gr-libs 0.1.6.post1__tar.gz → 0.1.8__tar.gz

{gr_libs-0.1.6.post1 → gr_libs-0.1.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gr_libs
-Version: 0.1.6.post1
+Version: 0.1.8
 Summary: Package with goal recognition frameworks baselines
 Author: Ben Nageris
 Author-email: Matan Shamir <matan.shamir@live.biu.ac.il>, Osher Elhadad <osher.elhadad@live.biu.ac.il>

gr_libs-0.1.8/evaluation/analyze_results_cross_alg_cross_domain.py

@@ -0,0 +1,267 @@
+import copy
+import sys
+import matplotlib.pyplot as plt
+import numpy as np
+import os
+import dill
+from scipy.interpolate import make_interp_spline
+from scipy.ndimage import gaussian_filter1d
+from gr_libs.ml.utils.storage import (
+    get_experiment_results_path,
+    set_global_storage_configs,
+)
+from scripts.generate_task_specific_statistics_plots import get_figures_dir_path
+
+
+def smooth_line(x, y, num_points=300):
+    x_smooth = np.linspace(np.min(x), np.max(x), num_points)
+    spline = make_interp_spline(x, y, k=3)  # Cubic spline
+    y_smooth = spline(x_smooth)
+    return x_smooth, y_smooth
+
+
+if __name__ == "__main__":
+
+    fragmented_accuracies = {
+        "graml": {
+            "panda": {
+                "gd_agent": {
+                    "0.3": [],  # every list here should have number of tasks accuracies in it, since we done experiments for L111-L555. remember each accuracy is an average of #goals different tasks.
+                    "0.5": [],
+                    "0.7": [],
+                    "0.9": [],
+                    "1": [],
+                },
+                "gc_agent": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+            },
+            "minigrid": {
+                "obstacles": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+                "lava_crossing": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+            },
+            "point_maze": {
+                "obstacles": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+                "four_rooms": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+            },
+            "parking": {
+                "gd_agent": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+                "gc_agent": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+            },
+        },
+        "graql": {
+            "panda": {
+                "gd_agent": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+                "gc_agent": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+            },
+            "minigrid": {
+                "obstacles": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+                "lava_crossing": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+            },
+            "point_maze": {
+                "obstacles": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+                "four_rooms": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+            },
+            "parking": {
+                "gd_agent": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+                "gc_agent": {"0.3": [], "0.5": [], "0.7": [], "0.9": [], "1": []},
+            },
+        },
+    }
+
+    continuing_accuracies = copy.deepcopy(fragmented_accuracies)
+
+    # domains = ['panda', 'minigrid', 'point_maze', 'parking']
+    domains = ["minigrid", "point_maze", "parking"]
+    tasks = ["L111", "L222", "L333", "L444", "L555"]
+    percentages = ["0.3", "0.5", "1"]
+
+    for partial_obs_type, accuracies, is_same_learn in zip(
+        ["fragmented", "continuing"],
+        [fragmented_accuracies, continuing_accuracies],
+        [False, True],
+    ):
+        for domain in domains:
+            for env in accuracies["graml"][domain].keys():
+                for task in tasks:
+                    set_global_storage_configs(
+                        recognizer_str="graml",
+                        is_fragmented=partial_obs_type,
+                        is_inference_same_length_sequences=True,
+                        is_learn_same_length_sequences=is_same_learn,
+                    )
+                    graml_res_file_path = (
+                        f"{get_experiment_results_path(domain, env, task)}.pkl"
+                    )
+                    set_global_storage_configs(
+                        recognizer_str="graql", is_fragmented=partial_obs_type
+                    )
+                    graql_res_file_path = (
+                        f"{get_experiment_results_path(domain, env, task)}.pkl"
+                    )
+                    if os.path.exists(graml_res_file_path):
+                        with open(graml_res_file_path, "rb") as results_file:
+                            results = dill.load(results_file)
+                            for percentage in accuracies["graml"][domain][env].keys():
+                                accuracies["graml"][domain][env][percentage].append(
+                                    results[percentage]["accuracy"]
+                                )
+                    else:
+                        assert (False, f"no file for {graml_res_file_path}")
+                    if os.path.exists(graql_res_file_path):
+                        with open(graql_res_file_path, "rb") as results_file:
+                            results = dill.load(results_file)
+                            for percentage in accuracies["graml"][domain][env].keys():
+                                accuracies["graql"][domain][env][percentage].append(
+                                    results[percentage]["accuracy"]
+                                )
+                    else:
+                        assert (False, f"no file for {graql_res_file_path}")
+
+    plot_styles = {
+        ("graml", "fragmented", 0.3): "g--o",  # Green dashed line with circle markers
+        ("graml", "fragmented", 0.5): "g--s",  # Green dashed line with square markers
+        (
+            "graml",
+            "fragmented",
+            0.7,
+        ): "g--^",  # Green dashed line with triangle-up markers
+        ("graml", "fragmented", 0.9): "g--d",  # Green dashed line with diamond markers
+        ("graml", "fragmented", 1.0): "g--*",  # Green dashed line with star markers
+        ("graml", "continuing", 0.3): "g-o",  # Green solid line with circle markers
+        ("graml", "continuing", 0.5): "g-s",  # Green solid line with square markers
+        (
+            "graml",
+            "continuing",
+            0.7,
+        ): "g-^",  # Green solid line with triangle-up markers
+        ("graml", "continuing", 0.9): "g-d",  # Green solid line with diamond markers
+        ("graml", "continuing", 1.0): "g-*",  # Green solid line with star markers
+        ("graql", "fragmented", 0.3): "b--o",  # Blue dashed line with circle markers
+        ("graql", "fragmented", 0.5): "b--s",  # Blue dashed line with square markers
+        (
+            "graql",
+            "fragmented",
+            0.7,
+        ): "b--^",  # Blue dashed line with triangle-up markers
+        ("graql", "fragmented", 0.9): "b--d",  # Blue dashed line with diamond markers
+        ("graql", "fragmented", 1.0): "b--*",  # Blue dashed line with star markers
+        ("graql", "continuing", 0.3): "b-o",  # Blue solid line with circle markers
+        ("graql", "continuing", 0.5): "b-s",  # Blue solid line with square markers
+        ("graql", "continuing", 0.7): "b-^",  # Blue solid line with triangle-up markers
+        ("graql", "continuing", 0.9): "b-d",  # Blue solid line with diamond markers
+        ("graql", "continuing", 1.0): "b-*",  # Blue solid line with star markers
+    }
+
+    def average_accuracies(accuracies, domain):
+        avg_acc = {
+            algo: {perc: [] for perc in percentages} for algo in ["graml", "graql"]
+        }
+
+        for algo in avg_acc.keys():
+            for perc in percentages:
+                for env in accuracies[algo][domain].keys():
+                    env_acc = accuracies[algo][domain][env][
+                        perc
+                    ]  # list of 5, averages for L111 to L555.
+                    if env_acc:
+                        avg_acc[algo][perc].append(np.array(env_acc))
+
+        for algo in avg_acc.keys():
+            for perc in percentages:
+                if avg_acc[algo][perc]:
+                    avg_acc[algo][perc] = np.mean(np.array(avg_acc[algo][perc]), axis=0)
+
+        return avg_acc
+
+    def plot_domain_accuracies(
+        ax,
+        fragmented_accuracies,
+        continuing_accuracies,
+        domain,
+        sigma=1,
+        line_width=1.5,
+    ):
+        fragmented_avg_acc = average_accuracies(fragmented_accuracies, domain)
+        continuing_avg_acc = average_accuracies(continuing_accuracies, domain)
+
+        x_vals = np.arange(1, 6)  # Number of goals
+
+        # Create "waves" (shaded regions) for each algorithm
+        for algo in ["graml", "graql"]:
+            fragmented_y_vals_by_percentage = []
+            continuing_y_vals_by_percentage = []
+
+            for perc in percentages:
+                fragmented_y_vals = np.array(fragmented_avg_acc[algo][perc])
+                continuing_y_vals = np.array(continuing_avg_acc[algo][perc])
+
+                # Smooth the trends using Gaussian filtering
+                fragmented_y_smoothed = gaussian_filter1d(
+                    fragmented_y_vals, sigma=sigma
+                )
+                continuing_y_smoothed = gaussian_filter1d(
+                    continuing_y_vals, sigma=sigma
+                )
+
+                fragmented_y_vals_by_percentage.append(fragmented_y_smoothed)
+                continuing_y_vals_by_percentage.append(continuing_y_smoothed)
+
+                ax.plot(
+                    x_vals,
+                    fragmented_y_smoothed,
+                    plot_styles[(algo, "fragmented", float(perc))],
+                    label=f"{algo}, non-consecutive, {perc}",
+                    linewidth=0.5,  # Control line thickness here
+                )
+                ax.plot(
+                    x_vals,
+                    continuing_y_smoothed,
+                    plot_styles[(algo, "continuing", float(perc))],
+                    label=f"{algo}, consecutive, {perc}",
+                    linewidth=0.5,  # Control line thickness here
+                )
+
+        ax.set_xticks(x_vals)
+        ax.set_yticks(np.linspace(0, 1, 6))
+        ax.set_ylim([0, 1])
+        ax.set_title(f"{domain.capitalize()} Domain", fontsize=16)
+        ax.grid(True)
+
+    fig, axes = plt.subplots(
+        1, 4, figsize=(24, 6)
+    )  # Increase the figure size for better spacing (width 24, height 6)
+
+    # Generate each plot in a subplot, including both fragmented and continuing accuracies
+    for i, domain in enumerate(domains):
+        plot_domain_accuracies(
+            axes[i], fragmented_accuracies, continuing_accuracies, domain
+        )
+
+    # Set a single x-axis and y-axis label for the entire figure
+    fig.text(
+        0.5, 0.04, "Number of Goals", ha="center", fontsize=20
+    )  # Centered x-axis label
+    fig.text(
+        0.04, 0.5, "Accuracy", va="center", rotation="vertical", fontsize=20
+    )  # Reduced spacing for y-axis label
+
+    # Adjust subplot layout to avoid overlap
+    plt.subplots_adjust(
+        left=0.09, right=0.91, top=0.79, bottom=0.21, wspace=0.3
+    )  # More space on top (top=0.82)
+
+    # Place the legend above the plots with more space between legend and plots
+    handles, labels = axes[0].get_legend_handles_labels()
+    fig.legend(
+        handles,
+        labels,
+        loc="upper center",
+        ncol=4,
+        bbox_to_anchor=(0.5, 1.05),
+        fontsize=12,
+    )  # Moved above with bbox_to_anchor
+
+    # Save the figure and show it
+    save_dir = os.path.join("figures", "all_domains_accuracy_plots")
+    if not os.path.exists(save_dir):
+        os.makedirs(save_dir)
+    plt.savefig(os.path.join(save_dir, "accuracy_plots_smooth.png"), dpi=300)

{gr_libs-0.1.6.post1 → gr_libs-0.1.8}/evaluation/create_minigrid_map_image.py

@@ -2,21 +2,25 @@ from minigrid.wrappers import RGBImgPartialObsWrapper, ImgObsWrapper
 import numpy as np
 import gr_libs.ml as ml
 from minigrid.core.world_object import Wall
-#from q_table_plot import save_q_table_plot_image
+
+# from q_table_plot import save_q_table_plot_image
 from gymnasium.envs.registration import register
 
 env_name = "MiniGrid-SimpleCrossingS13N4-DynamicGoal-5x9-v0"
 # create an agent and train it (if it is already trained, it will get q-table from cache)
-agent = ml.TabularQLearner(env_name='MiniGrid-Walls-13x13-v0',problem_name = "MiniGrid-SimpleCrossingS13N4-DynamicGoal-5x9-v0")
+agent = ml.TabularQLearner(
+    env_name="MiniGrid-Walls-13x13-v0",
+    problem_name="MiniGrid-SimpleCrossingS13N4-DynamicGoal-5x9-v0",
+)
 # agent.learn()
 
 # save_q_table_plot_image(agent.q_table, 15, 15, (10,7))
 
 # add to the steps list the step the trained agent would take on the env in every state according to the q_table
 env = agent.env
-env = RGBImgPartialObsWrapper(env) # Get pixel observations
-env = ImgObsWrapper(env) # Get rid of the 'mission' field
-obs, _ = env.reset() # This now produces an RGB tensor only
+env = RGBImgPartialObsWrapper(env)  # Get pixel observations
+env = ImgObsWrapper(env)  # Get rid of the 'mission' field
+obs, _ = env.reset()  # This now produces an RGB tensor only
 
 img = env.get_frame()
 
@@ -24,7 +28,7 @@ img = env.get_frame()
 from PIL import Image
 import numpy as np
 
-image_pil = Image.fromarray(np.uint8(img)).convert('RGB')
+image_pil = Image.fromarray(np.uint8(img)).convert("RGB")
 image_pil.save(r"{}.png".format(env_name))
 
 # ####### show image

{gr_libs-0.1.6.post1 → gr_libs-0.1.8}/evaluation/file_system.py

@@ -4,26 +4,36 @@ import random
 import hashlib
 from typing import List
 
+
 def get_observations_path(env_name: str):
     return f"dataset/{env_name}/observations"
 
+
 def get_observations_paths(path: str):
     return [os.path.join(path, file_name) for file_name in os.listdir(path)]
 
+
 def create_partial_observabilities_files(env_name: str, observabilities: List[float]):
-    with open(r"dataset/{env_name}/observations/obs1.0.pkl".format(env_name=env_name), "rb") as f:
+    with open(
+        r"dataset/{env_name}/observations/obs1.0.pkl".format(env_name=env_name), "rb"
+    ) as f:
         step_1_0 = dill.load(f)
 
-    number_of_items_to_randomize = [int(observability * len(step_1_0)) for observability in observabilities]
+    number_of_items_to_randomize = [
+        int(observability * len(step_1_0)) for observability in observabilities
+    ]
     obs = []
     for items_to_randomize in number_of_items_to_randomize:
         obs.append(random.sample(step_1_0, items_to_randomize))
     for index, observability in enumerate(observabilities):
         partial_steps = obs[index]
-        file_path = r"dataset/{env_name}/observations/obs{obs}.pkl".format(env_name=env_name, obs=observability)
+        file_path = r"dataset/{env_name}/observations/obs{obs}.pkl".format(
+            env_name=env_name, obs=observability
+        )
         with open(file_path, "wb+") as f:
             dill.dump(partial_steps, f)
-            
+
+
 def md5(file_path: str):
     hash_md5 = hashlib.md5()
     with open(file_path, "rb") as f:
@@ -31,6 +41,7 @@ def md5(file_path: str):
             hash_md5.update(chunk)
     return hash_md5.hexdigest()
 
+
 def get_md5(file_path_list: List[str]):
     return [(file_path, md5(file_path=file_path)) for file_path in file_path_list]
 
@@ -39,4 +50,4 @@ def print_md5(file_path_list: List[str]):
     md5_of_observations = get_md5(file_path_list=file_path_list)
     for file_name, file_md5 in md5_of_observations:
         print(f"{file_name}:{file_md5}")
-    print("")
+    print("")

gr_libs-0.1.8/evaluation/generate_experiments_results.py

@@ -0,0 +1,141 @@
+import copy
+import sys
+import matplotlib.pyplot as plt
+import numpy as np
+import os
+import dill
+from gr_libs.ml.utils.storage import (
+    get_experiment_results_path,
+    set_global_storage_configs,
+)
+from scripts.generate_task_specific_statistics_plots import get_figures_dir_path
+
+
+def gen_graph(
+    graph_name,
+    x_label_str,
+    tasks,
+    panda_env,
+    minigrid_env,
+    parking_env,
+    maze_env,
+    percentage,
+):
+
+    fragmented_accuracies = {
+        "graml": {
+            #'panda': [],
+            #'minigrid': [],
+            #'point_maze': [],
+            "parking": []
+        },
+        "graql": {
+            #'panda': [],
+            #'minigrid': [],
+            #'point_maze': [],
+            "parking": []
+        },
+    }
+
+    continuing_accuracies = copy.deepcopy(fragmented_accuracies)
+
+    # domains_envs = [('minigrid', minigrid_env), ('point_maze', maze_env), ('parking', parking_env)]
+    domains_envs = [("parking", parking_env)]
+
+    for partial_obs_type, accuracies, is_same_learn in zip(
+        ["fragmented", "continuing"],
+        [fragmented_accuracies, continuing_accuracies],
+        [False, True],
+    ):
+        for domain, env in domains_envs:
+            for task in tasks:
+                set_global_storage_configs(
+                    recognizer_str="graml",
+                    is_fragmented=partial_obs_type,
+                    is_inference_same_length_sequences=True,
+                    is_learn_same_length_sequences=is_same_learn,
+                )
+                graml_res_file_path = (
+                    f"{get_experiment_results_path(domain, env, task)}.pkl"
+                )
+                set_global_storage_configs(
+                    recognizer_str="graql", is_fragmented=partial_obs_type
+                )
+                graql_res_file_path = (
+                    f"{get_experiment_results_path(domain, env, task)}.pkl"
+                )
+                if os.path.exists(graml_res_file_path):
+                    with open(graml_res_file_path, "rb") as results_file:
+                        results = dill.load(results_file)
+                        accuracies["graml"][domain].append(
+                            results[percentage]["accuracy"]
+                        )
+                else:
+                    assert (False, f"no file for {graml_res_file_path}")
+                if os.path.exists(graql_res_file_path):
+                    with open(graql_res_file_path, "rb") as results_file:
+                        results = dill.load(results_file)
+                        accuracies["graql"][domain].append(
+                            results[percentage]["accuracy"]
+                        )
+                else:
+                    assert (False, f"no file for {graql_res_file_path}")
+
+    def plot_accuracies(accuracies, partial_obs_type):
+        plt.figure(figsize=(10, 6))
+        colors = plt.cm.get_cmap(
+            "tab10", len(accuracies["graml"]) * len(accuracies["graml"]["parking"])
+        )
+
+        # Define different line styles for each algorithm
+        line_styles = {"graml": "-", "graql": "--"}
+        x_vals = np.arange(3, 8)
+        plt.xticks(x_vals)
+        plt.yticks(np.linspace(0, 1, 6))
+        plt.ylim([0, 1])
+        # Plot each domain-env pair's accuracies with different line styles for each algorithm
+        for alg in ["graml", "graql"]:
+            for idx, (domain, acc_values) in enumerate(accuracies[alg].items()):
+                if acc_values and len(acc_values) > 0:  # Only plot if there are values
+                    x_values = np.arange(3, len(acc_values) + 3)
+                    plt.plot(
+                        x_values,
+                        acc_values,
+                        marker="o",
+                        linestyle=line_styles[alg],
+                        color=colors(idx),
+                        label=f"{alg}-{domain}-{partial_obs_type}-{percentage}",
+                    )
+
+        # Set labels, title, and grid
+        plt.xlabel(x_label_str)
+        plt.ylabel("Accuracy")
+        plt.grid(True)
+
+        # Add legend to differentiate between domain-env pairs
+        plt.legend()
+
+        # Save the figure
+        fig_path = os.path.join(f"{graph_name}_{partial_obs_type}.png")
+        plt.savefig(fig_path)
+        print(f"Accuracies figure saved at: {fig_path}")
+
+    print(f"fragmented_accuracies: {fragmented_accuracies}")
+    plot_accuracies(fragmented_accuracies, "fragmented")
+    print(f"continuing_accuracies: {continuing_accuracies}")
+    plot_accuracies(continuing_accuracies, "continuing")
+
+
+if __name__ == "__main__":
+    # gen_graph("increasing_base_goals", "Number of base goals", ['L1', 'L2', 'L3', 'L4', 'L5'], panda_env='gd_agent', minigrid_env='obstacles', parking_env='gd_agent', maze_env='obstacles')
+    # gen_graph("increasing_dynamic_goals", "Number of dynamic goals", ['L1', 'L2', 'L3', 'L4', 'L5'], panda_env='gc_agent', minigrid_env='lava_crossing', parking_env='gc_agent', maze_env='four_rooms')
+    gen_graph(
+        "base_problems",
+        "Number of goals",
+        ["L111", "L222", "L333", "L444", "L555"],
+        panda_env="gd_agent",
+        minigrid_env="obstacles",
+        parking_env="gc_agent",
+        maze_env="obstacles",
+        percentage="0.7",
+    )