Commit 9bb5ff657bf803e1ce5a403f9998e700bf3a3f72
1 parent
78b39d22dd
Exists in
master
Adding n argument to pred_distribution_wt_sel
Adding some comments
Showing 1 changed file with 71 additions and 20 deletions Inline Diff
volia/stats.py
| 1 | 1 | ||
| 2 | import argparse | 2 | import argparse |
| 3 | 3 | ||
| 4 | import os | 4 | import os |
| 5 | import core.data | 5 | import core.data |
| 6 | import math | 6 | import math |
| 7 | import numpy as np | 7 | import numpy as np |
| 8 | import scipy.stats | 8 | import scipy.stats |
| 9 | import pickle | 9 | import pickle |
| 10 | import matplotlib.pyplot as plt | 10 | import matplotlib.pyplot as plt |
| 11 | import matplotlib.colors as mcolors | 11 | import matplotlib.colors as mcolors |
| 12 | from utils import SubCommandRunner | 12 | from utils import SubCommandRunner |
| 13 | |||
| 14 | |||
| 15 | from cycler import cycler | 13 | from cycler import cycler |
| 16 | 14 | ||
| 15 | |||
| 17 | def pred_distribution(predictions: str, labels: str, labelencoder: str, outdir: str): | 16 | def pred_distribution(predictions: str, labels: str, labelencoder: str, outdir: str): |
| 17 | ''' | ||
| 18 | Distribution of the prediction. | ||
| 18 | 19 | ||
| 20 | For each label, we plot the distribution of the class predicted. | ||
| 21 | For example, for each character, we plot the distribution of the characters predicted. | ||
| 22 | Another example, for each speaker, we plot the distribution of the characters predicted. | ||
| 23 | |||
| 24 | ''' | ||
| 19 | predictions = core.data.read_id_values(args.predictions, float) | 25 | predictions = core.data.read_id_values(args.predictions, float) |
| 20 | labels = core.data.read_labels(args.labels) | 26 | labels = core.data.read_labels(args.labels) |
| 21 | 27 | ||
| 22 | le = None | 28 | le = None |
| 23 | with open(args.labelencoder, "rb") as f: | 29 | with open(args.labelencoder, "rb") as f: |
| 24 | le = pickle.load(f) | 30 | le = pickle.load(f) |
| 25 | stats = {} | 31 | stats = {} |
| 26 | 32 | ||
| 27 | print("PREDICTIONS ---------------------------") | 33 | print("PREDICTIONS ---------------------------") |
| 28 | for id_, predictions_ in predictions.items(): | 34 | for id_, predictions_ in predictions.items(): |
| 29 | label = labels[id_][0] | 35 | label = labels[id_][0] |
| 30 | if label not in stats: | 36 | if label not in stats: |
| 31 | stats[label] = { | 37 | stats[label] = { |
| 32 | "nb_utt": 1, | 38 | "nb_utt": 1, |
| 33 | "predictions": np.expand_dims(predictions_, axis=0) | 39 | "predictions": np.expand_dims(predictions_, axis=0) |
| 34 | } | 40 | } |
| 35 | else: | 41 | else: |
| 36 | stats[label]["nb_utt"] = stats[label]["nb_utt"] + 1 | 42 | stats[label]["nb_utt"] = stats[label]["nb_utt"] + 1 |
| 37 | stats[label]["predictions"] = np.append(stats[label]["predictions"], np.expand_dims(predictions_, axis=0), axis=0) | 43 | stats[label]["predictions"] = np.append(stats[label]["predictions"], np.expand_dims(predictions_, axis=0), axis=0) |
| 38 | |||
| 39 | 44 | ||
| 40 | print("CALCULATING ---------------------------") | ||
| 41 | |||
| 42 | |||
| 43 | colors = [ | 45 | colors = [ |
| 44 | "darkorange", | 46 | "darkorange", |
| 45 | "red", | 47 | "red", |
| 46 | "blue" | 48 | "blue" |
| 47 | ] | 49 | ] |
| 48 | custom_cycler = (cycler(color=list(mcolors.TABLEAU_COLORS)) * | 50 | custom_cycler = (cycler(color=list(mcolors.TABLEAU_COLORS)) * |
| 49 | cycler(linestyle=['-', '--', '-.'])) | 51 | cycler(linestyle=['-', '--', '-.'])) |
| 50 | 52 | ||
| 53 | print("CALCULATING ---------------------------") | ||
| 51 | 54 | ||
| 52 | for label, stats_ in stats.items(): | 55 | for label, stats_ in stats.items(): |
| 53 | 56 | ||
| 54 | plt.gca().set_prop_cycle(custom_cycler) | 57 | plt.gca().set_prop_cycle(custom_cycler) |
| 55 | stats_mean = np.mean(stats_["predictions"], axis=0) | 58 | stats_mean = np.mean(stats_["predictions"], axis=0) |
| 56 | stats_std = np.std(stats_["predictions"], axis=0) | 59 | stats_std = np.std(stats_["predictions"], axis=0) |
| 57 | 60 | ||
| 58 | #print(label) | 61 | #print(label) |
| 59 | #print(stats_mean) | 62 | #print(stats_mean) |
| 60 | #print(stats_std) | 63 | #print(stats_std) |
| 61 | kwargs = dict(alpha=0.5) | 64 | kwargs = dict(alpha=0.5) |
| 62 | 65 | ||
| 63 | for i in range(stats_["predictions"].shape[1]): | 66 | for i in range(stats_["predictions"].shape[1]): |
| 64 | label_str = le.inverse_transform([i])[0] | 67 | label_str = le.inverse_transform([i])[0] |
| 65 | #plt.hist(stats_["predictions"][:, i], bins=10, label=label_str, **kwargs) | 68 | #plt.hist(stats_["predictions"][:, i], bins=10, label=label_str, **kwargs) |
| 66 | mu = stats_mean[i] | 69 | mu = stats_mean[i] |
| 67 | variance = stats_std[i] * stats_std[i] | 70 | variance = stats_std[i] * stats_std[i] |
| 68 | sigma = stats_std[i] | 71 | sigma = stats_std[i] |
| 69 | # math.sqrt(variance) | ||
| 70 | print(f"{i}: mu {mu}, var {variance}, sigma {sigma}") | 72 | print(f"{i}: mu {mu}, var {variance}, sigma {sigma}") |
| 71 | 73 | ||
| 72 | #x_values = np.arange(-1, 5, 0.1) | 74 | #x_values = np.arange(-1, 5, 0.1) |
| 73 | 75 | ||
| 74 | #y_values = scipy.stats.norm(mu, variance) | 76 | #y_values = scipy.stats.norm(mu, variance) |
| 75 | #y = scipy.stats.norm.pdf(x,mean,std) | 77 | #y = scipy.stats.norm.pdf(x,mean,std) |
| 76 | 78 | ||
| 77 | #plt.plot(x_values, y_values.pdf(x_values,)) | 79 | #plt.plot(x_values, y_values.pdf(x_values,)) |
| 78 | 80 | ||
| 79 | #x, step = np.linspace(mu - 3*sigma, mu + 3*sigma, 1000, retstep=True) | 81 | #x, step = np.linspace(mu - 3*sigma, mu + 3*sigma, 1000, retstep=True) |
| 80 | x = np.linspace(0, 1, 1000) | 82 | x = np.linspace(0, 1, 1000) |
| 81 | #x = np.linspace(mu - 3*sigma, mu + 3*sigma, 1000) | 83 | #x = np.linspace(mu - 3*sigma, mu + 3*sigma, 1000) |
| 82 | #x, step = np.linspace(0, 1, 1000, retstep=True) | 84 | #x, step = np.linspace(0, 1, 1000, retstep=True) |
| 83 | 85 | ||
| 84 | P = scipy.stats.norm.cdf(x, mu, sigma) | 86 | P = scipy.stats.norm.cdf(x, mu, sigma) |
| 85 | #print(step) | 87 | #print(step) |
| 86 | plt.plot(x, P, label=label_str, **kwargs) | 88 | plt.plot(x, P, label=label_str, **kwargs) |
| 87 | #plt.savefig("simple_gaussian.pdf") | 89 | #plt.savefig("simple_gaussian.pdf") |
| 88 | 90 | ||
| 89 | plt.legend() | 91 | plt.legend() |
| 90 | plt.savefig(os.path.join(args.outdir, f"{label}_prediction_cdf.pdf")) | 92 | plt.savefig(os.path.join(args.outdir, f"{label}_prediction_cdf.pdf")) |
| 91 | plt.clf() | 93 | plt.clf() |
| 92 | 94 | ||
| 93 | print("Decisions") | 95 | print("Decisions") |
| 94 | 96 | ||
| 95 | 97 | ||
| 96 | def pred_distribution_wt_sel(predictions: str, labels: str, labelencoder: str, outdir: str): | 98 | def pred_distribution_wt_sel(predictions: str, n: int, labels: str, labelencoder: str, outdir: str): |
| 97 | 99 | ||
| 100 | ''' | ||
| 101 | Distribution of the predictions with selection process. | ||
| 102 | |||
| 103 | 1) For each dimension, select the n individus with the maximum values for the focused dimension. | ||
| 104 | We name S_i the set of n selected individus for the dimension i. | ||
| 105 | 2) For each subset S_i, we plot the distribution of each dimension. | ||
| 106 | ''' | ||
| 107 | |||
| 108 | le = None | ||
| 109 | with open(args.labelencoder, "rb") as f: | ||
| 110 | le = pickle.load(f) | ||
| 111 | |||
| 98 | keys_preds, matrix_preds = core.data.read_features_with_matrix(predictions) | 112 | keys_preds, matrix_preds = core.data.read_features_with_matrix(predictions) |
| 99 | n = 3 | 113 | |
| 100 | print(matrix_preds.shape) | 114 | colors = [ |
| 115 | "darkorange", | ||
| 116 | "red", | ||
| 117 | "blue" | ||
| 118 | ] | ||
| 119 | custom_cycler = (cycler(color=list(mcolors.TABLEAU_COLORS)) * | ||
| 120 | cycler(linestyle=['-', '--', '-.'])) | ||
| 121 | |||
| 122 | kwargs = dict(alpha=0.5) | ||
| 123 | |||
| 124 | stats_of = open(os.path.join(args.outdir, f"stats.txt"), "w") | ||
| 101 | for j in range(matrix_preds.shape[1]): | 125 | for j in range(matrix_preds.shape[1]): |
| 126 | |||
| 127 | label_focused = le.inverse_transform([j])[0] | ||
| 102 | indices = (-matrix_preds[:, j]).argsort()[:n] | 128 | indices = (-matrix_preds[:, j]).argsort()[:n] |
| 103 | print(f"INDICE: {j}") | 129 | |
| 104 | print("indices") | 130 | print(f"LABEL: {label_focused}", file=stats_of) |
| 105 | print(indices) | 131 | print(f"INDICE: {j}", file=stats_of) |
| 106 | print("Best values") | 132 | print("indices", file=stats_of) |
| 107 | print(matrix_preds[indices, j]) | 133 | print(indices, file=stats_of) |
| 108 | print("All dimensions of best values") | 134 | print("Best values", file=stats_of) |
| 109 | print(matrix_preds[indices]) | 135 | print(matrix_preds[indices, j], file=stats_of) |
| 110 | # Select the n best for each column | 136 | print("All dimensions of best values", file=stats_of) |
| 137 | print(matrix_preds[indices], file=stats_of) | ||
| 138 | |||
| 139 | # Use it to build a plot. | ||
| 140 | pred_ = matrix_preds[indices] | ||
| 141 | stats_mean = np.mean(pred_, axis=0) | ||
| 142 | stats_std = np.std(pred_, axis=0) | ||
| 143 | for i in range(matrix_preds.shape[1]): | ||
| 144 | label_str = le.inverse_transform([i])[0] | ||
| 145 | mu = stats_mean[i] | ||
| 146 | variance = stats_std[i] * stats_std[i] | ||
| 147 | sigma = stats_std[i] | ||
| 148 | |||
| 149 | print(f"{i}: mu {mu}, var {variance}, sigma {sigma}") | ||
| 150 | |||
| 151 | x = np.linspace(0, 1, 1000) | ||
| 152 | |||
| 153 | P = scipy.stats.norm.cdf(x, mu, sigma) | ||
| 154 | plt.plot(x, P, label=label_str, **kwargs) | ||
| 155 | |||
| 156 | plt.legend() | ||
| 157 | plt.savefig(os.path.join(args.outdir, f"{label_focused}_prediction_cdf.pdf")) | ||
| 158 | plt.clf() | ||
| 159 | stats_of.close() | ||
| 111 | pass | 160 | pass |
| 112 | 161 | ||
| 113 | 162 | ||
| 114 | def utt2dur(utt2dur: str, labels: str): | 163 | def utt2dur(utt2dur: str, labels: str): |
| 115 | if labels == None: | 164 | if labels == None: |
| 116 | pass | 165 | pass |
| 117 | else: | 166 | else: |
| 118 | pass | 167 | pass |
| 119 | 168 | ||
| 120 | durations = [] | 169 | durations = [] |
| 121 | with open(utt2dur, "r") as f: | 170 | with open(utt2dur, "r") as f: |
| 122 | for line in f: | 171 | for line in f: |
| 123 | splited = line.replace("\n", "").split(" ") | 172 | splited = line.replace("\n", "").split(" ") |
| 124 | durations.append(float(splited[1])) | 173 | durations.append(float(splited[1])) |
| 125 | 174 | ||
| 126 | durations = np.asarray(durations, dtype=float) | 175 | durations = np.asarray(durations, dtype=float) |
| 127 | print(durations.shape) | 176 | print(durations.shape) |
| 128 | mean = np.mean(durations) | 177 | mean = np.mean(durations) |
| 129 | std = np.std(durations) | 178 | std = np.std(durations) |
| 130 | 179 | ||
| 131 | print(f"mean: {mean}") | 180 | print(f"mean: {mean}") |
| 132 | print(f"std: {std}") | 181 | print(f"std: {std}") |
| 133 | 182 | ||
| 134 | 183 | ||
| 135 | if __name__ == "__main__": | 184 | if __name__ == "__main__": |
| 136 | 185 | ||
| 137 | # Parser | 186 | # Parser |
| 138 | parser = argparse.ArgumentParser(description="Statistics") | 187 | parser = argparse.ArgumentParser(description="Statistics") |
| 139 | subparsers = parser.add_subparsers(title="actions") | 188 | subparsers = parser.add_subparsers(title="actions") |
| 140 | 189 | ||
| 141 | # pred-distribution | 190 | # pred-distribution |
| 142 | parser_pred_dist = subparsers.add_parser("pred-distribution", help="plot distributions of prediction through labels") | 191 | parser_pred_dist = subparsers.add_parser("pred-distribution", help="plot distributions of prediction through labels") |
| 143 | parser_pred_dist.add_argument("--predictions", type=str, help="prediction file", required=True) | 192 | parser_pred_dist.add_argument("--predictions", type=str, help="prediction file", required=True) |
| 144 | parser_pred_dist.add_argument("--labels", type=str, help="label file", required=True) | 193 | parser_pred_dist.add_argument("--labels", type=str, help="label file", required=True) |
| 145 | parser_pred_dist.add_argument("--labelencoder", type=str, help="label encode pickle file", required=True) | 194 | parser_pred_dist.add_argument("--labelencoder", type=str, help="label encode pickle file", required=True) |
| 146 | parser_pred_dist.add_argument("--outdir", type=str, help="output file", required=True) | 195 | parser_pred_dist.add_argument("--outdir", type=str, help="output file", required=True) |
| 147 | parser_pred_dist.set_defaults(which="pred_distribution") | 196 | parser_pred_dist.set_defaults(which="pred_distribution") |
| 148 | 197 | ||
| 149 | # pred-distribution-with-selection | 198 | # pred-distribution-with-selection |
| 150 | parser_pred_dist_wt_sel = subparsers.add_parser("pred-distribution-with-selection", help="plot distributions of prediction through labels with a selection of the n best records by column/class prediction.") | 199 | parser_pred_dist_wt_sel = subparsers.add_parser("pred-distribution-with-selection", help="plot distributions of prediction through labels with a selection of the n best records by column/class prediction.") |
| 151 | parser_pred_dist_wt_sel.add_argument("--predictions", type=str, help="prediction file", required=True) | 200 | parser_pred_dist_wt_sel.add_argument("--predictions", type=str, help="prediction file", required=True) |
| 201 | parser_pred_dist_wt_sel.add_argument("-n", type=int, help="Number of maximum selected for each prediction y_i.") | ||
| 152 | parser_pred_dist_wt_sel.add_argument("--labels", type=str, help="label file", required=True) | 202 | parser_pred_dist_wt_sel.add_argument("--labels", type=str, help="label file", required=True) |
| 153 | parser_pred_dist_wt_sel.add_argument("--labelencoder", type=str, help="label encode pickle file", required=True) | 203 | parser_pred_dist_wt_sel.add_argument("--labelencoder", type=str, help="label encode pickle file", required=True) |
| 154 | parser_pred_dist_wt_sel.add_argument("--outdir", type=str, help="output file", required=True) | 204 | parser_pred_dist_wt_sel.add_argument("--outdir", type=str, help="output file", required=True) |
| 155 | parser_pred_dist_wt_sel.set_defaults(which="pred_distribution_with_selection") | 205 | parser_pred_dist_wt_sel.set_defaults(which="pred_distribution_with_selection") |
| 206 | |||
| 156 | # duration-stats | 207 | # duration-stats |
| 157 | parser_utt2dur = subparsers.add_parser("utt2dur", help="distribution of utt2dur") | 208 | parser_utt2dur = subparsers.add_parser("utt2dur", help="distribution of utt2dur") |
| 158 | parser_utt2dur.add_argument("--utt2dur", type=str, help="utt2dur file", required=True) | 209 | parser_utt2dur.add_argument("--utt2dur", type=str, help="utt2dur file", required=True) |
| 159 | parser_utt2dur.add_argument("--labels", type=str, default=None, help="labels file") | 210 | parser_utt2dur.add_argument("--labels", type=str, default=None, help="labels file") |
| 160 | parser_utt2dur.set_defaults(which="utt2dur") | 211 | parser_utt2dur.set_defaults(which="utt2dur") |
| 161 | 212 | ||
| 162 | # Parse | 213 | # Parse |
| 163 | args = parser.parse_args() | 214 | args = parser.parse_args() |
| 164 | 215 | ||
| 165 | # Run commands | 216 | # Run commands |