape)\n", "df = df[\n", " df[\"request_id\"].isin(\n", " df[\"request_id\"].value_counts().index[df[\"request_id\"].value_counts() == 2]\n", " )\n", "]\n", "print(df.shape)\n", "print(df.groupby([\"preference\"])[\"model_name\"].value_counts())\n", "assert df.shape[0] == df[\"request_id\"].nunique() * 2" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "with open(\"/home/tony/Data/Preference/up_v6/full_pair_quality.json\", \"r\") as file:\n", " full_pair_quality = json.load(file)\n", "print(\"Total pair quality scores:\", len(full_pair_quality))\n", "\n", "unpacked_pair_quality = {}\n", "for request_id, pairs_of_qualities in full_pair_quality.items():\n", " for clip_id, pair_quality in pairs_of_qualities.items():\n", " unpacked_pair_quality[clip_id] = pair_quality\n", "print(\"Total unpacked pair quality scores:\", len(unpacked_pair_quality))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "clip_diffs = []\n", "clip_ratios = []\n", "for request_id, pairs_of_qualities in full_pair_quality.items():\n", " mean_neg_scores = []\n", " mean_pos_scores = []\n", " for i, (clip_id, pair_quality) in enumerate(pairs_of_qualities.items()):\n", " if i % 2 == 0:\n", " mean_neg_scores.append(np.mean(pair_quality[\"ear_v2_quality_scores\"] if pair_quality else 0))\n", " if i % 2 == 1:\n", " mean_pos_scores.append(np.mean(pair_quality[\"ear_v2_quality_scores\"] if pair_quality else 0))\n", " ratios = [(pos - neg) / (pos + 0.0001) for pos, neg in zip(mean_pos_scores, mean_neg_scores)]\n", " pos_diffs = [(pos - prev_pos) / (prev_pos + 0.0001) for prev_pos, pos in zip(mean_pos_scores, mean_pos_scores[1:])]\n", " neg_diffs = [(neg - prev_neg) / (prev_neg + 0.0001) for prev_neg, neg in zip(mean_neg_scores, mean_neg_scores[1:])]\n", " \n", " for i in range(1, len(ratios)):\n", " clip_ratios.append(ratios[i])\n", " clip_diffs.append(pos_diffs[i - 1] - neg_diffs[i - 1])\n", "x = clip_ratios\n", "y = clip_diffs" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Create the 2D histogram (heatmap)\n", "plt.figure(figsize=(10, 8))\n", "\n", "# Create a 2D histogram\n", "bin_edges = np.linspace(-0.5, 0.5, 101) # 30 bins from -1 to 1\n", "hist, x_edges, y_edges = np.histogram2d(\n", " x, y, \n", " bins=[bin_edges, bin_edges], # Same bins for both x and y\n", " range=[[-0.5, 0.5], [-0.5, 0.5]] # Ensure range is from -1 to 1 for both axes\n", ")\n", "\n", "# Create a heatmap using pcolormesh for better control\n", "X, Y = np.meshgrid(x_edges[:-1], y_edges[:-1])\n", "plt.pcolormesh(X, Y, hist.T, cmap='viridis', shading='auto')\n", "\n", "# Add a color bar\n", "cbar = plt.colorbar()\n", "cbar.set_label('Counts', rotation=270, labelpad=20, fontsize=12)\n", "\n", "# Add labels and title\n", "plt.xlabel('Clip quality diff ratios', fontsize=12)\n", "plt.ylabel('Clip quality diff ratio difference with prev', fontsize=12)\n", "plt.title('2D Histogram (Heatmap) of Correlated Data', fontsize=14)\n", "\n", "# Show the plot\n", "plt.tight_layout()\n", "plt.savefig('2d_histogram.png', dpi=300) # Save to file (optional)\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "_ = plt.hist(clip_diffs, bins=np.linspace(-1, 1, 100))\n", "for percentage in [0.1, 0.2, 0.5, 0.8, 0.9]:\n", " print(percentage, \"--->\", np.quantile(sorted(clip_diffs), percentage))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "_ = plt.hist(clip_ratios, bins=np.linspace(-1, 1, 100))\n", "for percentage in [0.1, 0.2, 0.5, 0.8, 0.9]:\n", " print(percentage, \"--->\", np.quantile(sorted(clip_ratios), percentage))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "def get_audio_quality_measures(s3_id):\n", " audio_quality = unpacked_pair_quality.get(s3_id, [])\n", " if not audio_quality:\n", " return [None for _ in range(11)]\n", " return [\n", " np.mean(audio_quality[\"ear_v2_quality_scores\"]), # float(audio_quality[\"ear_v2_quality_scores\"]),\n", " float(audio_quality[\"shimmer_score\"]),\n", " float(audio_quality[\"loudness_factor\"]),\n", " audio_quality[\"spectral_character\"],\n", " float(audio_quality[\"spectral_centroid\"]),\n", " float(audio_quality[\"bass_ratio\"]),\n", " float(audio_quality[\"mid_ratio\"]),\n", " float(audio_quality[\"high_ratio\"]),\n", " float(audio_quality[\"stereo_width\"]),\n", " int(audio_quality[\"total_clips\"]),\n", " float(audio_quality[\"clips_per_second\"]),\n", " ]\n", "\n", "\n", "df[\n", " [\n", " \"pair_quality\",\n", " \"total_shimmer_score\",\n", " \"loudness_factor\",\n", " \"spectral_character\",\n", " \"spectral_centroid\",\n", " \"bass_ratio\",\n", " \"mid_ratio\",\n", " \"high_ratio\",\n", " \"stereo_width\",\n", " \"total_clips\",\n", " \"clips_per_second\",\n", " ]\n", "] = pd.DataFrame(df[\"s3_id\"].apply(get_audio_quality_measures).tolist(), index=df.index)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "print(df.shape)\n", "df = df.dropna(subset=[\n", " \"pair_quality\",\n", " \"total_shimmer_score\",\n", " \"loudness_factor\",\n", " \"spectral_character\",\n", " \"spectral_centroid\",\n", " \"bass_ratio\",\n", " \"mid_ratio\",\n", " \"high_ratio\",\n", " \"stereo_width\",\n", " \"total_clips\",\n", " \"clips_per_second\",\n", " ])\n", "print(df.shape)\n", "df = df[\n", " df[\"request_id\"].isin(\n", " df[\"request_id\"].value_counts().index[df[\"request_id\"].value_counts() == 2]\n", " )\n", "]\n", "print(df.shape)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Let's use the old selection for now -- for quality assurance\n", "# expand the metadata columns -- this takes forever...~ 6 mins\n", "# test_slice = df[\"metadata\"].apply(lambda x: ast.literal_eval(str(x)))\n", "# test_slice = df[\"metadata\"] # .apply(lambda x: custom_parse(x))\n", "# test_slice_series = test_slice.apply(pd.Series)\n", "# df = pd.concat([df, test_slice_series], axis=1, join=\"inner\")\n", "print(\"unique_requests\", df[\"request_id\"].nunique())" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "df = df.loc[:, ~df.columns.duplicated()].copy()\n", "# get the original duration of the clips, if they are concacted\n", "df[\"original_duration_s\"] = df[\"total_start_s\"] + df[\"duration\"]\n", "# classify the continue at behavoirs by the duration choice\n", "audio_prompt_id_to_continue_at = {}\n", "for _, row in df[~df[\"continued_parent\"].isna()].iterrows():\n", " audio_prompt_id = row[\"continued_parent\"]\n", " if audio_prompt_id not in audio_prompt_id_to_continue_at:\n", " audio_prompt_id_to_continue_at[audio_prompt_id] = row[\"continue_at\"]\n", " else:\n", " # pick the max\n", " audio_prompt_id = max(\n", " audio_prompt_id_to_continue_at[audio_prompt_id], row[\"continue_at\"]\n", " )\n", "print(len(audio_prompt_id_to_continue_at))\n", "df[\"has_continue_and_start_continue_at\"] = df[\"s3_id\"].apply(\n", " lambda x: audio_prompt_id_to_continue_at.get(x)\n", ")\n", "# we want continue at to be at most of the clip...\n", "df[\"good_continue_at\"] = (\n", " (df[\"has_continue_and_start_continue_at\"] / df[\"duration\"]) > 0.9\n", ") | df[\"has_continue_and_start_continue_at\"].isna()\n", "print(df[\"good_continue_at\"].value_counts())\n", "\n", "\n", "print(\n", " \"\\n Check some basics... \\n\",\n", " df[\"preference\"].value_counts(),\n", " df[\"model_name\"].value_counts(),\n", " df.groupby([\"preference\"])[\"model_name\"].value_counts(),\n", ")\n", "\n", "df = df.sort_values(by=[\"request_id\", \"preference\"])\n", "df[\"duration_rel_diff\"] = df[\"duration\"].diff()\n", "df[\"play_rel_diff\"] = df[\"reaction_play_count\"].diff()\n", "print(df[\"task\"].value_counts())" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "df = df.sort_values(by=[\"request_id\", \"preference\", \"diff_preference\"])\n", "df[\"pos_diff_preference\"] = df[\"diff_preference\"].diff()\n", "df[df[\"preference\"]][\"pos_diff_preference\"].value_counts()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "lookup_percentiles = [5, 10, 20, 50, 80, 90, 95]\n", "df[\"shimmer_score_diff\"] = df[\"total_shimmer_score\"].diff()\n", "percentiles = np.percentile(\n", " df[df[\"preference\"]][\"shimmer_score_diff\"].dropna(), lookup_percentiles\n", ")\n", "plt.hist(\n", " df[df[\"preference\"]][\"shimmer_score_diff\"],\n", " label=f\"pos, mean: {np.mean(df[df['preference']]['shimmer_score_diff']):.2f}\",\n", " bins=np.linspace(-1, 1, 100),\n", " alpha=0.5,\n", ")\n", "textstr = \"\\n\".join(\n", " [\n", " f\"{lookup_percentiles[i]}th: {percentile:.2f}\"\n", " for i, percentile in enumerate(percentiles)\n", " ]\n", ")\n", "plt.gcf().text(\n", " 0.15,\n", " 0.98,\n", " textstr,\n", " fontsize=10,\n", " verticalalignment=\"top\",\n", " horizontalalignment=\"left\",\n", " bbox=dict(facecolor=\"white\", alpha=0.5),\n", ")\n", "for percentile in percentiles:\n", " plt.axvline(x=percentile, color=\"r\", linestyle=\"dashed\", linewidth=1)\n", "plt.title(\n", " f\"Shimmer score difference --> {lookup_percentiles[-1]}th, {percentiles[-1]:.2f}\"\n", ")\n", "# plt.yscale(\"log\")\n", "plt.legend()\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "lookup_percentiles = [5, 10, 20, 50, 80, 90, 95]\n", "percentiles = np.percentile(\n", " df[df[\"preference\"]][\"pair_quality\"].dropna(), lookup_percentiles\n", ")\n", "plt.hist(\n", " df[df[\"preference\"]][\"pair_quality\"],\n", " label=f\"pos, mean: {np.mean(df[df['preference']]['pair_quality']):.2f}\",\n", " bins=np.linspace(10, 30, 100),\n", " alpha=0.5,\n", ")\n", "plt.hist(\n", " df[~df[\"preference\"]][\"pair_quality\"],\n", " label=f\"neg, mean: {np.mean(df[df['preference']]['pair_quality']):.2f}\",\n", " bins=np.linspace(10, 30, 100),\n", " alpha=0.5,\n", ")\n", "textstr = \"\\n\".join(\n", " [\n", " f\"{lookup_percentiles[i]}th: {percentile:.2f}\"\n", " for i, percentile in enumerate(percentiles)\n", " ]\n", ")\n", "plt.gcf().text(\n", " 0.15,\n", " 0.98,\n", " textstr,\n", " fontsize=10,\n", " verticalalignment=\"top\",\n", " horizontalalignment=\"left\",\n", " bbox=dict(facecolor=\"white\", alpha=0.5),\n", ")\n", "for percentile in percentiles:\n", " plt.axvline(x=percentile, color=\"r\", linestyle=\"dashed\", linewidth=1)\n", "plt.title(f\"Pair quality --> {lookup_percentiles[0]}th, {percentiles[0]:.2f}\")\n", "# plt.yscale(\"log\")\n", "plt.legend()\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "df[\"pair_quality_diff\"] = df[\"pair_quality\"].diff() / df[\"pair_quality\"]" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "lookup_percentiles = [5, 10, 20, 50, 80, 90, 95]\n", "percentiles = np.percentile(\n", " df[df[\"preference\"]][\"pair_quality_diff\"].dropna(), lookup_percentiles\n", ")\n", "plt.hist(\n", " df[df[\"preference\"]][\"pair_quality_diff\"],\n", " label=f\"pos, mean: {np.mean(df[df['preference']]['pair_quality_diff']):.2f}\",\n", " bins=np.linspace(-0.5, 0.5, 100),\n", " alpha=0.5,\n", ")\n", "textstr = \"\\n\".join(\n", " [\n", " f\"{lookup_percentiles[i]}th: {percentile:.2f}\"\n", " for i, percentile in enumerate(percentiles)\n", " ]\n", ")\n", "plt.gcf().text(\n", " 0.15,\n", " 0.98,\n", " textstr,\n", " fontsize=10,\n", " verticalalignment=\"top\",\n", " horizontalalignment=\"left\",\n", " bbox=dict(facecolor=\"white\", alpha=0.5),\n", ")\n", "for percentile in percentiles:\n", " plt.axvline(x=percentile, color=\"r\", linestyle=\"dashed\", linewidth=1)\n", "plt.title(f\"Pair quality diff --> {lookup_percentiles[0]}th, {percentiles[0]:.2f}\")\n", "# plt.yscale(\"log\")\n", "plt.legend()\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "subset_requests = df[(df[\"pair_quality_diff\"] > 0.2) & (df[\"preference\"])][\"request_id\"].unique()\n", "df[df[\"request_id\"].isin(subset_requests)][[\"id\", \"request_id\", \"pair_quality\", \"preference\", \"total_shimmer_score\"]]" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# CS said -1 means stero to mono; 1 means mono to stereo\n", "# only cut off the left side\n", "df[\"stereo_width_diff\"] = df[\"stereo_width\"].diff()\n", "lookup_percentiles = [5, 10, 20, 50, 80, 90, 95]\n", "percentiles = np.percentile(\n", " df[df[\"preference\"]][\"stereo_width_diff\"].dropna(), lookup_percentiles\n", ")\n", "plt.hist(\n", " df[df[\"preference\"]][\"stereo_width_diff\"],\n", " label=f\"pos, mean: {np.mean(df[df['preference']]['stereo_width_diff']):.2f}\",\n", " bins=np.linspace(-1, 1, 400),\n", " alpha=0.5,\n", ")\n", "textstr = \"\\n\".join(\n", " [\n", " f\"{lookup_percentiles[i]}th: {percentile:.2f}\"\n", " for i, percentile in enumerate(percentiles)\n", " ]\n", ")\n", "plt.gcf().text(\n", " 0.15,\n", " 0.98,\n", " textstr,\n", " fontsize=10,\n", " verticalalignment=\"top\",\n", " horizontalalignment=\"left\",\n", " bbox=dict(facecolor=\"white\", alpha=0.5),\n", ")\n", "for percentile in percentiles:\n", " plt.axvline(x=percentile, color=\"r\", linestyle=\"dashed\", linewidth=1)\n", "plt.title(\n", " f\"Stereo Width Difference --> {lookup_percentiles[0]}th, {percentiles[0]:.2f}\"\n", ")\n", "# plt.yscale(\"log\")\n", "plt.legend()\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# CS said no tails is good\n", "lookup_percentiles = [5, 10, 20, 50, 80, 90, 95]\n", "percentiles = np.percentile(\n", " df[df[\"preference\"]][\"spectral_centroid\"].dropna(), lookup_percentiles\n", ")\n", "percentiles = np.percentile(\n", " df[~df[\"preference\"]][\"spectral_centroid\"].dropna(), lookup_percentiles\n", ")\n", "plt.hist(\n", " df[df[\"preference\"]][\"spectral_centroid\"],\n", " label=f\"pos, mean: {np.mean(df[df['preference']]['spectral_centroid']):.2f}\",\n", " bins=np.linspace(0, 8000, 400),\n", " alpha=0.5,\n", ")\n", "plt.hist(\n", " df[~df[\"preference\"]][\"spectral_centroid\"],\n", " label=f\"neg, mean: {np.mean(df[~df['preference']]['spectral_centroid']):.2f}\",\n", " bins=np.linspace(0, 8000, 400),\n", " alpha=0.5,\n", ")\n", "textstr = \"\\n\".join(\n", " [\n", " f\"{lookup_percentiles[i]}th: {percentile:.2f}\"\n", " for i, percentile in enumerate(percentiles)\n", " ]\n", ")\n", "plt.gcf().text(\n", " 0.15,\n", " 0.98,\n", " textstr,\n", " fontsize=10,\n", " verticalalignment=\"top\",\n", " horizontalalignment=\"left\",\n", " bbox=dict(facecolor=\"white\", alpha=0.5),\n", ")\n", "for percentile in percentiles:\n", " plt.axvline(x=percentile, color=\"r\", linestyle=\"dashed\", linewidth=1)\n", "plt.title(\"Spectral Centroid\")\n", "# plt.yscale(\"log\")\n", "plt.legend()\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# CS said no tails is good\n", "# take the relative centroid diff\n", "df[\"spectral_centroid_diff\"] = df[\"spectral_centroid\"].diff() / df[\"spectral_centroid\"]\n", "lookup_percentiles = [5, 10, 20, 50, 80, 90, 95]\n", "percentiles = np.percentile(\n", " df[df[\"preference\"]][\"spectral_centroid_diff\"].dropna(), lookup_percentiles\n", ")\n", "plt.hist(\n", " df[df[\"preference\"]][\"spectral_centroid_diff\"],\n", " label=f\"pos, mean: {np.mean(df[df['preference']]['spectral_centroid_diff']):.2f}\",\n", " bins=np.linspace(-1, 1, 400),\n", " alpha=0.5,\n", ")\n", "textstr = \"\\n\".join(\n", " [\n", " f\"{lookup_percentiles[i]}th: {percentile:.2f}\"\n", " for i, percentile in enumerate(percentiles)\n", " ]\n", ")\n", "plt.gcf().text(\n", " 0.15,\n", " 0.98,\n", " textstr,\n", " fontsize=10,\n", " verticalalignment=\"top\",\n", " horizontalalignment=\"left\",\n", " bbox=dict(facecolor=\"white\", alpha=0.5),\n", ")\n", "for percentile in percentiles:\n", " plt.axvline(x=percentile, color=\"r\", linestyle=\"dashed\", linewidth=1)\n", "plt.title(\n", " f\"Spectral Centroid Difference ratio --> {lookup_percentiles[-1]}th, {percentiles[-1]:.2f}\"\n", ")\n", "# plt.yscale(\"log\")\n", "plt.legend()\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "df = df.loc[:, ~df.columns.duplicated()].copy()\n", "# get the original duration of the clips, if they are concacted\n", "df[\"original_duration_s\"] = df[\"total_start_s\"] + df[\"duration\"]\n", "# classify the continue at behavoirs by the duration choice\n", "audio_prompt_id_to_continue_at = {}\n", "for _, row in df[~df[\"continued_parent\"].isna()].iterrows():\n", " audio_prompt_id = row[\"continued_parent\"]\n", " if audio_prompt_id not in audio_prompt_id_to_continue_at:\n", " audio_prompt_id_to_continue_at[audio_prompt_id] = row[\"continue_at\"]\n", " else:\n", " # pick the max\n", " audio_prompt_id = max(\n", " audio_prompt_id_to_continue_at[audio_prompt_id], row[\"continue_at\"]\n", " )\n", "print(len(audio_prompt_id_to_continue_at))\n", "df[\"has_continue_and_start_continue_at\"] = df[\"s3_id\"].apply(\n", " lambda x: audio_prompt_id_to_continue_at.get(x)\n", ")\n", "# we want continue at to be at most of the clip...\n", "df[\"good_continue_at\"] = (\n", " (df[\"has_continue_and_start_continue_at\"] / df[\"duration\"]) > 0.9\n", ") | df[\"has_continue_and_start_continue_at\"].isna()\n", "print(df[\"good_continue_at\"].value_counts())\n", "\n", "\n", "print(\n", " \"\\n Check some basics... \\n\",\n", " df[\"preference\"].value_counts(),\n", " df[\"model_name\"].value_counts(),\n", " df.groupby([\"preference\"])[\"model_name\"].value_counts(),\n", ")\n", "\n", "df = df.sort_values(by=[\"request_id\", \"preference\"])\n", "df[\"duration_rel_diff\"] = df[\"duration\"].diff()\n", "df[\"play_rel_diff\"] = df[\"reaction_play_count\"].diff()\n", "print(df[\"task\"].value_counts())" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "plt.hist(\n", " df[df[\"preference\"]][\"total_shimmer_score\"],\n", " label=f\"pos, mean: {np.mean(df[df['preference']]['total_shimmer_score']):.2f}\",\n", " bins=np.linspace(0, 10, 100),\n", " alpha=0.5,\n", ")\n", "plt.hist(\n", " df[~df[\"preference\"]][\"total_shimmer_score\"],\n", " label=f\"neg, mean: {np.mean(df[~df['preference']]['total_shimmer_score']):.2f}\",\n", " bins=np.linspace(0, 10, 100),\n", " alpha=0.5,\n", ")\n", "percentiles = np.percentile(df[df[\"preference\"]][\"total_shimmer_score\"].dropna(), [50, 75, 90])\n", "for percentile in percentiles:\n", " # print(percentile)\n", " plt.axvline(x=percentile, color=\"r\", linestyle=\"dashed\", linewidth=1)\n", "# plt.yscale(\"log\")\n", "plt.title(f\"Shimmer score\")\n", "plt.legend()\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "normal_pos_play_count = 3\n", "# this is lower, cause a concat is probably already ensuring that it is good\n", "concat_pos_play_count = 2\n", "# this is a filter on the concated clip\n", "concat_total_play_count = 3\n", "\n", "neg_filter_selection_mask = (\n", " (~df[\"preference\"]) # get basics aligned\n", " & (df[\"reaction_play_count\"] >= 1) # has to be played once\n", " # & (df[\"play_count\"] <= 3) # if it is actually bad, shouldn't be listened often\n", " & (df[\"duration\"] >= 30) # can't be too short, otherwise it is obvious\n", " # & (df[\"duration\"] <= 60) # can't be badly long\n", " & (df[\"has_continue_and_start_continue_at\"].isna()) # won't have any continues\n", " & (df[\"norm_play_frac\"] <= 2.1)\n", " # & (df[\"sum_total_play_duration_5\"] >= 31)\n", " # & (df[\"dislike_count\"] >= 1) # this is kinda strict\n", " # & (\n", " # (df_slice[\"is_in_playlist\"] == False)\n", " # & (df_slice[\"concat_in_playlist\"] == False)\n", " # ) # can't be part of a playlist -- otherwise there are some like signal in it?\n", ")\n", "pos_filter_selectin_mask = (\n", " (df[\"preference\"]) # get basics aligned\n", " & (\n", " df[\"good_continue_at\"]\n", " ) # if continue, needs to continue off a certain percentage\n", " & (df[\"reaction_play_count\"] >= 1)\n", " & (df[\"play_rel_diff\"] >= 0) # this is more like quality assurance\n", " & (df[\"duration\"] >= 30) # can't be too short, otherwise it is obvious\n", " & (df[\"dislike_count\"] == 0) # can't have dislikes\n", " & (df[\"flag_count\"] == 0) # can't have issues\n", " & (\n", " (\n", " (df[\"part_of_concat\"])\n", " & (df[\"reaction_play_count\"] >= concat_pos_play_count)\n", " & (df[\"concat_play_counts\"] >= concat_total_play_count)\n", " )\n", " | (\n", " (~df[\"part_of_concat\"])\n", " & (df[\"reaction_play_count\"] >= normal_pos_play_count)\n", " & (df[\"norm_play_frac\"] >= 2.1) # this is a bit of a luxury cut...\n", " & (df[\"sum_total_play_duration_5\"] >= 31)\n", " )\n", " )\n", " # & (df[\"norm_play_frac\"] >= 1.9)\n", " # & (df[\"user_n_clips\"] >= 100) # user needs to have genereated at least 20\n", " # & (df[\"duration_rel_diff\"] < 10) # positive isn't just longer\n", " # & ((df[\"task\"] == \"\") | (df[\"task\"] == \"extend\"))\n", " & (\n", " (df[\"upvote_count\"] >= 1)\n", " | (df[\"reaction_play_count\"] >= 5)\n", " | (df[\"concat_play_counts\"] >= 5)\n", " )\n", " # & (df[\"pos_diff_preference\"] == 2)\n", " # & ((0 < df[\"similarity\"]) & (df[\"similarity\"] <= 0.99))\n", " # & (\n", " # (df[\"cer_diff_preference\"] < 0.25) & (df[\"cer\"] < 0.8)\n", " # ) # cut on hoot cer difference and abs cer\n", " # & (df[\"pair_quality\"] > 0.31) # bottom 5%\n", " # & ((df[\"total_shimmer_score\"] < 1) | (df[\"shimmer_score_diff\"] < 0.4))\n", " # & (df[\"stereo_width_diff\"] > -0.2) # cut off bottom 5%\n", " # & (df[\"spectral_centroid_diff\"] < 0.25) # crop off the top 5%\n", ")\n", "print(\n", " \"negative\",\n", " sum(neg_filter_selection_mask),\n", " \"positive\",\n", " sum(pos_filter_selectin_mask),\n", ")\n", "\n", "neg_filter_requests = df[neg_filter_selection_mask][\"request_id\"].unique()\n", "pos_filter_requests = df[pos_filter_selectin_mask][\"request_id\"].unique()\n", "# looking for very strong signal here:\n", "# listen to the positive/negative more than once\n", "# disliked one of the clips\n", "unique_requests = set(pos_filter_requests).intersection(neg_filter_requests)\n", "print(\n", " \"total pair requests\",\n", " df[\"request_id\"].nunique(),\n", " \"selected pair requests\",\n", " len(unique_requests),\n", " f\"frac {len(unique_requests) / df['request_id'].nunique():.3f}\",\n", ")" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "df_slice = df[df[\"request_id\"].isin(set(unique_requests))].copy()\n", "print(\n", " f\"{os.path.basename(OUT_DATA_DIR)} requests\",\n", " df_slice[\"request_id\"].nunique(),\n", " \"clips\",\n", " df_slice.shape[0],\n", " f\"total khrs {sum(df_slice['duration'] / 3600 / 1000):.3f};\",\n", " f\"N gpus for 1000 iters {df_slice.shape[0] / 8 / 2 / 1000:.3f};\",\n", " f\"4 gpus for x iters {df_slice.shape[0] / 8 / 2 / 4:.3f};\",\n", " f\"n unique users {df_slice['user_id'].nunique()}\",\n", " f\"n pro users {df_slice[df_slice['is_pro_user']]['user_id'].nunique()}\",\n", ")\n", "# up t7 requests 37735 clips 75470 total khrs 3.964; N gpus for 1000 iters 4.717; 4 gpus for x iters 1179.219; n unique users 19093 n pro users 16217\n", "# up t17 requests 49262 clips 98524 total khrs 5.187; N gpus for 1000 iters 6.158; 4 gpus for x iters 1539.438; n unique users 24093 n pro users 20326\n", "# up v2 t1 requests 6008 clips 12016 total khrs 0.642; N gpus for 1000 iters 0.751; 4 gpus for x iters 187.750; n unique users 3846 n pro users 3645\n", "# up v2 t2 requests 10772 clips 21544 total khrs 1.154; N gpus for 1000 iters 1.347; 4 gpus for x iters 336.625; n unique users 6527 n pro users 6027\n", "# up v3 t10 requests 27102 clips 54204 total khrs 2.938; N gpus for 1000 iters 3.388; 4 gpus for x iters 846.938; n unique users 13932 n pro users 12187\n", "# up v4 t1 requests 3201 clips 6402 total khrs 0.343; N gpus for 1000 iters 0.400; 4 gpus for x iters 100.031; n unique users 2363 n pro users 2321\n", "# up v4 t2 requests 12818 clips 25636 total khrs 1.371; N gpus for 1000 iters 1.602; 4 gpus for x iters 400.562; n unique users 7753 n pro users 7524\n", "# up v4 t3 requests 15332 clips 30664 total khrs 1.637; N gpus for 1000 iters 1.917; 4 gpus for x iters 479.125; n unique users 8960 n pro users 8656\n", "# up v4 t4 requests 18368 clips 36736 total khrs 1.962; N gpus for 1000 iters 2.296; 4 gpus for x iters 574.000; n unique users 10427 n pro users 10049\n", "# up v4 t5 requests 22037 clips 44074 total khrs 2.347; N gpus for 1000 iters 2.755; 4 gpus for x iters 688.656; n unique users 12121 n pro users 11551\n", "# up v4 t6 requests 27374 clips 54748 total khrs 2.915; N gpus for 1000 iters 3.422; 4 gpus for x iters 855.438; n unique users 14463 n pro users 13678\n", "# up v4 t7 requests 31797 clips 63594 total khrs 3.384; N gpus for 1000 iters 3.975; 4 gpus for x iters 993.656; n unique users 16398 n pro users 15325\n", "# up v5 t1 requests 17035 clips 34070 total khrs 1.788; N gpus for 1000 iters 2.129; 4 gpus for x iters 532.344; n unique users 10858 n pro users 8065\n", "# up v5 t2 requests 20030 clips 40060 total khrs 2.108; N gpus for 1000 iters 2.504; 4 gpus for x iters 625.938; n unique users 12395 n pro users 9156\n", "# up v6 t1 requests 17941 clips 35882 total khrs 1.889; N gpus for 1000 iters 2.243; 4 gpus for x iters 560.656; n unique users 11471 n pro users 8374\n", "# up v6 t2 requests 23563 clips 47126 total khrs 2.473; N gpus for 1000 iters 2.945; 4 gpus for x iters 736.344; n unique users 14550 n pro users 10353\n", "# up v6 t3 requests 34761 clips 69522 total khrs 3.647; N gpus for 1000 iters 4.345; 4 gpus for x iters 1086.281; n unique users 20264 n pro users 14000" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "test_mask = (df_slice[\"preference\"]) & (\n", " (df_slice[\"is_in_playlist\"]) | (df_slice[\"concat_in_playlist\"])\n", ")\n", "print(\"positive in playlist\", df_slice[test_mask].shape)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# save positive ids\n", "# positive_preference_ids = df_slice[df_slice[\"preference\"] == False][\"s3_id\"].to_json(orient='values')\n", "# with open('/home/tony/Data/Preference/7b_v2/7v_v20_full_recut_id_negative.json', 'w') as file:\n", "# file.write(positive_preference_ids)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# def modify_model_name(model_name, metadata):\n", "# if (\n", "# model_name.startswith(\"chirp-v3p5-engine-t\")\n", "# or model_name.startswith(\"chirp-v3p5-engine-s\")\n", "# or model_name.startswith(\"chirp-v4\")\n", "# or model_name.startswith(\"chirp-v3p5-h-s-31\")\n", "# ):\n", "# if \"param_experiment\" in metadata:\n", "# exp = metadata.get(\"param_experiment\", \"\")\n", "# if exp:\n", "# return f\"{model_name}_{exp}\"\n", "# return model_name\n", "\n", "# metrics_check_df_slice = df_slice.copy()\n", "# metrics_check_df_slice[\"model_name\"] = metrics_check_df_slice.apply(\n", "# lambda row: modify_model_name(row[\"model_name\"], row[\"metadata\"]), axis=1\n", "# )\n", "# get_preference_counts(metrics_check_df_slice)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "print(df_slice[\"source\"].value_counts())" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# df_slice.to_csv(\"/home/tony/Data/Preference/13b_v0/interesting_clips_v3p5_s_8_20240828_slice.csv\")\n", "BREAK" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Need to kick out the ones has gpt prompt -- these are pairs with different text inputs" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "final_filtered_requests = df_slice[\"request_id\"].unique()\n", "print(len(final_filtered_requests))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "train_requests, val_requests = train_test_split(\n", " sorted(list(final_filtered_requests)), test_size=0.01, random_state=42\n", ")\n", "print(len(train_requests), len(val_requests))\n", "\n", "train_df = df_slice[df_slice[\"request_id\"].isin(set(train_requests))].copy()\n", "val_df = df_slice[df_slice[\"request_id\"].isin(set(val_requests))].copy()\n", "train_df = train_df.sort_values(by=[\"request_id\", \"preference\"])\n", "train_df = train_df # .reset_index()\n", "val_df = val_df.sort_values(by=[\"request_id\", \"preference\"])\n", "val_df = val_df # .reset_index()\n", "\n", "print(train_df.shape, val_df.shape)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# BREAK" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Actually make" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# val_df[[\"request_id\", \"metadata\", \"updated_at\", \"user_id\", \"preference\"]].head()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "total_duration = 0\n", "for i, row in tqdm(train_df.iterrows(), total=len(train_df)):\n", " # we need to alternate between preference: neg, pos\n", " # print(i, row)\n", " try:\n", " assert row[\"preference\"] == (i % 2 == 1)\n", " except:\n", " print(i, row)\n", " total_duration += row[\"duration\"]\n", "print(\n", " f\"{round(total_duration / 60 / 60):,} hours of {train_df.shape[0]} clips, {train_df.shape[0] / 8 / 2 / 1000} nodes, {train_df.shape[0] / 8 / 2 / 4} steps\"\n", ")" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "make_dataset(\n", " val_df,\n", " OUT_DATA_DIR,\n", " is_val=True,\n", " npz_dir=NPZ_DIR,\n", " do_extend_chunks=True,\n", " clip_id_to_quality_scores=unpacked_pair_quality,\n", ")" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "make_dataset(\n", " train_df,\n", " OUT_DATA_DIR,\n", " is_val=False,\n", " npz_dir=NPZ_DIR,\n", " do_extend_chunks=True,\n", " clip_id_to_quality_scores=unpacked_pair_quality,\n", ")" ] }, { "cell_type": "markdown", "metadata": { "ExecuteTime": { "end_time": "2024-01-29T19:46:47.549860Z", "start_time": "2024-01-29T19:46:47.548015Z" } }, "source": [ "# Validation" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# verify\n", "metas_val = read_jsonl(os.path.join(OUT_DATA_DIR, \"metas_val.jsonl\"))\n", "print(len(metas_val) / 2)\n", "mm_semantic_val = np.memmap(\n", " os.path.join(OUT_DATA_DIR, \"data_semantic_val.bin\"), dtype=np.uint16, mode=\"r\"\n", ")\n", "mm_vae_val = np.memmap(\n", " os.path.join(OUT_DATA_DIR, \"data_vae_val.bin\"), dtype=np.float16, mode=\"r\"\n", ")\n", "\n", "\n", "mm_vae_val = mm_vae_val.reshape(-1, VAE_MEMMAP_SIZE, VAE_DIM)\n", "print(mm_vae_val.shape)\n", "\n", "mm_semantic_val = mm_semantic_val.reshape(-1, SEMANTIC_MEMMAP_SIZE)\n", "print(mm_semantic_val.shape)\n", "\n", "assert len(metas_val) == mm_vae_val.shape[0] == mm_semantic_val.shape[0]" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# # load codec for decoding\n", "# from suno_utils.tasks.dac_vae_100hz_peaq import ( # NOTE: works for 25hz as well\n", "# preload_models as preload_codec_models,\n", "# decode as codec_decode,\n", "# encode as codec_encode,\n", "# get_embedding_rate,\n", "# load_model as load_codec_model,\n", "# )\n", "\n", "# CODEC_FILEPATH = \"s3://suno-data/christian/25hz_vae_peaq_kl_0.005.pth\"\n", "# preload_codec_models(CODEC_FILEPATH)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# # decode some audio\n", "idx = 108\n", "# # ensure even index\n", "assert idx % 2 == 0\n", "# print(metas_val[idx])\n", "# print(\"negative\")\n", "# audio = codec_decode(mm_vae_val[idx])\n", "# audio.normalize_volume().play()\n", "\n", "# print(metas_val[idx + 1])\n", "# print(\"positive\")\n", "# audio = codec_decode(mm_vae_val[idx + 1])\n", "# audio.normalize_volume().play()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import torch" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "torch.equal(torch.tensor(mm_semantic_val[idx]), torch.tensor(mm_semantic_val[idx + 1]))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# original_npz_path = f\"/app/suno/data/dpo/7b_npz/{test_metas[idx]['id']}.npz\"\n", "# original_npz_path = \"/app/suno/data/dpo/7b_npz/729c3011-f672-4ccd-8d82-1cbf2b52ff69.npz\"\n", "# original_arr = np.load(original_npz_path)[\"v2_raw\"]" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "def validation_on_metas(input_metas):\n", " total_bad = 0\n", " total_good = 0\n", " for idx in range(len(input_metas)):\n", " if idx % 2 == 0:\n", " pos_idx = idx + 1\n", " if input_metas[idx].get(\"tags\") != input_metas[pos_idx].get(\"tags\"):\n", " # print(test_metas[idx].get(\"text\") == test_metas[pos_idx].get(\"text\"), test_metas[idx].get(\"tags\"), test_metas[pos_idx].get(\"tags\"))\n", " total_bad += 1\n", " else:\n", " total_good += 1\n", " print(total_good, total_bad)\n", " return" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "metas_tr = read_jsonl(os.path.join(OUT_DATA_DIR, \"metas_tr.jsonl\"))\n", "validation_on_metas(metas_tr)\n", "print(len(metas_tr))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "sum(len(meta[\"tags\"][0]) == 0 for meta in metas_tr)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "!cd /home/tony/Work/tony/slurm/diffusion && sbatch run_diffusion.sh" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import shutil\n", "\n", "# Basic file copy\n", "shutil.copy(\n", " \"/home/tony/Work/tony/Preference/make_dataset_diff_upsample_v1_r5_comb.ipynb\",\n", " os.path.join(OUT_DATA_DIR, \"make_dataset.ipynb\"),\n", ")\n", "print(\"Cache kept!\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Inspections " ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# df[df[\"preference\"] & (df[\"shimmer_score_diff\"] > 3)][\n", "# [\n", "# \"index\",\n", "# \"s3_id\",\n", "# \"total_shimmer_score\",\n", "# \"shimmer_score_diff\",\n", "# \"request_id\",\n", "# \"preference\",\n", "# ]\n", "# ].tail()\n", "\n", "# df[df[\"preference\"] & (df[\"pair_quality\"] < 0.1)][\n", "# [\n", "# \"index\",\n", "# \"s3_id\",\n", "# \"total_shimmer_score\",\n", "# \"shimmer_score_diff\",\n", "# \"pair_quality\",\n", "# \"request_id\",\n", "# \"preference\",\n", "# ]\n", "# ].tail()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# test_pair_df = df[df[\"request_id\"] == \"621c8b02-a905-48f1-a2d5-a4e8423d1505\"]\n", "# print(\n", "# test_pair_df[\n", "# [\n", "# \"s3_id\",\n", "# \"total_shimmer_score\",\n", "# \"pair_quality\",\n", "# \"request_id\",\n", "# \"preference\",\n", "# \"prompt_text\",\n", "# ]\n", "# ]\n", "# )\n", "# negative_audio = Audio.from_s3(\n", "# f\"s3://suno-data-uploads/studio/uploads/{test_pair_df['s3_id'].values[0]}.mp3\",\n", "# n_channels=2,\n", "# )\n", "# print(\"negative\")\n", "# negative_audio.get_segment(0, 30).play()\n", "# positive_audio = Audio.from_s3(\n", "# f\"s3://suno-data-uploads/studio/uploads/{test_pair_df['s3_id'].values[1]}.mp3\",\n", "# n_channels=2,\n", "# )\n", "# print(\"positive\")\n", "# positive_audio.get_segment(0, 30).play()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# total_dict = {}\n", "# total_dict.update(pair_quality_dict)\n", "# total_dict.update(pair_quality_1_dict)\n", "# total_dict.update(pair_quality_2_dict)\n", "# total_dict.update(pair_quality_3_dict)\n", "# len(total_dict)\n", "# with open(\n", "# os.path.join(\"/home/tony/Data/Preference/up_v1\", \"pair_quality.json\"), \"w\"\n", "# ) as fp:\n", "# json.dump(total_dict, fp, indent=4)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# import numpy as np" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# test_arr = np.load(\"/home/tony/Data/test_npz/diffusion_input_tensor([ 18, 182]).npy\")\n", "# test_arr.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# mm_vae_val = np.memmap(\n", "# os.path.join(OUT_DATA_DIR, \"data_vae_val.bin\"), dtype=np.float16, mode=\"r\"\n", "# )\n", "\n", "# mm_vae_val = mm_vae_val.reshape(-1, VAE_MEMMAP_SIZE, VAE_DIM)\n", "# print(mm_vae_val.shape)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# print(\"negative\")\n", "# audio = codec_decode(test_arr[0].T / 2.5)\n", "# audio.normalize_volume().play()\n", "\n", "# print(\"positive\")\n", "# audio = codec_decode(test_arr[1].T / 2.5)\n", "# audio.normalize_volume().play()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# print(\"negative\")\n", "# audio = codec_decode(mm_vae_val[18])\n", "# audio.normalize_volume().play()\n", "\n", "# print(\"positive\")\n", "# audio = codec_decode(mm_vae_val[19])\n", "# audio.normalize_volume().play()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import torch\n", "\n", "rng = torch.quasirandom.SobolEngine(1, scramble=True, seed=0)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "t = rng.draw(4)[:, 0].to(torch.bfloat16)\n", "print(t)\n", "# Replace 1% of t with ones to ensure training on terminal SNR\n", "t = torch.where(torch.rand_like(t) < 0.5, torch.ones_like(t), t)\n", "print(t)\n", "t = torch.repeat_interleave(t, repeats=2, dim=0)\n", "print(t)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "(t * 32).to(int) / 32" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "t[0] = 0.99" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "torch.round(t * 32) / 32" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# import json\n", "# # with open(f\"/home/tony/Data/Preference/up_v6/full_pair_quality.json\", \"r\") as f:\n", "# # result = json.load(f)\n", "# result = {}\n", "# print(len(result))\n", "# for job_idx in range(8):\n", "# with open(f\"/home/tony/Data/Preference/up_v6/full_pair_quality_{job_idx}.json\", \"r\") as fp:\n", "# current_result = json.load(fp)\n", "# result.update(current_result)\n", "# print(len(result))\n", "# with open(f\"/home/tony/Data/Preference/up_v6/full_pair_quality.json\", \"w\") as f:\n", "# json.dump(result, f, indent=4)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# import torch\n", "# semantic_codes_chunk = torch.ones((1, 100))\n", "# semantic_skip_phase = 0\n", "# semantic_skip_factor = 4\n", "# mask = torch.ones_like(semantic_codes_chunk, dtype=torch.bool)\n", "# indices = (\n", "# torch.arange(semantic_codes_chunk.size(1)) + semantic_skip_phase\n", "# ) % semantic_skip_factor == 0\n", "# mask[:, indices] = False\n", "# mask" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "df_slice.columns" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# df_slice[(df_slice[\"preference\"]) & (df_slice[\"total_shimmer_score\"] < 0.5)][[\"s3_id\", \"total_shimmer_score\"]].head()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# df_slice[(df_slice[\"preference\"]) & (df_slice[\"total_shimmer_score\"] > 2)][[\"s3_id\", \"total_shimmer_score\"]].head()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "language_info": { "name": "python" } }, "nbformat": 4, "nbformat_minor": 4 }